The Numbers — AI Search Statistics 2026

Share of its cited sources ChatGPT Search replaces every week, against 56% for Google, across 82,619 prompts re-sampled weekly for 17 weeks (1,548,213 snapshots, six countries) in a SISTRIX study — with a stable core of a few domains present for 86% of prompts and everything outside it rotating at 89% per week.

SISTRIX, 2026-05 SISTRIX study — vendor-published, not replicated; window closed 2026-04-08, before later model changes single-source · on the wire · source ↗

Share of the time Google AI Overviews and Google AI Mode cite different domains for the same prompt, per SISTRIX — two surfaces from one company that agree on source selection less than a fifth of the time, so a citation-share number measured on one is not a reading of the other.

SISTRIX, 2026-05 SISTRIX study — vendor-published, not replicated; window closed 2026-04-08, before later model changes single-source · on the wire · source ↗

Share of news articles cited as sources that remained permanently in the citation set across 17 weeks of weekly re-sampling, per SISTRIX — for a publisher an AI citation behaves less like an asset that accrues than a placement that expires. By contrast, 43% of brand queries kept the brand's own domain present in all 17 weeks while its co-citations rotated at 70% per week.

SISTRIX, 2026-05 SISTRIX study — vendor-published, not replicated; window closed 2026-04-08, before later model changes single-source · on the wire · source ↗

Shared search-query strings between two runs of an identical, hash-verified ChatGPT prompt eight minutes apart on Sept. 17, 2026, per Seer Interactive — the same instability SISTRIX measured weekly in citations, here on a timescale of minutes in the underlying search queries. Between Sept. 3 and Sept. 16, the same prompt's retrieved pipe-language query data dropped from four lines to zero while its unique retrieved URLs rose from 14 to 40.

Seer Interactive, 2026-09 unchecked · on the wire · source ↗

YouTube's mention share in Google AI Overviews — the most-cited domain, ahead of Reddit (18.5%), Facebook (10.7%), Google itself (7.1%) and Instagram (5.8%).

Ahrefs, 2026-07 unchecked · on the wire · source ↗

Google's own rank among the most-cited domains in AI Mode answers, driven by Google-hosted Business Profile and Product Knowledge Panel cards.

Search Engine Journal, 2026-07 unchecked · on the wire · source ↗ · study ↗

Share of estimated ChatGPT search demand sitting in categories with no single clearly leading brand — only 15.2% of 1,094 categories analyzed have one, per a Kevin Indig analysis of six months of Semrush citation data; the most-cited domain matched the most-mentioned brand only 20.8% of the time.

Search Engine Journal, 2026-07 unchecked · on the wire · source ↗ · study ↗

Share of AI Mode passages reused 100 or more times that come from pages already ranking #1 organically, per a Pillarbase study of 15.7 million citations; the most-recycled passages average 117 words, answer in the first sentence about 80% of the time, and are self-contained about 85% of the time.

Search Engine Land, 2026-07 Pillarbase study — vendor-published single-source · on the wire · source ↗ · study ↗

Perplexity's rate of "ghost citations" — pulling from content without naming the source brand — the highest of seven engines Writesonic measured, followed by Google AI Mode (49%), AI Overviews (41%), ChatGPT (37%), Gemini (25%), Grok (22%) and Microsoft Copilot (19%), across roughly 16 million brand appearances.

Search Engine Land, 2026-07 Writesonic study — vendor-published single-source · on the wire · source ↗ · study ↗

How much more often AI models searched for brands they already knew versus unfamiliar ones (55.7% vs. 17.4%) across 66 U.S. buyer prompts tested on ChatGPT, Gemini and Claude; 63% of brand-specific searches surfaced one of each model's five most-familiar brands.

Search Engine Land, 2026-07 geoSurge study — vendor-published single-source · on the wire · source ↗ · study ↗

Trust in an AI-generated answer when it cites a fully trusted news source, per an AOP/Ipsos survey of 1,000 UK adults — versus about 25% for a neutral source and roughly 10% for a distrusted one; 37% of respondents (45% among ages 45-54) didn't realize AI tools can fabricate sources.

Press Gazette AI, 2026-07 AOP-commissioned study — publisher trade body unchecked · on the wire · source ↗ · study ↗

Verticals where Google AI Mode's citation mix shifted from publishers toward retailers over the 30 days ending July 15, 2026 — publisher mentions fell nearly in half while mentions of retailers like Home Depot and Wayfair roughly tripled and quadrupled respectively, per Kevin Indig's Growth Memo analysis of the Search Signals Index; AI Mode's overall mention volume contracted 12.9% over the same window while AI Overviews held flat.

Growth Memo (Kevin Indig), 2026-08 Search Signals Index — Kevin Indig/Growth Memo analysis, self-published data unchecked · on the wire · source ↗

Share of AI-cited pages that looked fresh by last-modified date versus by original publish date, among 4,124 pages with both dates in a Seer Interactive analysis of 47,097 citations across ChatGPT, Gemini and Perplexity — refreshed old pages outcited newly published ones. Overall, 75% of cited pages had been updated within the past year, 88% within two years; recency requirements varied by engine (Gemini 78%, ChatGPT 73%, Perplexity 65%).

Seer Interactive, 2026-07 Seer Interactive study — vendor-published single-source · on the wire · source ↗

UGC/community platforms' share of domains cited in ChatGPT answers to SaaS buying queries — more than four times publishers' 4.0% and ahead of review platforms (7.4% at discovery to 13.2% at evaluation) at every buyer-journey stage, with Wikipedia alone accounting for 10.1 to 14.0 of the 17.1 points, per a Growth Memo analysis of roughly 35,000 citation URLs tracked in Profound over December 2025.

Growth Memo (Kevin Indig), 2026-08 Growth Memo/Profound analysis — self-published data unchecked · on the wire · source ↗

Favorability percentile on OpenAI's GPT 5.6 Terra for real brands with positive- versus negative-sounding names in a controlled test, per a Peec AI analysis of 904 companies across ChatGPT, Gemini, AI Overviews, AI Mode and Copilot — well-known brands like Discord and Slack were unaffected.

Peec AI blog, 2026-08 Peec AI study — vendor-published single-source · on the wire · source ↗

Higher odds of selection for a Reddit AI search comment one standard deviation more formal than average, while personal-experience language cut the odds to 0.789, per a University of Illinois Urbana-Champaign preprint auditing 30,000 Reddit AI search answers across 14.68 million comments in 20 advice subreddits. Selected comments ranked at the 91st percentile of their thread's vote count versus the 45th percentile for passed-over comments, and first-person words like "I" and "my" fell from 3.3% of quoted comments to 0.06% of final answers.

Search Engine Journal, 2026-09 UIUC preprint — not peer reviewed, single source unchecked · on the wire · source ↗ · study ↗

Growth in Reddit's tracked search footprint between June 2025 and February 2026, with its AI citations up another 84% from February to July 2026 before plateauing at a level several times higher than a year earlier — Google AI Overviews drives roughly 88% of Reddit's citation volume, per a BrightEdge analysis using its AI Hyper Cube, Data Cube X and Expert Services datasets.

BrightEdge Blog, 2026-08 BrightEdge study — vendor-published contested · on the wire · source ↗

-86.4% per Promptwatch — Reddit's ChatGPT Search citation share fell 86.4% relative over the same window, against 11.3% on AI Overviews and 30.5% on AI Mode. Not a contradiction: the plateau is an AI Overviews plateau, and Reddit's position across engines is not settled. (wire item)

Share of Google AI Overviews' Instagram Reels citations going to posted content (89%, and 53% for Facebook video) versus roughly 80% of ChatGPT's LinkedIn citations going to profile and company pages instead — the two engines pull from structurally different parts of the same social platforms, per a BrightEdge analysis of five platforms across nine industries over twelve weekly observations.

BrightEdge Blog, 2026-08 BrightEdge study — vendor-published single-source · on the wire · source ↗

Share of brands referenced by AI answers that were named by all three of GPT-4o, Gemini 2.5 Flash and Claude Sonnet 4.6 in a Fractl analysis of 4,320 AI responses across 96 prompts — 77% were named by just one model. In the travel category, Booking.com led with 285 AI mentions ahead of Airbnb (227) and Expedia (215).

Search Engine Land, 2026-08 Fractl study — vendor-published, author is a Fractl co-founder single-source · on the wire · source ↗ · study ↗

How often a brand ChatGPT itself named in its own search query reached the final answer (68.9%) versus a brand surfacing only on retrieved pages (2.1%) — a roughly 33x gap — per an independent traffic analysis of a ChatGPT Plus account across 27 conversations; only 3.1% of 3,554 retrieved pages earned a citation at all, with citation rates falling from 5.2% at the first-listed domain to 0.3% from the sixth position on.

PPC Land, 2026-08 single-source — confirmation pending single-source · on the wire · source ↗ · study ↗

Business websites' share of citations in Gemini's local AI search results — nearly 60% of 14,472 citations across 1,487 queries in the 50 largest U.S. metros, with Reddit second at 13.7% — per a Steady Demand analysis. Identical queries returned overlapping sources only about 40% of the time and named the same top business in just 7% of repeat searches, versus roughly 90% listing consistency in Google's local pack; Gemini and ChatGPT cited the same domains for the same queries only 8% of the time.

Search Engine Land, 2026-08 Steady Demand study — vendor-published single-source · on the wire · source ↗ · study ↗

Share of Google AI Overviews' YouTube citations linking to a timestamped moment within the video, versus 0.6% for ChatGPT — Google also cites YouTube Shorts 17.5% of the time versus 1.1% for ChatGPT — per a BrightEdge analysis of cited YouTube URLs across nine industries over twelve weekly observations (May 17–Aug 2, 2026); BrightEdge says the pattern shows Google navigates within a video while ChatGPT cites it whole.

BrightEdge Blog, 2026-08 BrightEdge study — vendor-published single-source · on the wire · source ↗

Growth in search volume behind top prompts citing Yahoo Finance in Google AI Overviews between January and July 2026 — versus just 1.1x growth in the number of distinct prompts — after AI Overviews expanded into the finance category in April 2026 and shifted volume toward ticker lookups; Yahoo's brand-mention share rose to roughly half of tracked mentions by July while Forbes, which led in January, fell by about a third, per a BrightEdge analysis.

BrightEdge Blog, 2026-08 BrightEdge study — vendor-published single-source · on the wire · source ↗

Share of vendor-authored listicles that still gained estimated organic traffic even as Google's share of listicles holding position 1 fell from roughly 19-20% in January 2026 to about 15-16% in August — Google also shows roughly 4.5x more listicles for queries that explicitly ask for a list than for implicit category queries, per a Growth Memo analysis of 5.32 million Google result rows across 60,000 queries.

Growth Memo (Kevin Indig), 2026-08 Growth Memo/SE Ranking analysis — self-published data unchecked · on the wire · source ↗

UK B2B software firms that never appeared in any of 24 shortlist-style answers each was eligible for, across 1,608 queries Recrawled ran against ChatGPT, Gemini, Google AI Overviews and AI Mode — all 40 were named correctly when searched directly by brand. How the buyer question was worded swung results far more than engine or review volume did: a UK-specific shortlist question named 63% of the 40 firms, a market-leader-versus-rival question named none. Exclusion rates also varied sharply by engine — Gemini left out 42% of the firms named somewhere, AI Mode 35%, AI Overviews 19%, ChatGPT 16% — with no correlation to a firm's review-site volume.

PPC Land, 2026-10 Recrawled study — vendor-published (UK answer-engine agency) single-source · on the wire · source ↗ · study ↗

Reduction in a listicle's odds of reaching the Google top 3 when it carries an old or unreadable publish date — the strongest predictor measured, ahead of author bios, schema markup and word count; counted listicles ("17 Best Tools") also beat unnumbered editorial roundups in 56.8% of head-to-head matchups, per a Growth Memo analysis of nearly 60,000 Google results pages.

Growth Memo (Kevin Indig), 2026-08 Growth Memo analysis — self-published data unchecked · on the wire · source ↗

Recommendation rate for the well-known brand over an otherwise-identical rival product, testing GPT-4o-mini, Claude Sonnet and Gemini on skincare recommendations — but a rival needed less than a 0.1-star ratings edge to overturn that advantage; when every brand adopted the same marketing tactics, the incumbent's edge collapsed from a payoff of +0.802 to +0.007, and brands that opted out received no recommendations at all, per an arXiv paper by Xi Chu and Yupeng Hou.

arXiv GEO, 2026-06 single skincare-category test — not verified across other verticals unchecked · on the wire · source ↗

Relative drop in Reddit's share of ChatGPT Search citations — an average 3.83% (July 18-Aug 7) down to 0.52% (Aug 14-17) — against just an 11.3% decline on Google AI Overviews (2.37% to 2.10%) and 30.5% on AI Mode over the same window, per Promptwatch. It attributes the ChatGPT-only break to an August 8 change that pushed use of the site: operator in ChatGPT's background search queries from 0.37% to 16.8%, and calls the decline provisional pending confirmation it isn't a data-collection issue.

Promptwatch, 2026-08 Promptwatch study — vendor-published and called provisional by its author single-source · on the wire · source ↗

Exact word-match rate between a shopper's prompt and the brand named in Reddit-cited answers — 43.5-52.0% for restaurants and 36.4-49.6% for ecommerce across Google AI Overviews and ChatGPT, with a median of four brands per Reddit-cited answer — evidence Reddit citations skew toward comparison and purchase-decision prompts rather than top-of-funnel discovery, per a BrightEdge analysis spanning May 17-Aug 2, 2026 across nine industries.

BrightEdge Blog, 2026-08 BrightEdge study — vendor-published single-source · on the wire · source ↗

ChatGPT's top five cited domains for health questions are all government agencies or hospital systems, with YouTube absent from its top ten, while Google AI Overviews' YouTube citations for health queries run 1.8x above their June level and outrank clinical sources; for finance questions, AI Overviews concentrates 51% of citations in its top source versus a 19/15/14/14/11% spread across ChatGPT's top five, which includes the smaller valuation platform SimplyWall.st, per a BrightEdge analysis using its AI Hyper Cube citation tracking.

BrightEdge Blog, 2026-08 BrightEdge study — vendor-published single-source · on the wire · source ↗

Product pages' share of citations in a Ten Speed analysis of 170 B2B buyer-evaluation prompts across ChatGPT, Perplexity, Claude and Gemini (7,387 citation appearances, tracked via Peec AI) — the largest single category, ahead of blog/PR content (17.4%) and comparison pages/listicles (about 13% each); Reddit, YouTube and forums combined for just 4.2%.

Search Engine Journal, 2026-09 Ten Speed study — vendor-published; SEJ's own review found the comparison descriptive only, without statistical testing, and a prompt-count discrepancy (220 vs. 170) the vendor had to correct single-source · on the wire · source ↗ · study ↗

Share of English-language citations held by the ten most-cited domains, in a cross-platform audit of ChatGPT, Perplexity and Google AI Overview on 20 English mental-health questions plus six additional languages — 15,942 citations across 1,140 responses and 1,713 unique domains. Government, commercial-health and academic sources each drew roughly 22% of citations overall, but the three platforms diverged sharply in consistency and in which source types they favored; explicitly asking a platform to show its sources produced only modest changes in what it cited.

arXiv AI-search citations, 2026-08 arXiv preprint, not yet peer reviewed unchecked · on the wire · source ↗

Rate at which ChatGPT, Gemini and Claude recommended the same brand to two different personas asking the identical question, versus roughly 2 in 5 when the same persona repeated the question — across 71,147 responses built from income, age, gender and occupation combinations on credit-card, clothing and furniture queries, per Profound. On Gemini clothing queries, high-income personas drew 24 percentage points more brand-owned citations and 17 points fewer earned-media citations than low-income personas; brand mentions per response rose from 5.8 for low-income personas to 7.5 for high-income ones.

Profound, 2026-09 Profound study — vendor-published single-source · on the wire · source ↗

Reddit's share of ChatGPT's Education-category citations over two weeks in August 2026, per a BrightEdge analysis tracking a fixed weekly panel of U.S. education prompts from week 28 through week 37 (July 6-Sept 13). Government, university and professional-association domains — state.gov, asu.edu, apa.org, aacnnursing.org — took the vacated positions and still held them five weeks later. BrightEdge attributes the drop to Reddit's robots.txt policy; this wire's Aug 27 coverage found a change in ChatGPT's own background-query behavior preceded the decline by about six days, leaving the single-cause account unsettled.

BrightEdge Blog, 2026-08 BrightEdge study — vendor-published; causal claim (Reddit's robots.txt) is BrightEdge's own attribution, not independently confirmed single-source · on the wire · source ↗

Drop in the average number of B2B tech brands ChatGPT named per answer, from 5.36 in March 2026 to 3.87 in September, tracked across a fixed panel of the same non-branded prompts held constant over six months, per BrightEdge. The drop was steepest in informational answers (5.15 to 3.50 brands, down 32%), where Microsoft, Google and Amazon lost the most ground and IBM gained the most; transactional answers moved the opposite way, rising from 2.02 to 3.34 brands per answer, a reversal BrightEdge's own analysis excludes from its headline finding.

BrightEdge Blog, 2026-09 BrightEdge study — vendor-published single-source · on the wire · source ↗

Share of fan-out queries — the sub-queries ChatGPT runs while answering — in which ChatGPT-5.6 Luna, OpenAI's new default model, used the site: operator to limit a query to one domain, up from 21.6% for its predecessor GPT-5.5 — Luna also consulted 43% fewer sources overall (3,481 versus 6,134) and generated about 13% fewer inline citations, even as the total number of fan-out queries held steady, per a Seer Interactive analysis of 346 matched prompts across six industries.

Seer Interactive, 2026-09 Seer Interactive study — vendor-published contested · on the wire · source ↗

6.8% per Ahrefs — Ahrefs measured the site: operator at 6.8% of 15 million fan-out queries across 66.6 million logged-out responses — far below Seer's 76% on a 346-prompt set. Ahrefs' own comparison table attributes the gap to sample scope and prompt mix, not a methodology error in either study. (wire item)

Reddit's share of ChatGPT citations, which fell from about 1 in 5 answers in March 2026 to 1 in 43 by late September — an 88% drop — while Wikipedia fell from 1 in 7 to 1 in 35, down 80%, per Ahrefs' Brand Radar data tracking 2.4 to 3.8 million ChatGPT answers a month. Over the same stretch the combined citation share of OpenAI's 17 licensed content partners more than doubled, from 7.5% in March to a July peak of 15.8%, easing to 12.5% by late September. Reddit still ranks ninth among domains ChatGPT searches directly with the site: operator, which Ahrefs reads as the model still fetching Reddit for opinions while citing it less.

Ahrefs Blog, 2026-09 Ahrefs study — vendor-published single-source · on the wire · source ↗

Lift in a brand's AI recommendation rate when a self-published 'best of' or comparison listicle it wrote gets cited by an AI assistant, versus when it isn't — across ChatGPT, Claude, Gemini, Microsoft Copilot, Perplexity, Google AI Overviews and AI Mode. Citing the listicle also roughly doubled how often the AI named the brand outright (39% vs. 19%), though Scrunch says only the recommendation-rate figure survived a falsification test it ran on the data; a named rival was itself recommended in 24.3% of citing answers, an effect that held up in testing only when a lower-visibility brand named lower-visibility rivals, and 63.9% of citing answers recommended no brand at all — per a Scrunch analysis of nearly 10,000 AI-cited URLs sampled May-June 2026.

PPC Land, 2026-09 Scrunch study — vendor-published single-source · on the wire · source ↗ · study ↗

Share of AI answers to product-comparison questions that cited at least one commercially disclosed source, across 106,758 citations on ChatGPT, Google AI Overview and Perplexity — affiliate disclosures appeared on 21.4% of the English-language citations, and TechRadar, Forbes Advisor and PCMag drew the most citations of any publisher in the dataset, per GetCited.

PPC Land, 2026-09 GetCited sells placement on the publishers it studied, per PPC Land, which also found the dataset reuses citation data DataPulse first published in German in May 2026 without disclosing the earlier run unchecked · on the wire · source ↗ · study ↗

Next-run ChatGPT mention rate for a brand with both a prior mention and current own-domain citation exposure (80.5%) versus neither (1.6%) — Gemini's equivalent gap ran 83.7% versus 1.9% — across 34,960 unbranded prompt-engine observations from 75 anonymized client projects and 2,854 monitored prompts, per an arXiv analysis of AI-visibility platform Aiso's client data, June-September 2026. Own-domain citation alone lifted mention rate from 2.8% to 49.0% on GPT and 3.8% to 58.4% on Gemini; pairing it with a branded citation pushed rates to 91.4% and 100%.

arXiv AI-search citations, 2026-09 arXiv preprint analyzing vendor (Aiso) client data — not yet peer reviewed, single-source single-source · on the wire · source ↗

Healthcare.gov's share of ChatGPT's insurance-question citations over a month, in a BrightEdge analysis of a tracked insurance prompt panel across four question types. Insurers' own domains were the cited source only on branded searches — when a consumer names the company; on definitional and coverage questions government and clinical sources dominate instead, with drugs.com (20.6%) and nih.gov (19.1%) gaining share on coverage questions as consumer sites like WebMD (1.4%) declined.

BrightEdge Blog, 2026-09 BrightEdge study — vendor-published; panel size and prompt count undisclosed single-source · on the wire · source ↗

Median time for an AI-cited page to lose half its peak citation share, with the full rise-to-half-peak arc running 42 days, across 883,000 pages tracked over seven AI engines (ChatGPT, Perplexity, Claude, Google AI Overviews, Google AI Mode, Gemini, Microsoft Copilot) from September 2025 to September 2026. Durability varies by engine: a sustained page on Claude is 2.8 times as likely as one on ChatGPT to still hold half its peak share two months out, and pages cited by five or more engines are 5.5 times as likely as single-engine pages to stay above half their peak after eight weeks — per Profound.

Profound research blog, 2026-09 Profound study — vendor's own product data, not independent research unchecked · on the wire · source ↗

Share of Google AI Overview answers that omitted a self-ranked brand from its own "best of" comparison page in September 2026, up from 69% in June — with a competitor recommended instead in 70% of cases — per SEO consultant Lily Ray of Amsive, tracking a fixed set of 100 "best B2B software" queries.

PPC Land, 2026-09 Lily Ray/Amsive analysis — self-published, 100 queries in one vertical contested · on the wire · source ↗ · study ↗

62% per Growth Memo (Kevin Indig) — Indig's August analysis found 62% of vendor-authored listicles still gained estimated organic traffic even as Google's share of listicles holding position 1 fell — a traffic measurement, not a citation-omission rate, but it points the opposite direction from Ray's suppression finding. (wire item)

24.3% per Scrunch — Scrunch measured a named rival recommended in just 24.3% of answers citing a self-ranked listicle, well under Ray's 70% competitor-recommendation figure, across a differently built sample of nearly 10,000 AI-cited URLs. (wire item)

How many times ChatGPT-User kept pulling from a site publishing 50,000 pages of GPT-4.1-fabricated statistics over 30 days — more than 10,000 distinct AI sessions — despite Google deindexing the site within weeks of publication, per a Peec AI experiment that cost about $50 to run. The fabricated numbers went on to appear uncredited in academic papers and an investment research note from Qatar's largest financial group.

Peec AI blog, 2026-10 Peec AI study — vendor-published, single experiment single-source · on the wire · source ↗

Share of Best Buy's Black Friday citations in ChatGPT going to a dated press release versus the retailer's current store hub page, per BrightEdge — Home Depot splits a similar 63.1% to 13.3%. Across 13,915 holiday prompts tracked against ChatGPT over six major US retailers from June to August 2026, about 11% of cited holiday pages carried a year that had already passed, weighted across all six; nothing ages a dated press release out of the citation pool until the retailer replaces it.

BrightEdge Blog, 2026-10 BrightEdge study — vendor-published single-source · on the wire · source ↗

Share of an AI-search platform's citations captured by its own top 20 cited domains, across 17,211 citation instances spanning 10 platforms and 6,356 source domains in a new study — concentrated enough that posting on a handful of accessible domains can seed what a platform cites. The same study got 8 of 10 platforms to cite a fabricated concept within a week of such posting.

arXiv AI-search citations, 2026-10 arXiv preprint — not yet peer reviewed unchecked · on the wire · source ↗

How much more a native-language health query pulls locally sourced AI Overview citations than the same query asked in English, across 1,920 health queries spanning 12 countries and 4 languages on Google and Baidu, per an arXiv study — both engines also showed "vertical integration," routing citations toward their own company's platforms rather than a diverse set of primary health sources, and smaller, less-localized countries received fewer domestically sourced references regardless of language.

arXiv AI Overviews & brand visibility, 2026-09 arXiv preprint — not yet peer reviewed unchecked · on the wire · source ↗

Median cosine similarity between Baidu and Google AI-overview answers for matched query intents in a cross-lingual audit using MS MARCO queries and their Chinese translations — yet the two platforms' overviews showed little overlap in which domains they cited, showing answer similarity and source exposure are separate things to track.

arXiv AI-search citations, 2026-09 arXiv preprint — accepted at WAC@EMNLP 2026 workshop; not yet peer-reviewed published unchecked · on the wire · source ↗

How far the top AI-recommended brand swung between languages for the same question, testing Claude Sonnet 5 on six shopping and travel prompts across English, Arabic, Russian, Chinese and Persian (1,650 responses) — Claude picked British Airways 52.5% of the time in Persian for a London-to-Tenerife flight and never in Russian, where it favored easyJet 72.5% of the time instead. A smaller comparison test found Gemini's top pick matched Claude's in 5 of 10 language-category combinations, ChatGPT in just 1.

Search Engine Journal, 2026-09 single-source — confirmation pending; author cautions some results may sit within margin of error on the 1,650-response dataset single-source · on the wire · source ↗

Asda's rank among ChatGPT's most-mentioned UK brands in July 2026 (8th) versus its rank for citations (606th) — Tesco ranked 9th for mentions and 25th for citations, Ford.com 18th and 1,055th — a mention/citation split Semrush's first UK benchmark of its AI Visibility Index found held across ChatGPT, Gemini, Google AI Mode and Google AI Overviews. The same release found any two of the four platforms shared just 63 to 70 of their top 100 mentioned UK brands, versus 70 to 78 in the US.

PPC Land, 2026-10 Semrush study — vendor-published single-source · on the wire · source ↗ · study ↗

Share of ChatGPT's holiday-shopping citations going to a retailer's newsroom or customer-support pages, versus 18.1% for Google AI Overviews on the same retailers and queries — AI Overviews instead put 64.6% of its citations on commercial product pages versus 35.4% for ChatGPT, a 12.5x gap on help-page citations in ChatGPT's favor. BrightEdge compared the top 50 cited rows per domain on each engine for Black Friday, Cyber Monday, Christmas and Thanksgiving prompts across four US retailers; ChatGPT's holiday prompts also averaged 6.0-word logistics questions versus 4.4-word keyword queries on Google.

BrightEdge Blog, 2026-10 BrightEdge study — vendor-published single-source · on the wire · source ↗

Share of e-commerce brands that never appear at all in their own category across ChatGPT, Gemini and Perplexity — rising to 84% on problem-solving prompts — in a CommerceV3 study of more than 1 million prompts (12 million prompt-runs) spanning seven categories and roughly 25,000 brands, March-July 2026. The same study found 75% of recommendation slots go to the same handful of brands per category, and that AI platforms source answers mostly from editorial sites (34%) and Reddit/forums (22%), with a brand's own site cited only 14% of the time.

CommerceV3, 2026-10 CommerceV3 study — vendor-published (AI-visibility vendor) single-source · on the wire · source ↗

Share of Google AI Overview answers citing at least one social or community source (roughly two in three) versus Google AI Mode (about one in three), across 100,000 keywords spanning 20 niches. YouTube was the single most-cited domain overall (16.2% of citations); LinkedIn's share jumped to 21.42% on career/jobs-topic answers versus 3.15% across all topics; AI Mode cited social sources in 48.58% of non-branded answers versus 39.34% for branded ones — per SE Ranking, published via Planable's blog (SE Ranking's own SE Visible tool ran the AI-citation side of the study).

SE Ranking, 2026-10 SE Ranking study — vendor-published single-source · on the wire · source ↗

Share of matched AI-answer bank comparisons across Southeast Asia that named a different bank first when only the AI service changed and the question stayed the same — 706 of 1,628 eligible comparisons across 11 AI and search services in six markets. In Indonesia, BCA appeared in 86% of tested AI answers, the highest overall presence of any bank in the market, and led most often on personal-banking and trust questions, while Bank Mandiri led on business-banking and economic-contribution questions despite lower overall presence — per PRecious Communications' AI Authority Index: Southeast Asia Banking 2026, covered by ContentGrip. The study also covered Singapore, Malaysia, the Philippines, Thailand and Vietnam across 16 banks and 8 languages, but PRecious's own landing page withholds those country-level figures behind a lead-capture form; only the Indonesia breakdown is public.

ContentGrip, 2026-10 PRecious Communications study — vendor-published single-source · on the wire · source ↗ · study ↗

Croatia's official tourism site, croatia.hr, did not rank among the 50 most-cited sources out of more than 3,000 citations drawn from 628 sites in a Determ study of 556 AI responses to UK seaside-holiday questions across ChatGPT, Gemini and Google AI Overviews (Oct 1-16, 2026) — Facebook (143), YouTube (100) and Instagram (83) led, while Spain's and Greece's official tourism sites did rank. Croatia itself appeared in 67% of seaside-holiday answers overall (80% on ChatGPT, 67% on Gemini, 53% on AI Overviews), fourth behind Spain (71%) and Greece/Italy (70% each).

Croatia Week, 2026-10 Determ study — vendor-published single-source · on the wire · source ↗ · study ↗

Share of B2B marketers who say their brand gets cited in AI-generated answers among those who integrate influencer marketing with AI search optimization (48%) versus those who don't (26%) — the same split held for referral traffic from AI assistants, 48% versus 21% — per a TopRank Marketing/Ascend2 survey of 424 B2B marketers at US companies with 100 or more employees, fielded mid-May to mid-June 2026 for the 2027 B2B Influencer & Creator Marketing Report. Only 37% of respondents had made the connection between influencer marketing and AI search optimization, even though 86% called improving generative AI visibility important this year.

TopRank Marketing / Ascend2, 2026-10 TopRank/Ascend2 study — vendor-published single-source · on the wire · source ↗

Hilton's visibility share among hotel brands on Google AI Overviews for holiday-travel queries (28.3%) versus Marriott's share on ChatGPT instead (29.4%) — leadership splits by engine rather than converging on one winner, echoing BrightEdge's Oct. 16 finding for retail citations. Among booking platforms, Expedia outpaces Kayak on AI Overviews, 31.3% to 14.4%, and booking-focused sites appear in 78% of budget-related AI Overviews answers. Per BrightEdge's AI Hyper Cube, covering inspiration, comparison, booking-intent and budget queries collected August-September 2026.

BrightEdge, 2026-10 BrightEdge study — vendor-published single-source · on the wire · source ↗

Share of ChatGPT's answers to local-topic questions (crime, planning, housing, council decisions, local events) citing a local news source (0.9%) versus a government or council site (81.3%), across 30,000 prompts covering 100 UK towns — on crime questions specifically, police force websites alone supplied 57.8% of answers, and of ChatGPT's top 20 news sources for local queries only one, Wales Online, was a local title. Per a Searchable study reported by Press Gazette; Searchable's CEO said ChatGPT draws on an average of 12.3 websites per query but cites only four, with one in seven background searches reaching a local news provider against one in three reaching an official body.

Press Gazette, 2026-10 Searchable study — vendor-published single-source · on the wire · source ↗ · study ↗

Share of citations going to Lebanese sources in Arabic-language AI search answers about Lebanon (56.7%) versus English-language answers to the same questions (31.6%) — a 25.1-point gap — testing ChatGPT Search, Gemini, Perplexity and Copilot on 40 equivalent questions (320 responses total). Gemini showed the widest language gap at 34.8 percentage points, ahead of Perplexity (26.8), Copilot (20.4) and ChatGPT Search (18.5); international organizations drew 21% of English citations versus 12% of Arabic ones, while news and media sources drew 24% of Arabic citations versus 10% of English. Per computational researcher Mohamed Soufan's own write-up.

HackerNoon, 2026-10 single-source — confirmation pending single-source · on the wire · source ↗ · study ↗

Share of Australian vet clinics with 200 or more reviews that ChatGPT, Gemini and Google AI Mode recommended when asked for a local vet (99%) versus clinics with fewer than 50 reviews (37%), across 360 AI answers to pet-owner questions spanning 60 locations and 2,785 active clinic listings. Claimed Google Business profiles were recommended 79% of the time versus 17% for unclaimed ones. Per Seenmetric, reported by Pet Industry News.

Pet Industry News, 2026-10 vendor study — vendor-published (Seenmetric is an AI-visibility consultancy; study covers its own service category) single-source · on the wire · source ↗

How much more often Google Shopping appeared than the next-largest comparison service in the Comparison Sites section that replaced Google's free product grid across Germany, France, Spain and Italy, landing in the top four results 71% of the time overall. The grid vanished from all 14 EEA markets BrightEdge tracks on Sept. 16, 2026, the same day the section appeared, following the European Commission's €460 million DMA fine against Google for favoring its own shopping results. Per BrightEdge's analysis of daily search results, Sept. 1 to Oct. 13, 2026.

BrightEdge, 2026-10 BrightEdge study — vendor-published single-source · on the wire · source ↗

Share of Google traffic reaching news publishers via Discover versus Web Search in Q4 2025, across 400+ publishers tracked by NewzDash — reversed from 2023, when Search sent 51% and Discover 37%. NewzDash CEO John Shehata calls it "the Great Flip." Discover carries no queries, so most of a news publisher's Google traffic now arrives from a surface with nothing to optimize a keyword against.

Advanced Web Ranking, 2025-12 NewzDash analysis — vendor-published, not independently replicated; Q4 2025 data single-source · on the wire · source ↗

How often AI Overviews trigger on US breaking and political news queries versus health queries, per John Shehata — with sports around 18%. Exposure to AI answers is a property of the beat rather than of news as a category, so a single figure for "news" averages desks in materially different businesses.

Advanced Web Ranking, 2026-01 NewzDash analysis — vendor-published, not independently replicated; Q4 2025 data single-source · on the wire · source ↗

Weekly clicks Google says its AI Search features send to websites — a claim published without methodology, against third-party studies showing CTR declines.

Search Engine Land, 2026-07 no methodology published unchecked · on the wire · source ↗ · study ↗

Alphabet's Q2 2026 Search revenue, up 17% year over year — the first deceleration after four quarters of accelerating growth.

Search Engine Journal, 2026-07 unchecked · on the wire · source ↗ · study ↗

Commercial-keyword AI Mode queries showing ads in a July analysis — while advertisers were rarely among the answer's cited sources.

Search Engine Journal, 2026-07 unchecked · on the wire · source ↗ · study ↗

Share of more than 50,000 commercial ChatGPT prompts across 20 niches that showed ads, per an SE Ranking study — with 14.35% of those ads unrelated to the surrounding conversation, and only 3.63% of advertisers whose ads ran also cited as a source in the answer above them.

Search Engine Land, 2026-08 SE Ranking study — vendor-published single-source · on the wire · source ↗ · study ↗

Share of U.S. Google searches that sent a click to the open web in June 2026 (40.7% in the EU/UK) — a record low — while Google AI Mode held at just 0.13% of U.S. search-related visits, per Datos and SparkToro's Q2 State of Search report.

Search Engine Journal, 2026-08 Datos/SparkToro study — cited via Search Engine Journal, methodology not published unchecked · on the wire · source ↗ · study ↗

Year-over-year decline in People Inc.'s Google search traffic — now 21% of its total traffic, down from 25% the prior quarter, with core sessions off 22% YoY — even as the publisher's CEO says it won't block Google's AI crawlers because that "would turn off search."

Digiday, 2026-08 unchecked · on the wire · source ↗ · study ↗

Le Monde's English-language traffic growth in H1 2026 versus H1 2025, roughly seven weeks after Google's AI Overviews launched in France — CEO Louis Dreyfus also cited 24,000 new subscribers in each of July and August 2026. He contrasted this with French publishers' group APIG's reported 33-38% search-referral loss, saying search drives only about half of a major news site's visits and that the steepest AI Overviews damage is concentrated in service-content sectors — health, food, automotive — down as much as 50%, not general news.

Search Engine Journal, 2026-09 single-source — confirmation pending single-source · on the wire · source ↗ · study ↗

Estimated decline in monthly Google search referrals to English Wikipedia since AI Overviews became the US default in 2024, per a University of Washington study by Mehrzad Khosravi and Hema Yoganarasimhan — one of the few non-vendor-funded estimates of AI Overviews' referral impact. Google disputes the figure, saying Wikimedia's public referral data mixes traffic from all search engines and can't isolate Google's effect; the authors counter that other engines carry too little share to explain the gap.

Search Engine Journal, 2026-09 single academic study, disputed by Google unchecked · on the wire · source ↗ · study ↗

~15% as of 2026-09 — the authors' own earlier draft, using daily pageviews instead of monthly referrals, before they added international controls and revised down (wire item)

Year-over-year growth in AI-driven traffic and orders to Shopify merchant stores in Q2 2026, per Shopify president Harley Finkelstein — with half of AI-referred sessions landing directly on product pages, 2.5x the rate of traditional search, while traditional search sessions stayed up 1.3x over two years. Shopify frames AI as additive to search, not a replacement.

TechCrunch AI, 2026-08 single-source — confirmation pending single-source · on the wire · source ↗ · study ↗

Share of U.S. holiday shoppers who say they regularly or occasionally consult AI tools such as ChatGPT, Claude or Gemini for gift ideas, with 49% doing so regularly — per an Optimove survey of 648 U.S. consumers; 37% cited data privacy as the top barrier to using AI tools more, even though nearly three-quarters trust AI-generated recommendations as much as or more than other sources.

PPC Land, 2026-08 Optimove survey — vendor-published, n=648 single-source · on the wire · source ↗ · study ↗

Google's share of foundation model API revenue as of December 2025, with Google, OpenAI and Anthropic together controlling 88% — cited in an amicus brief urging the D.C. Circuit to toughen, not weaken, the remedy in Google's search antitrust appeal.

PPC Land, 2025-12 figure as cited in an advocacy brief, not independently verified unchecked · on the wire · source ↗

Blood Cancer UK's year-over-year drop in page views for its leukaemia information page, with declines of 17-45% across its other blood cancer pages and community-forum visits up 86% over the same span, per Press Gazette; Save the Children logged a 303% YoY rise in AI Overview appearances in July 2026 while clicks and impressions fell, and the Brain Tumour Charity reported a roughly 27% site-wide traffic drop.

PPC Land, 2026-08 single-source — confirmation pending single-source · on the wire · source ↗

Rise in Perplexity's India revenue from February through mid-August 2026 versus the period when its free 12-month Pro offer via Airtel was still available — usage has settled near 14 million monthly active users, still about 5x pre-deal levels, even as new downloads fell 90% after redemptions closed in January 2026, per Sensor Tower data cited by TechCrunch.

TechCrunch AI, 2026-08 Sensor Tower estimate via TechCrunch — not Perplexity-reported unchecked · on the wire · source ↗

Drop in Google AI Overviews' position-1 click-through rate in 2026, up from a 34.5% drop in 2025, per an Ahrefs analysis spanning roughly 15 million data points.

Search Engine Journal, 2026-08 Ahrefs study — vendor-published single-source · on the wire · source ↗ · study ↗

Average share of Google AI Overview citation-click traffic misattributed to the Direct channel instead of Organic Search in GA4, per a nine-month tracking study of 51,200 clicks across 1,661 AI Overview-cited snippets — worst in May 2026 at 29.3%. The same tracking found AI Overviews drove 7.53% of organic sessions overall, peaking at 16-17% in February-March 2026 before falling to 2-4% by August.

Search Engine Land, 2026-08 single-source — confirmation pending single-source · on the wire · source ↗

Share of observed US ChatGPT ad placements going to retail and fashion queries versus their 24% share of query volume, per an Adthena analysis of nearly 850,000 US and UK AI search queries from March to May 2026 — ads overall appeared on 4.47% of US queries at 1.06 items per response, below the 3.53-item average Adthena measured on Google's AI surfaces; logistics (12.41%), home-and-garden (11.99%) and beauty (10.03%) had the highest per-category ad frequency.

PPC Land, 2026-08 Adthena study — vendor-published single-source · on the wire · source ↗ · study ↗

ChatGPT Ads' annualized revenue run rate, reached in under 200 days after launch, per OpenAI — a self-reported, unaudited figure. OpenAI opened self-service Ads Manager access to India, Europe, the Middle East and North Africa the same day, previously a managed-sales and partner-led market.

OpenAI News, 2026-08 OpenAI-reported figure — annualized run rate, not audited revenue unchecked · on the wire · source ↗

Share of AI-referred shoppers who switch to a competitor when the AI-recommended website disappoints them — 16% abandon the purchase outright and only 3% buy anyway — per a Contentsquare survey of 2,000 US and French consumers; 21% said they'd seen AI-provided product information contradict what a brand's own website showed.

PPC Land, 2026-09 Contentsquare survey — vendor-published; fieldwork dates and margin of error undisclosed single-source · on the wire · source ↗ · study ↗

How much higher Google AI Mode's shown prices ran versus the same items in Search's product carousel, per a Productrise study of 2 million-plus product listings tracked August 9-31, 2026 — the two surfaces showed the identical product on only 1.28% of matched queries, and listed a different lead seller 49.6% of the time when products did match. Google told Search Engine Journal all its shopping surfaces draw on the same Shopping Graph data but did not address the price or seller gap.

Search Engine Journal, 2026-09 Productrise study — vendor-published single-source · on the wire · source ↗ · study ↗

Financial services' share of total US ChatGPT ad spend, April to August 2026, as the category grew from one top-100 advertiser to four of the top ten spenders — while shopping and retail's share fell from 37% to 21% over the same span, per Sensor Tower. Ad density (ads shown per user per hour) rose 163% over the period, averaging 26% monthly growth, with roughly 1,200 unique advertisers active by August.

PPC Land, 2026-09 Sensor Tower study — vendor-published single-source · on the wire · source ↗

Share of hotel-related ChatGPT prompts carrying a sponsored link, March to May 2026 — a fourfold increase in three months — per Comscore, which began tracking sponsored ChatGPT links using opt-in panel data on real prompts rather than synthetic tests. Panel size and composition undisclosed.

PPC Land, 2026-09 Comscore data — vendor-published single-source · on the wire · source ↗ · study ↗

Share of gaming-related ChatGPT prompts carrying a sponsored ad, March to June 2026 — the steepest category climb Comscore has tracked, ahead of the 6% to 24% hotel-prompt rise over March to May — per Comscore's Q2 2026 AI Intelligence Report. Toys-and-hobbies rose 5% to 24% and books-music-video 9% to 22% over the same window. Panel size and composition undisclosed.

Search Engine Journal, 2026-09 Comscore study — vendor-published; panel size not disclosed single-source · on the wire · source ↗

Feed-integrated retrieval's single-day jump in ChatGPT Shopping mode product recommendations on July 10, 2026 — the day ChatGPT 5.6 shipped — reaching roughly 65% of recommendations by late August and overtaking web search as the leading source, per a Profound analysis of 687 customers. Top-10 merchants' share of shopping-mode product cards rose from 22.5% to 41.8% over the same period.

Profound, 2026-09 Profound — vendor's own product data, not independent research unchecked · on the wire · source ↗

Decline in search-referral traffic to mid-sized regional publishers, December 2024 to December 2025 — roughly double the decline measured at larger national outlets — per industry data cited in the Seattle Times and Newsday's joint copyright and trademark-dilution complaint against OpenAI and Microsoft.

Chat GPT Is Eating the World, 2025-12 figure as cited in a litigation complaint — underlying study and methodology not identified unchecked · on the wire · source ↗

Google Shopping ads' median click-through rate, mid-2025 to mid-2026, even as impressions fell from roughly 1.85 million to 1.4 million — the inverse of the "crocodile effect" seen in organic search, where AI Overviews have lifted impressions while cutting clicks. Analysis of 175 billion impressions by Mike Ryan of Smarter Ecommerce; Google has not confirmed his theory that AI Overviews are increasingly served on lower-click-probability queries.

Search Engine Roundtable, 2026-09 Smarter Ecommerce study — vendor-published single-source · on the wire · source ↗ · study ↗

Estimated Google paid search spend by Similarweb's Top 100 Media publishers in July 2026, up 41% year over year and 274% over three years — Forbes alone accounted for $72.2 million, about two-thirds of the total, even as its organic traffic fell 26.7% year over year; CNN and USA Today posted organic declines of 28.9% and 24.1% over the same period, per Similarweb data reported by Adweek.

PPC Land, 2026-09 Similarweb data — vendor-published, reported by PPC Land and Adweek single-source · on the wire · source ↗

CTR lost by French sites most exposed to AI Overviews (over 20% of their queries triggering one), comparing the 28 days before to the 9 days after AI Overviews launched in France on July 22, 2026; the median site across 963 tracked domains lost 5.7% of CTR and 8.8% of clicks, and nearly 48% of sites lost more than 10% of daily clicks within nine days. Health queries lost the most — a 22.2% AI Overview trigger rate and a 20.4% CTR decline. Ahrefs' Juliette Begue, using Google Search Console data, found roughly "one point of CTR lost per point of exposure."

Ahrefs Blog, 2026-09 Ahrefs analysis of Google Search Console data — vendor-published, 9-day post-launch window, not independently replicated single-source · on the wire · source ↗

Reduction in outbound clicks per search when Google shows an AI Overview — from 0.62 down to 0.37 — in the first randomized field experiment on the question: 1,065 U.S. desktop searchers randomly assigned to see results with or without AI Overviews over two weeks in early 2026. Self-reported search satisfaction held flat at 4.0 out of 5 either way, and just 7.1% of outbound clicks on AI Overview-bearing pages came from links inside the AI-generated answer itself, with the rest going to ordinary organic results on the same page.

Digital Content Next, 2026-09 SSRN working paper by Saharsh Agarwal and Ananya Sen — not yet peer reviewed; first posted April 2026, revised June 2026 unchecked · on the wire · source ↗ · study ↗

Washington Post readers in a randomized field experiment who got AI-generated answers with article citations placed above conventional results — they consumed a wider range of topics, individually and as a group, while also increasing what they had in common, mostly through the AI answers themselves rather than added article clicks.

arXiv AI-search citations, 2026-09 arXiv preprint by Heeseung Andrew Lee, Dokyun Lee, Gwanhoo Lee and Dongwon Lee — not yet peer reviewed unchecked · on the wire · source ↗

Rise in ChatGPT's weekly advertiser count over eight weeks, from 7,306 in the week of July 9, 2026 to 12,075 by August 31 — peaking at 13,009 in early August — per an Adthena analysis of AI search queries.

PPC Land, 2026-08 Adthena study — vendor-published contested · on the wire · source ↗

~1,200 unique advertisers active per Sensor Tower — Same late-August window, a different method: Sensor Tower estimates active advertiser accounts from ad-spend data, Adthena counts distinct advertisers observed across a weekly sample of AI search queries. Which better answers 'how many advertisers are on ChatGPT' is unresolved. (wire item)

Share of Google AI Overview ad appearances generated by one- and two-word search queries, in every market Adthena measured as of late August 2026 — reversing a June 2026 pattern in which three- and four-word queries had the majority of AI Overview ad appearances.

PPC Land, 2026-08 Adthena study — vendor-published single-source · on the wire · source ↗

Average share of advertising revenue Google pays roughly 100 publishers in its Search Console AI contribution pilot, per The Information — payouts range from under $1,000 over several months to more than $1 million a year depending on the publisher, and participants say Google hasn't explained how the monthly totals are calculated.

Search Engine Roundtable, 2026-09 The Information reporting via Search Engine Roundtable — payment rate and publisher count previously undisclosed by Google unchecked · on the wire · source ↗ · study ↗

Invalid-click rate on ChatGPT ads by advertiser, per ad-verification vendor TrafficGuard's analysis of 40,021 ChatGPT ad clicks across 78 websites run by 47 enterprise advertisers, August 24-September 14, 2026. Hosting, proxy and malicious-IP traffic made up 3.34% of ChatGPT ad clicks versus 0.92% of the same advertisers' Google Ads clicks over the same period — a 3.6x gap — and four hosting and proxy networks reached 25 of the 47 advertisers while driving 47% of flagged clicks. The figures count clicks that reached advertiser websites, not clicks OpenAI billed for, so the study doesn't establish whether advertisers were charged for the flagged traffic.

PPC Land, 2026-10 TrafficGuard study — vendor-published (ad-verification vendor); counts clicks reaching advertiser sites, not clicks OpenAI billed single-source · on the wire · source ↗

U.S. Google AI Mode ad placements carrying a prescription-drug brand in Adthena's August 2026 tracking, out of roughly 50,000 placements — until Adthena spotted an unbranded Eli Lilly ad for an oral GLP-1 weight-loss pill inside an AI Mode answer on October 11, the first live pharma ad it has found in the surface. The ad never named the drug, an unbranded format that sidesteps the risk disclosures a named-drug ad would carry under FDA rules.

Adthena, 2026-10 vendor-published — Adthena ad-intelligence tracking data single-source · on the wire · source ↗

Share of ChatGPT answers with cited links carrying a tagged chatgpt.com source link, which fell from about 98% on October 8, 2026 to a low of 5-10% on October 14 evening, then recovered through 16% on October 15 and 51% early on October 16 to about 99% by October 16, per Profound's 30-minute-interval tracking. Profound initially tied the swing to the rollout of its Intelligent UI answer format rather than a bug; Adobe senior director Cedric Huesler says OpenAI told him on October 14 it was in fact a bug, with a fix rolling out around 2 a.m. PT on October 15 and tagged-click volume back to roughly 95-100% of the prior week's levels by 9 a.m. PT that day. OpenAI has not commented publicly.

Search Engine Journal, 2026-10 secondhand account via an Adobe exec; OpenAI has not confirmed publicly unchecked · on the wire · source ↗

98% to 66% as of 2026-10 — the initial drop, which Profound attributed to the Intelligent UI rollout (wire item)

16% to ~99% as of 2026-10 — the recovery Profound tracked while still attributing the cause to Intelligent UI rather than a bug (wire item)

Google Search referral traffic to publishers fell 40.2% year-over-year from July 2025 to July 2026, nearly double the prior year's 21.9% decline, across Chartbeat's client network — which skews toward news and media publishers — per Chartbeat's 2026 Publisher Playbook. AI referrals grew eightfold over two years but still made up just 0.01% of publisher traffic, and declined across ChatGPT, Claude, Perplexity, Copilot and Gemini over 2026: AI traffic has not come close to offsetting what Search stopped sending.

Chartbeat, 2026-09 Chartbeat 2026 Publisher Playbook — vendor-published, catch-up (published Sep 24); an earlier version of this item misattributed a publisher-size breakdown to this report that the report does not contain, corrected 2026-10-17 and removed here single-source · on the wire · source ↗

Share of Direct-channel sessions carrying an identifiable AI trace, per a Growth Memo analysis of GA4 data across 12 websites from January through September 2026 — illustrated by one site that logged 409,000 Direct sessions of which only 198 counted as engaged. Raw Direct traffic rose 140.9% over that period; engaged (non-bot) Direct traffic fell 1.0% in raw monthly counts but rose 2.3% once normalized for days per month — in both readings, the growth was overwhelmingly bot activity, not hidden AI referrals. Testing 632 comparisons of AI-visibility metrics against branded-search and Direct traffic across four brands, correlations as high as 0.72 failed to hold up against randomized control data, suggesting the pattern is likely spurious rather than causal.

Growth Memo (Kevin Indig), 2026-10 single-source — one practitioner's 12-site GA4 analysis; the wire covered the same Growth Memo analysis a second time on 2026-10-21; that copy was merged into this item on 2026-10-23 (its 2.3%/409,000-sessions detail is in the item's correction) single-source · on the wire · source ↗

Share of 1,000 UK and US digital and e-commerce decision-makers who say AI-generated search experiences have already cut their website traffic over the past 12 months, with 82% saying AI-powered recommendations are changing how customers find brands and 77% saying traditional SEO alone no longer guarantees visibility, per a Cloudways survey fielded by Censuswide.

Cloudways, 2026-10 Cloudways survey — vendor-published, self-reported, n=1,000 single-source · on the wire · source ↗

ChatGPT referral traffic converted at 13.8% against Google organic's 3.02% in a 145-million-session GA4 analysis across 100 websites over 12 months, per Marie Haynes — even though ChatGPT traffic equaled just 0.62% of Google organic's volume and organic still produced 96.5% of the tracked conversions. A second analysis, by SEO Works across 95 sites (72,095 AI referral sessions against 2.8 million organic, September 2025 to August 2026), found the same direction at a smaller gap: AI referrals converted at 7.18% against organic's 6.08%, while organic delivered 39 times the traffic.

Marie Haynes, 2026-10 practitioner GA4 analyses, not peer-reviewed; two different site panels and methodologies corroborated by SEO Works · on the wire · source ↗

Bloomsbury's Academic & Professional adjusted profit margin fell from 23.9% to 13.9% (profit down from £11.0m to £6.8m) in the six months to Aug. 31, 2026, because its non-exclusive AI content-licensing agreement continued into the new year at a lower value, per Bloomsbury's own interim results — the first look on the record at AI licensing income as something that resets rather than simply recurs. The release does not name the AI licensing counterparty.

Bloomsbury Publishing (interim results), 2026-10 unchecked · on the wire · source ↗

Mentions of Philippine brands in Google AI Overviews rose more than 23-fold in 2025, per a 12-month Truelogic study tracking 80 brands across seven sectors (retail, e-commerce and real estate saw the strongest gains), while organic branded web traffic across the panel dipped 3-8% over the same period — a correlation the study does not claim as causal. Among 40 of Truelogic's own client projects, AI visibility rose 750% between January and July 2025, against a 400% nationwide average it measured.

adobo Magazine, 2026-10 Truelogic study — vendor-published; no primary study page found on Truelogic's own site, only press-release pickup across trade outlets single-source · on the wire · source ↗

Conversations in Google's AI & Economy ATLAS report, which found AI questions about government services occur roughly 20x more often than time spent on them would predict.

Google / Search Engine Journal, 2026-07 vendor study unchecked · on the wire · source ↗ · study ↗

Share of medical, legal, financial and government queries to Google's AI surfaces arriving outside work hours — high-friction tasks, handled at night and on weekends.

Google / Search Engine Journal, 2026-07 vendor study unchecked · on the wire · source ↗ · study ↗

Share of ChatGPT conversations carrying commercial intent as of June 2026, up from 13.9% a year earlier — weekly commercial conversations grew from roughly 243 million to 533 million over the same span, a pace Profound says puts ChatGPT near 28 billion commercial conversations a year. Automotive (59.1%) and travel/hospitality (56.1%) topped the industry list; education and non-profit topics were lowest at near 6%.

Profound research blog, 2026-08 Profound study — vendor's own product data, not independent research unchecked · on the wire · source ↗

Average share of AI retrievability and trust criteria met by 71 audited small businesses — and 17% had no AI-retrievable digital presence at all.

Search Engine Land, 2026-07 audit study — vendor-published single-source · on the wire · source ↗

Domains in Ahrefs' llms.txt adoption analysis — the dataset behind its free llms.txt generator.

Ahrefs, 2026-07 unchecked · on the wire · source ↗

Share of sites' llms.txt files AI crawlers never read at all, per a separate Ahrefs analysis spanning roughly 15 million data points — which also found adding schema markup produced no measurable lift in AI citations across 1,885 pages tested over 30 days, and that 88.46% of AI citations still traced to pages already in the general search index.

Search Engine Journal, 2026-08 Ahrefs study — vendor-published single-source · on the wire · source ↗ · study ↗

Share of 584,107 llms.txt files in Common Crawl's July 2026 archive produced by templates or SEO plugins rather than hand-authored — Wix alone accounts for 41% of the corpus, and only about half fully follow the spec's structure of a title, summary line, and link sections.

Common Crawl Blog, 2026-08 unchecked · on the wire · source ↗

Share of the top 1,000 most-visited sites serving a valid llms.txt file with correct headers as of October 2026, per LintLab — a third independent dataset, after Ahrefs and Common Crawl, finding adoption still stuck in single digits; nearly two-thirds of the sites that attempted one returned a 404 instead.

PPC Land, 2026-10 LintLab study — vendor-published, data not independently verified single-source · on the wire · source ↗ · study ↗

Share of the 400 top-1,000 sites with an XML sitemap that failed at least one of LintLab's checks — most often a sitemap redirecting to a different URL (23%, 92 sites) — in the same October 2026 crawl.

PPC Land, 2026-10 LintLab study — vendor-published, data not independently verified single-source · on the wire · source ↗ · study ↗

Studies in the first critical survey of generative engine optimization research, 2023–2026 — the discipline now has a literature.

arXiv, 2026-07 unchecked · on the wire · source ↗

Share of brands that currently track their AI visibility, per the IAB's new "Measuring Visibility in the AI Era" framework — even as more than 20 vendors now sell tools for it.

PPC Land, 2026-08 unchecked · on the wire · source ↗ · study ↗

Share of 10,095 live Google Search and Gemini pages carrying detectable generative-engine-optimized content in a GEOFlagBench study — rising to 16.36% among pages modified in 2026; the paper's ModernBERT-based detector reached 0.944 F1 across a 3,200-page benchmark spanning eight GEO optimizer families.

arXiv GEO, 2026-08 unchecked · on the wire · source ↗

Maximum relative reduction in attack success rate from three off-the-shelf safety guardrails (Granite Guardian, Llama Guard 3, NeMo Self-Check Fact-Checking) tested against GEO-optimized misinformation in a new Counter-GEO-Bench benchmark of 247 human-verified queries — Granite Guardian's reduction wasn't statistically significant. The guardrails are built to catch policy violations, not fluent, factually wrong text; the authors' own proposed defense, C-GEO Guard, cut attack success by 47.6% relative with near-zero utility loss.

arXiv GEO, 2026-09 arXiv preprint accepted to EMNLP 2026 — the three victim LLMs tested are not named unchecked · on the wire · source ↗

Attack success rate against generative search systems after GEO Defender — a reranker trained to demote GEO-rewritten documents paired with a training-free source-selection guide — versus 50.32% unprotected, tested across five LLMs and seven GEO attack variants while retaining 94.12% of benign-source usage.

arXiv GEO, 2026-09 arXiv preprint — not yet peer reviewed unchecked · on the wire · source ↗

Within-query Spearman correlation between a deterministic GEO content-quality score and actual AI citations, per a study that re-tested the field's original 2023 citation-lever effect sizes against ten modern AI engine families and found none of them moved citations — positioning such scores as content-quality filters rather than citation predictors. On a 500-source adversarial benchmark, amplifying the levers that survived recalibration gained an attacker at most 6 points.

arXiv GEO, 2026-09 arXiv preprint — not yet peer reviewed unchecked · on the wire · source ↗

How much more domain-level AI authority mattered than the next-strongest page feature in SHAP value, per a study of roughly 2 million AI citations across 10,000 pages from 19 B2B SaaS workspaces — prompt-content alignment (word overlap with real user prompts) was the strongest per-page predictor tested (beta +0.37), while traditional SEO signals like FAQ blocks, structured data and Core Web Vitals showed positive citation correlations that reversed or fell to zero once domain-level effects were controlled for.

arXiv, 2026-09 Discovered Labs study — vendor-published (the study's authors run an AEO/GEO agency), arXiv preprint not yet peer reviewed single-source · on the wire · source ↗

Citation-odds multiplier for a page carrying the current year in its title, versus a matched, uncited page on the same domain — the strongest of 43 page-level features tested, per Sistrix's conditional logistic regression across 48,527 pages cited by ChatGPT, Google AI Mode and AI Overviews in Q2 2026, matched against uncited same-domain pages (96,418 URLs, 17,380 domains). Other factors: a table of contents, used by only 14% of pages, carried a 1.7x factor; Article schema markup 1.6x; five or more H2 headings 1.6x; FAQPage schema 1.5x. Pages carrying many of these features were cited 2.1 times as often as minimally optimized ones; on ChatGPT specifically, a question-phrased title carried a separate 1.9x factor. The dataset covers the German-speaking (DACH) market.

Sistrix Blog, 2026-10 Sistrix study — vendor-published; DACH market only contested · on the wire · source ↗

no measurable lift per Ahrefs — Ahrefs found schema markup produced no measurable AI-citation lift across 1,885 pages over 30 days. Sistrix's design, which matches each cited page against an uncited page on the same domain, still finds a 1.6x factor for Article schema and 1.5x for FAQPage schema. (wire item)

fell to zero once domain controlled per arXiv (Discovered Labs) — That study found schema and FAQ blocks' apparent positive correlation with citations collapsed once domain-level AI authority was controlled for. Sistrix's same-domain matched-pair design controls for domain by construction and still finds schema (1.6x) and FAQPage (1.5x) factors — a disagreement between two domain-controlled studies, not a reconciliation. (wire item)

Share of JavaScript-only linked pages found by GPTBot, ClaudeBot and Bingbot in a 41-day crawler-discovery test on a 2,400-page site — versus 2% for Googlebot and 48% for Google's separate, JS-capable GoogleOther crawler; converting the links to plain HTML let GPTBot recover 250 previously unreachable pages within 48 hours.

Search Engine Land, 2026-08 single-source — confirmation pending single-source · on the wire · source ↗

Share of English-language webpages published after ChatGPT's November 2022 launch that showed significant signs of AI authorship in a July 2026 Common Crawl snapshot (10% across all sampled pages) — commercial (.com) domains showed AI-authorship signs at roughly 10 times the rate of .edu or .gov domains (near 1% each), with .org domains at 4.6%, per Pew Research.

Pew Research, 2026-07 AI authorship detected via the Open Pangram classifier — detection accuracy not independently verified unchecked · on the wire · source ↗

Share of 4,776 Bali food-service venues that ChatGPT, Gemini, Claude and Perplexity failed to recommend across 2,208 queries in a Norly Research market census — having a business website nearly doubled a venue's odds of recommendation, while star rating had no effect on whether a venue appeared at all; the four systems also recommended 14 permanently closed venues a combined 93 times.

PPC Land, 2026-08 Norly Research study — vendor-published, funded and conducted by a review-management/AI-visibility vendor single-source · on the wire · source ↗ · study ↗

Share of 343 U.S. marketing decision-makers who still refer to their AI search strategy internally as "SEO," with only 27% having formally adopted newer terminology, even as they allocate an average of 24% of search and content budgets to AI search visibility (43% allocating more than a fifth), per a Fractl survey.

Search Engine Land, 2026-08 Fractl study — vendor-published, author is a Fractl co-founder single-source · on the wire · source ↗ · study ↗

Share of the top 20,000 websites currently set up to serve AI agents optimized content formats such as Markdown — 97% aren't, per a BrightEdge audit published alongside its new Agent Edge product, which detects agent requests at the CDN layer and serves a lightweight Markdown version of pages without changing what human visitors see.

BrightEdge Blog, 2026-08 BrightEdge study — vendor-published, audit accompanies its own product launch single-source · on the wire · source ↗

30-day transaction volume on x402, the HTTP-402-based agent payment settlement protocol — 75.41M transactions across 94.06K buyers and 22K sellers, per the project's own site, which lists adopters including AWS, Cloudflare, Stripe and Vercel.

x402, 2026-08 self-reported by the project, not independently audited single-source · on the wire · source ↗

Major websites that scored zero on agent-transaction protocols like OAuth discovery, the steepest shortfall of 12 machine-readability signals in a SALT.agency audit of 50 sites across retail, SaaS, travel, publishing and finance. Sites averaged 56.6% overall (Airbnb highest at 79.2%), only 11 of 50 had published an llms.txt file, and just 5 of 50 had adopted Cloudflare's Content Signals Policy.

Search Engine Journal, 2026-09 SALT.agency audit — vendor-published single-source · on the wire · source ↗

How much more often AI assistants clearly recommended businesses whose websites an AI agent could successfully fetch and read, in a simulation of 37,927 buyer-question journeys across four agent harnesses and 1,056 real businesses — agent-ready businesses had their AI answers built from their own pages 78% of the time versus 56% for the rest, and were 41% more accurate, while answers about non-agent-ready businesses were 3.7 times more likely to omit facts the buyer had asked for. Only 7-10% of the finished answer came from the model's training knowledge either way.

arXiv GEO, 2026-09 preprint study — vendor-adjacent; co-author co-founded Tavily, an AI-agent web-search company acquired by Nebius — no affiliation disclosed in the paper unchecked · on the wire · source ↗

AI-search platforms a new study got to cite a fabricated concept within seven days of posting it on preferred, easy-to-access domains — a $14 commercial GEO purchase separately got one platform citing planted content within an hour. The same analysis, of 17,211 citations across 6,356 source domains, found 15 of 22 publishing platforms tied to each engine's top-cited domains have low or medium barriers to posting.

arXiv AI-search citations, 2026-10 arXiv preprint — not yet peer reviewed unchecked · on the wire · source ↗

Share of more than 500 U.S. and U.K. marketing leaders who plan to deprioritize or stop investing in traditional SEO, while 44% plan to increase spending on answer engine optimization instead — 62% said their brand is trading content quality for AI citations, and 61% said the same of authenticity, per a Typeface survey.

PPC Land, 2026-10 Typeface study — vendor-published, methodology disclosure thinner than Typeface's June 2026 survey on the same topic single-source · on the wire · source ↗ · study ↗

Growth in purchases placed through AI shopping agents since April 2026 — rising about 43% a month, per BrightEdge data cited by The Wall Street Journal — while only about 3% of the ecommerce companies BrightEdge works with let an agent complete a transaction, as retailers report losing visibility into who initiated a purchase and why.

The Wall Street Journal, 2026-10 BrightEdge data — vendor-published, via WSJ reporting single-source · on the wire · source ↗

Task success rate on a simulated Amazon AI-tool market after seller AI agents — told only to win more buyer-agent selections, never instructed to lie — spontaneously began exaggerating their listed capabilities, down from 87-89% before that drift; a parallel Instagram tool market fell from 74-80% to 40%. Across 166,320 recorded buyer-agent choices, tools that answered fewer than 10% of questions correctly still won a quarter to two-fifths of selections once their listings read like capable tools', per researchers at Google, Google DeepMind and the University of Toronto.

arXiv GEO, 2026-10 arXiv preprint — not yet peer reviewed unchecked · on the wire · source ↗

Trials across seven AI models the authors ran to measure how much seller-controlled content — badges, endorsements, framing — shifts AI shopping agents' product picks, introducing "price-equivalent auditing" (raising a cued product's price until the model picks it only half the time) because two cues that win at an equal rate can still differ widely in how much price premium they're worth. A decision-native model tested, Jev, resisted most cues but stayed susceptible to the strongest social-proof claims — a different profile than any text-based model in the study.

arXiv GEO, 2026-10 arXiv preprint — not yet peer reviewed unchecked · on the wire · source ↗

Share of U.S. desktop searches across 41 search-heavy websites handled by Google versus all AI tools combined in Q4 2025, per SparkToro and Datos — traditional engines about 80%, commerce sites about 10%, social about 5.5%. Counts searches wherever they happen, not referrals among search engines, which is why Google sits below Statcounter's 90%-plus. Amazon, Bing and YouTube each out-searched ChatGPT.

SparkToro / Datos, 2025-12 desktop panel only; 41 domains chosen editorially; Datos is a Semrush company unchecked · on the wire · source ↗

ChatGPT's search-like volume as a share of Google's, per an Ahrefs estimate: 2.5 billion prompts a day against Google's 13.7 billion searches, with 65% of prompts classed as search — 8.33 Google searches for every search-like ChatGPT query. Falls to 4.4% if only pure fact-finding prompts count. Google sends about 190 times more traffic to websites.

Ahrefs, 2026-02 vendor-published; the 65% is the author's own classification single-source · on the wire · source ↗

Share of traffic to 74,752 websites sent by all AI chatbots combined in March 2026, against 28.12% from Google, per Ahrefs's chatgpt-vs-google.com tracker — referral visits, not searches.

Ahrefs, 2026-03 vendor-published, from Ahrefs's own panel single-source · on the wire · source ↗

ChatGPT referral traffic as a share of Google organic traffic, across 100 websites and 145 million GA4 sessions over 12 months, per Marie Haynes (0.73% for all LLMs combined) — a different denominator than Ahrefs's 0.28% reading of AI chatbots' share of 74,752 sites' total traffic above, but the same order of magnitude.

Marie Haynes, 2026-10 practitioner GA4 analysis across her own audit panel — not independently replicated unchecked · on the wire · source ↗

Share of Google searches surfacing an AI Overview, per a Similarweb analysis.

TechCrunch AI / Similarweb, 2026-07 Similarweb study — vendor-published single-source · on the wire · source ↗ · study ↗

15% as of 2025-07 — Same Similarweb series, one year earlier — nearly a threefold rise in twelve months. (wire item)

Share of Google's People Also Ask boxes answered by an AI Overview in the first week of September 2026, per AlsoAsked's analysis of 19.2 million English-language queries — up from 86% in August, the last major SERP feature AI Overviews hadn't fully absorbed.

Search Engine Roundtable / AlsoAsked, 2026-09 AlsoAsked study — vendor-published contested · on the wire · source ↗ · study ↗

86% as of 2026-08 — Same AlsoAsked series, one month earlier. (wire item)

~12% as of 2025-07 — Both trackers put the figure at roughly 12% about 14 months earlier. (wire item)

100% per Allintitle — Allintitle separately measured full saturation as of August 2026, a month before AlsoAsked's 97% reading — close but not identical trackers of the same convergence. (wire item)

Growth in AI Overview coverage of manufacturing search terms, January to July 2026, per a Semrush analysis of 458 manufacturing keywords, clickstream data and AI citation tracking across ChatGPT, Gemini, AI Mode and AI Overviews — a faster rise than the 15% to 43% climb measured across all Google searches over the prior year. AI sources drove under half a percent of manufacturing site traffic against direct and organic search's roughly 80% share.

Semrush Blog, 2026-09 Semrush study — vendor-published single-source · on the wire · source ↗

Growth in AI Overview coverage of commercial search queries, September 2025 to August 2026, across 400 tracked keywords spanning banking, software and ecommerce on US desktop, per Semrush — weighted by estimated search volume, August coverage reached 97.08%. By query type in August: "best X for Y" searches 96.12%, reviews 90.25%, brand-vs-brand comparisons 86.23%, price/cost queries 85.51%, alternatives queries 82.05%.

Semrush Blog, 2026-08 Semrush study — vendor-published single-source · on the wire · source ↗

Share of roughly 100 large brand names that triggered a Google AI Overview when searched directly — including Reddit, Salesforce, Amazon, Adobe, Wikipedia, Samsung, Apple and Yahoo — well above the 43% aggregate rate across all Google searches, per researcher Chris Long. Most placements sat mid-page or lower rather than at the top; Google showed none for searches on its own name or on news publishers' names. Long found AI Overviews far less common on the same kind of brand search when he ran a similar test roughly two years earlier.

Search Engine Roundtable, 2026-09 single-source — roughly 100 brands, one researcher's test, confirmation pending contested · on the wire · source ↗

63 of 100 per Ahrefs Blog — Ahrefs' weekly panel of 100 bare brand names (Interbrand's Best Global Brands) found an AI Overview on 63 at its September 30 checkpoint, against Long's 93% the same week. The two also disagree on placement: 48 of Ahrefs' 63 sat in first position, where Long found most mid-page or lower. Long's is one test; Ahrefs' is 14 weekly ones. Ahrefs' 82.91% is a different measure (every search containing a brand name) and does not contest this figure. (wire item)

Share of US desktop Google results for branded keywords carrying an AI Overview as of September 29, 2026, per Ahrefs' analysis of 232.8 million search crawls — a September monthly average of 73.3%, up from 61.3% in July. Branded keywords here means any search containing a brand name. In a separate Ahrefs panel of 100 bare brand names, 63 showed an AI Overview by September 30, 48 of them (76.2%) in the page's first position, ahead of the brand's own website; Wikipedia was cited in 41 of the 57 whose sources were visible (71.9%), YouTube in 22 (38.6%) and LinkedIn in 13 (22.8%).

Ahrefs Blog, 2026-09 Ahrefs study — vendor-published single-source · on the wire · source ↗

13.1% as of 2025-11 — An earlier Ahrefs study, ten months before; Ahrefs says the two samples differ, so read it as direction, not a like-for-like series. (wire item)

Share of branded Google searches showing an AI Overview, per DemandSphere's tracking — 26.12% on September 1, 2026, rising to 82.06% by September 29, a jump DemandSphere says it cannot explain and Google has not addressed. Ahrefs independently measured 82.91% on the same day, up from 65.1% in August, per Semrush Blog.

Semrush Blog, 2026-09 DemandSphere study — vendor-published corroborated by Ahrefs · on the wire · source ↗ · study ↗

Share of U.S. ChatGPT desktop queries that included a source citation as of May 2026 — even as citation volume in AI answers overall rose more than fivefold over the year.

TechCrunch AI / Similarweb, 2026-07 Similarweb study — vendor-published single-source · on the wire · source ↗ · study ↗

Share of U.S. news searches (17.5% in the U.K.) where Google now places the Top Stories carousel inside the AI Overview — meaning publishers who use the Search Console AI opt-out could lose Top Stories placement there too. Google hasn't confirmed the link.

Search Engine Journal, 2026-07 single-source — confirmation pending single-source · on the wire · source ↗ · study ↗

Share of ChatGPT's 494 million users who also used Google search during the same March-to-May 2026 span, per a Similarweb analysis — evidence AI search is layering on top of traditional search, not replacing it. The same analysis found 58.8% of ChatGPT referral traffic lands on publisher homepages even though 65% of the URLs it cites sit two to three folders deep, and that U.S. ad penetration on ChatGPT jumped from 14% in May to 26% in June.

Search Engine Journal, 2026-07 Similarweb study — vendor-published single-source · on the wire · source ↗ · study ↗

Share of web traffic that was automated (bot) requests in June 2026, per Cloudflare data cited in IAB Australia's new crawler decision matrix — with 52% of AI crawler requests serving model training versus just 2.6% for live AI-answer queries.

PPC Land, 2026-07 unchecked · on the wire · source ↗ · study ↗

Gen Z consideration of Claude, which nearly doubled from 14.2% to 28.1% between Q1 and Q2 2026 (OpenAI: 10.1% to 22.1%), per YouGov BrandIndex — even as just 28% of Americans overall say they trust AI assistants for answers, a 42-point gap behind trust in traditional search engines.

Search Engine Journal, 2026-08 unchecked · on the wire · source ↗ · study ↗

Daily requests donor-database site PatronView logged from Amazon's Amzn-SearchBot before blocking it in early August 2026, per founder Nick Gray's published server logs — after already blocking Anthropic's Claude-SearchBot in June over a measured 35,000-to-1 ratio of crawls to human-referred visitors; a 48-hour CAPTCHA challenge in the same period drew 106,437 attempts and only 252 solves, a 0.24% rate.

PPC Land, 2026-08 single-source — confirmation pending single-source · on the wire · source ↗ · study ↗

Trust in Google AI Overviews and ChatGPT among US shoppers, tied for the top score and ahead of TikTok, Instagram, Facebook and influencers, though still behind friends and family at 59% — per an RTB House survey of 1,840 US and international consumers. 42% of shoppers said AI tools lengthen their purchase research, but only 35% want to approve a purchase before an AI agent completes it, and millennials' comfort with a $250 autonomous AI spend fell from 42% to 34% once a return guarantee was removed.

PPC Land, 2026-08 RTB House study — vendor-published single-source · on the wire · source ↗

Monthly active users of the Gemini app, which Google calls the fastest-growing product in its history — with 63% of users engaging by voice, over 150 million images generated daily, and more than 100 million active iOS users, per Google's own figures.

Google Keyword Blog, 2026-08 Google-reported figure — not independently verified unchecked · on the wire · source ↗

Share of AI page fetchers TollBit tracked in the first half of 2026 that reached European publisher URLs marked disallowed in robots.txt — with OpenAI's ChatGPT-User among the bots most often crossing blocks aimed specifically at it, as OpenAI expands ChatGPT ads into the EEA and Switzerland.

PPC Land / TollBit, 2026-08 TollBit data as cited via PPC Land unchecked · on the wire · source ↗

Share of 592 sites with a declared GPTBot ban in robots.txt that still served GPTBot a live page when tested directly, per HasData's AI Crawler Block Index of 10,894 domains — a declaration-level echo of TollBit's per-fetch 15% crossing rate above. The same study found 118 sites, among them the Smithsonian and Vatican News, lost their Cloudflare-managed AI opt-out and GPTBot-ban lines entirely when Cloudflare retired its older "Managed Robots.txt" and "Block AI Bots" features on September 15, 2026; the replacement, Bot Preference Sync, leaves robots.txt blank until the site owner reviews and confirms new rules, a step Cloudflare's own announcements did not flag to existing customers.

PPC Land, 2026-10 HasData study — vendor-published single-source · on the wire · source ↗ · study ↗

Share of 6,735 working robots.txt files, among Chrome's top 10,000 sites, using the IETF's draft Content-Usage directive — the structured vocabulary for separately flagging AI training, AI use and search crawling. Just 13 sites used it, and 12 of those 13 were country editions of one company, Vinted. By comparison, 950 of the same sites (14.1%) carry a named block on OpenAI's GPTBot alone, per a ProxyPicker check run October 8, 2026.

ProxyPicker, 2026-10 ProxyPicker study — vendor-published single-source · on the wire · source ↗

Unique sources readers have picked as a Google Preferred Source as of August 2026, up from over 345,000 at the feature's May 2026 launch — Google now lets publishers embed a button on their own pages to prompt the pick, and says preferred sources surface more often in Top Stories, AI Overviews and AI Mode.

Google, 2026-08 Google-reported figure — not independently verified unchecked · on the wire · source ↗

Average monthly EU users the European Commission credits ChatGPT with, the basis for designating it a Very Large Online Search Engine under the Digital Services Act — the third such designation after Google Search (364M) and Bing (119M), both from April 2023, and the first non-traditional search product on the list.

European Commission, 2026-08 Commission-published figure — not independently verified unchecked · on the wire · source ↗

ChatGPT's share of AI assistant prompt volume, January to June 2026, while Gemini's share nearly doubled from 17% to 30% and Claude's climbed from 2% to 11%, per Comscore's Q2 2026 AI Intelligence Report. ChatGPT still led on raw volume — 168 million desktop conversations in June — but its own desktop conversations fell from an April peak of 40 million to 22.3 million by June, after surging from 1.1 million the previous October.

Digiday Media, 2026-06 Comscore data — vendor-published, opt-in panel of 500,000 to 1 million respondents single-source · on the wire · source ↗ · study ↗

Share of AI-platform referral traffic to websites that went to ChatGPT in August 2026, a 2026 high — Gemini took 2.4%, Perplexity 1.1% and Claude 0.8%, a combined 4.3% — per BrightEdge. ChatGPT's own referral traffic rose 101% between January and August 2026, including a 36% jump in August alone. This measures referral clicks, not prompt volume: Comscore's prompt-share reading above has ChatGPT's share falling as Gemini and Claude gain, a different axis entirely.

BrightEdge Blog, 2026-08 BrightEdge study — vendor-published; methodology (panel size, site count) undisclosed single-source · on the wire · source ↗

People who used YouTube's Ask AI feature — which answers viewer questions about the video on screen using a custom Gemini model — in June 2026, up more than 500% from December 2025, per a YouTube product VP's remarks at the company's Made on YouTube 2026 event. YouTube disclosed neither the December baseline nor how it defines a user.

PPC Land, 2026-06 single-source — confirmation pending; company-reported figure with no disclosed baseline or measurement method single-source · on the wire · source ↗

Average share of Google SERPs missing more than half their usual top-10 sites after Google began blocking bulk SERP collection outright on September 13, 2026, up from a 0.17% baseline and peaking at 27.0% on September 29 — across 27.5 million SERPs collected in 49 countries, per rank-tracking vendor Nozzle. Social platforms including Reddit (+8.0 points) and Facebook (+5.5) were over-represented on the altered pages, which mostly revert within three days; Nozzle says third-party rank trackers are caught in a countermeasure aimed at bulk AI-grounding scrapers.

Nozzle, 2026-10 Nozzle study — vendor-published (rank-tracking vendor); catch-up (published Oct 10) single-source · on the wire · source ↗

Publisher sites that allowed Anthropic's Claude-SearchBot in an OpenAEO measurement, versus 11 of 12 that allowed OpenAI's OAI-SearchBot — AP, Forbes, The Guardian, The New Yorker, The Verge and Vox were among those blocking Claude's crawler while permitting OpenAI's. OpenAEO attributes the gap to publishers having learned OpenAI runs separate search and training crawlers, and not having learned the same about Anthropic.

OpenAEO, 2026-08 OpenAEO study — vendor-published, catch-up (published Aug 11); narrowed from 131 domains to 12 after a correction removed a parked domain single-source · on the wire · source ↗

Meta's share of AI agent traffic across DataDome's network in Q3 2026 — its Meta-ExternalAgent and Meta-WebIndexer crawlers combined for 12.4 billion of nearly 25 billion requests — down from 51.5% in Q2 even as the raw total grew 41.5% quarter over quarter, because non-Meta agents grew traffic faster (47% quarter over quarter) than Meta's own crawlers did. Meta's new agentic browser, Muse, separately logged 29.1 million requests in its first 12 days of availability (Sept. 22-Oct. 3).

Help Net Security, 2026-10 DataDome study — vendor-published; the wire published the same DataDome report twice within minutes on 2026-10-19 under a second headline; that copy was merged into this item on 2026-10-23 single-source · on the wire · source ↗ · study ↗

Sites on AEO Wire's 52-site crawler panel disallowing Anthropic's Claude-SearchBot versus OpenAI's OAI-SearchBot in robots.txt, read October 19, 2026 — PerplexityBot was disallowed on 30. The gap tracks OpenAI licensing deals: of 17 panel sites with a public OpenAI deal, 16 let OAI-SearchBot in and only 3 let Claude-SearchBot in; of 35 sites without one, the three bots were disallowed at similar rates (15, 14 and 15). Googlebot and Bingbot were each disallowed only by Reddit, which disallows every crawler in robots.txt while licensing content to Google and OpenAI.

AEO Wire crawler panel, 2026-10 wire's own measurement — robots.txt rules on 52 of a fixed 55-site panel, read October 19; policy as stated, not what each server enforces unchecked · on the wire · source ↗

Monthly reach of Google AI Overviews, per DataReportal's Digital 2027 report — applying Similarweb's finding that over 4 in 10 Google searches now surface an AI Overview to Google's 2.6 billion-plus monthly search audience.

DataReportal, 2026-10 DataReportal estimate (compiled with Meltwater and We Are Social) applying Google's reported search audience to Similarweb's AI Overview share — not a direct measurement unchecked · on the wire · source ↗

How much more likely online adults are to visit a search engine than a stand-alone AI platform when researching a brand, per DataReportal's Digital 2027 report — evidence AI platforms still trail search for commercial research even as AI Overviews reach a majority of Google's audience.

DataReportal, 2026-10 DataReportal report (compiled with Meltwater and We Are Social) — vendor-published single-source · on the wire · source ↗

Share of facts frontier LLMs — including Gemini-3-Pro and GPT-5 — encode internally but fail to directly recall, per Google research that measured 95-98% encoding rates; reversing a fact's usual subject/object order makes recall harder, and recall failures account for more than 70% of GPT-5.2's errors. Extended "thinking" recovered 40-65% of the otherwise-unrecalled facts.

Search Engine Journal, 2026-08 Google research paper as covered by Search Engine Journal — researchers stop short of confirming a content-phrasing implication unchecked · on the wire · source ↗ · study ↗

How much more likely Claude is than ChatGPT to produce an inaccurate brand claim, across more than 158,000 AI-generated brand claims Profound analyzed with its FactCheck tool — overall inaccuracy rates ran 4.3-6.8% across engines. Pricing and billing was the single largest error driver: 12% of claims evaluated but 24% of inaccurate ones, and the top error theme for 66% of brands with at least ten pricing claims checked.

Profound research blog, 2026-08 Profound study — vendor's own product data, not independent research unchecked · on the wire · source ↗

Win rate for AI Overviews' generated interfaces over the top organic search result in head-to-head comparisons, per Google's own research paper — though custom-built websites still beat generative UI 50% to 35.3%, and on information-seeking queries generative UI beat the top result 73.5% to 19%, per Search Engine Journal's review.

Search Engine Journal, 2026-08 Google's own research paper — self-reported test results single-source · on the wire · source ↗ · study ↗

Share of responses in which Claude searched the web, versus just 13% for Claude Code — the two products' agents also visited different kinds of pages (Claude Code mostly documentation and pricing pages, Claude mostly robots.txt files, sitemaps and homepages) and shared only about one in five cited brands, per a Profound analysis of 24,135 responses collected July 13-23, 2026.

Profound research blog, 2026-08 Profound study — vendor's own product data, not independent research unchecked · on the wire · source ↗

Accuracy of AI answers to branded questions when a platform chose to respond (95.9%) versus how often it declined to answer at all (31.2%), across 1,562 branded prompts and 28,123 responses on six platforms, per a Seer Interactive analysis — Claude was most accurate of any platform when it answered (98%) but had the lowest total accuracy (52.6%) because it declined most often, while AI Mode posted the highest total accuracy (74.2%). On comparative prompts, platforms declined 80.3% of the time and were right only 18.8% of the time they answered.

Seer Interactive, 2026-08 Seer Interactive study — single-brand self-study, not a cross-brand sample unchecked · on the wire · source ↗

Average accuracy of six leading AI models — including GPT-4o, Claude and Gemini variants — answering questions about global development statistics, across a UNICEF benchmark of 133,000 questions; roughly 60% of answers lacked a usable number at all, per TechCrunch. The finding prompted the UN to launch its own agent-queryable Data Commons platform with Google on September 17, 2026.

TechCrunch AI, 2026-09 unchecked · on the wire · source ↗

Share of product-recommending responses in which ChatGPT, Gemini and Google AI Overviews state a first-person product preference — ChatGPT 79%, Gemini 7%, AI Overviews 2% — across 1,536 responses to 2,528 real buying queries in the ConsumerQ dataset, posted to arXiv. The same audit found ChatGPT and Gemini's web interfaces shared only 5.4% of cited domains on average, with none in common at all in 76.7% of query comparisons — a gap the researchers say narrows when comparing the same models' raw APIs instead of their consumer interfaces.

arXiv AI Overviews & brand visibility, 2026-09 arXiv preprint — not yet peer reviewed unchecked · on the wire · source ↗

Share of 66.6 million free, logged-out ChatGPT-5.6 Luna responses that triggered a web search at all in August and September 2026, averaging 1.6 fan-out queries — the sub-queries ChatGPT runs while answering — each, per Ahrefs. Of the resulting 15 million fan-out queries, 6.8% used the site: operator to search one domain directly, spread across roughly 116,000 domains; apple.com, ikea.com and nike.com topped the list by mentions.

Ahrefs Blog, 2026-09 Ahrefs study — vendor-published single-source · on the wire · source ↗

Share of answers in which Claude Sonnet 5.5 asked the reader for context about their specific situation, versus 12.5% for Sonnet 5, across 711 prompts run 20 times each on Anthropic's API with web search enabled (28,440 answers), per Profound. Sonnet 5.5 also named the current year in 59% of its query fan-outs — more than double Sonnet 5's rate — and its most-cited pages skewed newer, with a median age of 107 days against 164 days for Sonnet 5.

Profound research blog, 2026-10 Profound study — vendor's own product data, not independent research unchecked · on the wire · source ↗

Share of ChatGPT Web's exact-URL citations recovered when replaying the same queries through OpenAI's Responses API and Google Search as external proxies, combined — the two together still miss about three-quarters of what ChatGPT Web actually cites. A model combining both signals matched citations more closely at the hostname and entity level (AUC above 0.74 for GPT-5.3 and GPT-5.5), and pages retrieved more often across repeated runs of GPT-5.2 through GPT-5.5 were also more likely to be cited.

arXiv, 2026-10 arXiv preprint — not yet peer reviewed unchecked · on the wire · source ↗

Share of within-site variation in which pages ChatGPT cites explained by whether ChatGPT's own rewritten search queries retrieved the page (32%, 95% CI 24-42%) versus by page content and structure (2.8%) — a gap the study calls statistically decisive, across 90 days of ChatGPT, Gemini and Perplexity answers to commercial questions (158,099 answers, 468,037 citations), per University of Toronto researchers. In controlled experiments holding every version of a product's facts identical, revealing the brand name changed ChatGPT's top-three ranking rate 2.1 times as much as simply repeating the prompt, and 3.3 times as much for Gemini.

arXiv, 2026-10 arXiv preprint — not yet peer reviewed unchecked · on the wire · source ↗

Share of the pages ChatGPT's own search retrieved that were absent from Google's top 50 (97%) and from Bing's top 30 (99%) for the same queries. ChatGPT splits a question into about 4 keyword-style queries, adds the current year to 23% of them and a brand the user did not name to 28%; it cites 43.8% of the pages it ranks first, 12.2% at ranks 6-10 and 3.5% at ranks 11-20; and two runs of the same question on the same day overlap by 2.7% in queries and 15% in cited pages (Sections 6.2-6.3), per University of Toronto researchers.

arXiv, 2026-10 arXiv preprint — not yet peer reviewed; the replay used ChatGPT's API, not the app unchecked · on the wire · source ↗

Answers a controlled audit collected to compare how ChatGPT and Gemini localize results: 300 questions asked of ChatGPT across 19 declared (app) locations, and both engines asked from five network locations. ChatGPT follows the location an app declares down to the city — localizing results, and the cited sources and business recommendations in them, even for questions that don't mention a place — while ignoring network location entirely; Gemini does the opposite, following the network's country and ignoring declared location.

arXiv GEO, 2026-10 arXiv preprint — not yet peer reviewed unchecked · on the wire · source ↗

Mean reduction in answer overlap from Gap Detection, a proposed framework that searches a shared embedding space for relevant ideas a language model's existing answers haven't already covered, across five curated topics and 90 queries from the INFINITY-CHAT benchmark — aimed at LLMs' tendency to give near-identical answers to open-ended questions. The authors also instantiated the method for generative engine optimization across four unnamed product markets, where it cut answer overlap from 0.324 to 0.298 against undirected methods.

arXiv GEO, 2026-10 arXiv preprint — not yet peer reviewed; GEO application tested on four unnamed markets, not real citation data unchecked · on the wire · source ↗

Safety-and-relevance scores out of 1.00 for Google's AI Overviews (0.76) and Featured Snippets (0.71) across 1,508 first-aid queries, weakest on yes/no and how-to questions, per a peer-reviewed audit using RADAR-FirstAid, an automated black-box auditing pipeline built by researchers at Wuzhou Medical College; a manual check on a 200-query subset backed up the automated scores.

Discover Artificial Intelligence (Springer Nature), 2026-10 single-source — confirmation pending single-source · on the wire · source ↗