Last Week in AI Search: November 9

The Story

Are ChatGPT ads worth it yet? A new study values a click at 26 cents

Last Monday night, at a dinner in New York, OpenAI gave every marketing deck a new slide.

Adam Fry, who leads consumer product for ChatGPT, said that for every five search-style questions people ask Google, they now ask ChatGPT about one.

And starting today, November 9, ChatGPT ads get easier to buy. You tell ChatGPT what you’re willing to pay, on average, for each sale or sign-up, and it decides how much to pay for each ad.

That’s the pitch: a lot of people asking, and an easier way to reach them. So are ChatGPT ads worth it yet?

Questions aren’t visits

Fry counts a search each time ChatGPT looks online for answers, so one conversation can count several times. A question also isn’t a visit to your website. Ahrefs, an SEO software company, tracks visits to 118,671 sites that use its free analytics tool. In October, for every 100 visits Google sent those sites, Ahrefs recorded about 1.5 from ChatGPT. Since ChatGPT’s links stopped saying where a visit came from, some of its visits get counted as something else, so the real number is probably higher.

What an ad visit is worth

Also last Monday, Maximilian Kaiser and Christian Schulze, two business-school researchers, published an early study of online stores that track every sale. Of those stores, 87 bought ads on both ChatGPT and Google. Their data runs through August 22, before today’s new buying option.

A visit from a ChatGPT ad brought the store 89 cents in sales on average. A visit from a Google ad brought $2.45. Fewer than 4 in 1,000 ChatGPT ad visits ended in a purchase, against about 12 in 1,000 for Google.

The sales were lopsided. Of the 87 stores, 73 recorded no sales from ChatGPT ads at all. One store made 66% of all the purchases ChatGPT ads brought in.

By the authors’ math, for a ChatGPT ad click to pay off as well as a typical Google ad, a store could pay about 26 cents per click. That’s the price that matches Google’s return, not a break-even price.

The ads may cost you even if you never buy one. In the 13 weeks after ChatGPT started showing ads, the authors estimate the free visits ChatGPT sent to 1,898 sites fell about 6% to 8%, compared with the EU, where ChatGPT had no ads yet. Those visits ended in a purchase about as often as before.

What advertisers say

Of 30 big brands that shared their results with Adthena, an ad-tracking company, 52% said they spent more than they got back. But 26% said ChatGPT did as well as Google or better. Barn2, a small software company, spent $477 on ChatGPT ads over six weeks, got 370 clicks and recorded no sales. Its free ChatGPT visits converted at its usual rate.

OpenAI says that less than 200 days after launch, its ads are bringing in money at a pace of $1 billion a year. So far, the money is easier to find on OpenAI’s side.

So, worth it yet?

Not as a dependable place to move your budget. Most stores in the study earned far less per visit than from Google ads, and most recorded no sales at all. These numbers measure sales, not profit, and none of them can forecast your business. They make a case for a small, closely measured test. If you run one:

• Make sure you can count the sales. Your analytics probably miss some ChatGPT visits, and ten of Adthena’s 30 brands said tracking sales was one of their biggest problems. OpenAI warns the new buying option can leave budget unspent if it can’t see your sales. If ChatGPT shows zero, check your tracking before you believe it.
• Decide what a customer can cost before you start. Set a test budget and the result you’d need to keep going. Cheap clicks aren’t enough.
• Keep paid and free visits apart. One combined number would have hidden Barn2’s story.

Fry’s figure is OpenAI’s own and unpublished, and its ad revenue pace isn’t audited. Adthena’s results are self-reported by the brands, and Adthena sells ad-tracking software. The Kaiser & Schulze study isn’t peer reviewed. Its paid-ad comparison looks at stores that chose to buy ChatGPT ads, so it can’t prove the ads caused those results. The free-visit figure compares ad markets with the EU and is still an estimate. The data came free from Grips Intelligence, and a Grips adviser is one of the authors.

The Follow-Up

Three AI engines ranked the same pages and often picked different winners

Last week’s Story found that about half the websites an AI tool cites one day are gone the next. A new test shows the AI tools don’t agree with each other either.

Researchers put up product pages for 80 made-up brands, each rewritten in different ways to win AI recommendations. They gave ChatGPT, Gemini and Grok the links and asked each to rank the products. Each pair of engines agreed on the top product for only 48.8% to 66.3% of the brands, roughly half to two-thirds of the time.

Even the order of the links mattered. ChatGPT put the best-performing rewrite first 73.8% of the time with the links in one order, and 93.8% with the order reversed. The engines only ranked pages they were handed. The test didn’t check whether they’d find those pages on their own. So a win in one doesn’t guarantee a win in another. (Early study, not peer reviewed; made-up brands, product pages only.)

The Numbers

• 26% drop in People Inc.’s main traffic count, which it calls Core Sessions, in the third quarter, to 1.61 billion. The owner of PEOPLE and Allrecipes blamed AI search tools, and its CEO said it has lost nearly 70% of its Google search traffic over the last few years. (Company results and earnings call) Read →
• 64% of the 282 brand and product names BrightEdge tracks showed Google’s AI summary at the top of their searches in late October, up from 19% in mid-August. (BrightEdge study) Read →
• 0.75 correlation (1.0 is a perfect match) between how often a brand comes up in Reddit threads that show in Google results and how often AI tools name that brand. It was the strongest of seven kinds of mention Ahrefs checked across 10,000 U.S. brands. One change to watch: Reddit said last week it will start testing a logged-in-only site outside the U.S. this quarter, and hasn’t decided how much of a thread visitors from search will see. (Ahrefs study) Read →

The Rest

• An adviser to the EU’s top court said an AI chatbot’s answer can violate the rights EU law gives news publishers. That’s when the chatbot got around a paywall or a block on bots, or when its answer copies an article’s wording so closely that readers don’t need the original. Facts stay free to use. It’s the court’s first case about generative AI. The adviser’s opinion isn’t binding, and the judges will rule later.
• Anthropic signed its first news licensing deal, with the Associated Press for election data, Adweek reported, citing two people familiar with it. Neither company disclosed terms.
• The New York Times sued Brave, the search engine company. The Times says Brave copied its articles with a bot publishers can’t identify, then sold its search index to AI companies, Anthropic among them.
• Corrections — We corrected nine items, most to tie a figure to what it actually measured. People Inc.’s 21% was a share of its Core Sessions, not all traffic. Ahrefs’ 14.4% was the share of ChatGPT answers that ran extra background searches, not the share that searched the web. We also merged six stories the wire had run twice. All corrections →

---

All 18 items from November 2–8, dated and sourced: the week on the wire · State of Search · Corrections · Archive. Seen something the record should hold? Reply to this email; tips are credited by name.

The full wire

19 items published November 2–8. Every item dated, sourced and linked. An item can carry more than one lane, so lane totals exceed the count.

Google Search · 6

• Nov 2 — Google AI Overviews for brand searches reported moving to right column — Search Engine Roundtable
• Nov 3 — Google turns off Search and Gemini for K-12 student accounts by default — Google (The Keyword) (company announcement)
• Nov 4 — Google AI Mode adds home services booking and quote links — Search Engine Journal
• Nov 4 — Google UCP direct checkout comes to AI Mode in Canada — Google Canada Blog (company announcement)
• Nov 5 — Adthena: 175,100 Ads Counted in Google AI Overviews, AI Mode — Adthena (vendor study)
• Nov 6 — BrightEdge: AI Overviews on 64% of brand searches — BrightEdge Blog (vendor study)

Research · 5

• Nov 2 — ChatGPT Ads study: $0.89 per paid visit vs $2.45 on Google — Kaiser & Schulze (ecommerceresearch.org) (study)
• Nov 3 — GEO study tests five optimization strategies on live pages in ChatGPT, Gemini and Grok — arXiv GEO (study)
• Nov 3 — Ahrefs: Reddit mentions track AI visibility most closely (0.75) — Ahrefs Blog (vendor study)
• Nov 3 — Semrush: Listicle affiliate content is cited in AI answers most, coupon pages least — Semrush Blog (vendor study)
• Nov 8 — StackScan: 6.88% of robots.txt files block GPTBot — StackScan (vendor study)

Law & Money · 3

• Nov 2 — New York Times sues Brave over search index sold to AI companies — TechCrunch AI
• Nov 6 — People Inc. core sessions fall 26% as AI Overviews spread — People Incorporated
• Nov 8 — EU court adviser: AI chatbot answers can infringe publishers — Court of Justice of the European Union (press release 155/26) (court filing)

Engines · 3

• Nov 5 — ChatGPT Ads adds Target CPA and Target CPC bidding — Search Engine Roundtable
• Nov 8 — Anthropic signs first news licensing deal, with AP — Adweek (single source)
• Nov 8 — OpenAI: ChatGPT handles a fifth of Google’s search volume — Big Technology

Tools & Vendors · 2

• Nov 4 — Profound adds Meta’s Muse Spark to Answer Engine Insights tracking — Profound research blog (vendor study)
• Nov 8 — BBC robots.txt now blocks Anthropic’s Claude-SearchBot — Crawler access panel (first-party data)

Read the complete archive →