Graphite: ChatGPT's no-search brand recall predicts its search answers
What this answers
- Does ChatGPT name a brand more often with search if pages it cites already mention it?
- Only partly, per Graphite: among brands substantially mentioned on at least 15% of cited pages, those ChatGPT named in under a quarter of its no-search answers reached a median 11% visibility when it used search. Brands it named more often without search were also named more often with search, at every level of cited-page mention rate.
- How often do ChatGPT's background searches include brand names?
- In Graphite's sample of about 957,000 ChatGPT fan-out queries, issued for 22,953 non-branded comparison prompts, 59.6% contained a brand and 22.7% used a site: operator; none of the prompts themselves named a brand.
Graphite, a growth agency that sells AEO services, reported on October 16 that brands logged-out ChatGPT rarely named without search were also rarely named when it searched, even when many cited pages mentioned them. The comparison covers 3,115 entity-question pairs across 57 comparison questions collected in August. Among brands substantially mentioned on at least 15% of cited pages, those ChatGPT named in 0–25% of no-search answers had a median 11% visibility with search; Graphite says two replications (100 API prompts with search switched on and off, and 200 tracked prompts over three weeks) showed comparable effects.
Graphite also counted brands in about 957,000 ChatGPT fan-out queries (the background searches run before answering) from 22,953 non-branded comparison prompts: 59.6% contained a brand and 22.7% used a “site:” operator. In its fixed set of comparison prompts, the share of logged-out ChatGPT responses with citations rose from 56.6% (week of August 3) to 88.5% (week of September 28), against 79.0% to 81.2% for logged-out Gemini. Per Search Engine Journal, the write-up does not say how Graphite separated no-search responses in the main sample, and does not control for how well known each brand is.
Why it matters: If the result holds, getting mentioned on pages ChatGPT cites is not sufficient on its own, because a brand the model rarely recalls from training stays scarce in search-backed answers too — though the vendor selling the remedy (training-data work it calls PKO) is also the one reporting the finding.
Glossary: AEOQuery fan-out
Via Graphite ↗ · Search Engine Journal ↗
Posted to the wire October 26, 2026. Edited by Joe Balewski.