A Graphite study of logged-out ChatGPT answers found that appearing on the pages ChatGPT referenced did not reliably earn a brand a mention in the answer itself — citation coverage alone did not predict whether the model named you.
2026-10-26 · by Moe Ameen
Graphite, a growth agency that sells answer-engine-optimization (AEO) work, published an analysis on October 16, 2026 arguing that being cited by ChatGPT and being named by ChatGPT are two different things — and that the first does not buy you the second. The write-up, covered by Search Engine Journal's Matt G. Southern, looked at logged-out ChatGPT answers to comparison prompts, run both with web search and without it, and defined a brand's "visibility" as the share of responses that mention the brand or product by name.
The headline finding: brands ChatGPT rarely named when it was NOT searching also rarely showed up when it DID search — and that held even for brands that appeared on many of the pages ChatGPT cited. Graphite grouped brands into three tiers by how often they appeared on cited pages; the pattern held in all three. In one cut of the data, among brands that appeared on at least 15% of cited pages, median visibility with search was only about 11% for brands that had been named in under a quarter of the no-search answers. In other words, showing up on the source pages did not translate into being named in the answer.
The main dataset was 3,115 entity-question pairs across 57 questions, collected in August, with two replications (100 prompts through the API with search explicitly on and off, and 200 prompts tracked over a separate three-week window). A fan-out analysis of roughly 957,000 background queries drawn from 22,953 non-branded comparison prompts found about 59.6% contained a brand name (labeled by Claude Haiku 4.5). Graphite also noted that citations themselves have surged: logged-out ChatGPT responses carrying citations rose from 56.6% in early August to 88.5% by late September.
Treat the specifics as directional. Graphite sells the service its data recommends, the coverage notes the write-up does not fully explain how non-search responses were identified, there were no controls for pre-existing brand awareness, and a citation does not by itself prove a search happened. The broader 2026 research points the same way — ChatGPT cites far more pages than it names brands, and named mentions cluster in third-party listicles, comparisons, and review roundups rather than a brand's own pages — but the exact size of the gap varies by category, prompt type, and whoever ran the study.
The study’s uncomfortable lesson is that an AI engine names the brands it already recognizes. You do not earn that by shipping one well-optimized page and hoping it lands on a cited list — you earn it by attaching your brand name to a steady, recognizable stream of content across the surfaces these models read, so the name becomes familiar rather than merely referenced. That volume problem is exactly what [Kompozy](/) exists to solve.
Here is the concrete move. Set your [Persona Brief](/glossary/persona-brief) so every output carries the same brand name, voice, and positioning, then generate the full spread from one idea — text posts, a blog article, an email newsletter, carousels, quote graphics, and persona/avatar video — all consistently branded. [Autopilot](/glossary/autopilot) then schedules and publishes that set across the eight social platforms plus blog and email behind a per-post review, week after week. The point is cadence and consistency: LinkedIn posts, YouTube video, and long-form blog content are among the surfaces answer engines lean on, and a brand that shows up on them repeatedly with a coherent identity is the kind of name a model carries into an answer. Kompozy cannot promise ChatGPT names you — nobody honestly can — but it turns "publish everywhere, consistently, on brand" from a team’s full-time job into a workflow, which is the groundwork the research says precedes the mention. For the strategy behind this, see our guides on [AI brand citations](/guides/ai-brand-citations) and [why AI recommends your competitor](/guides/why-ai-recommends-your-competitor), and the playbook on [getting your content cited by ChatGPT](/how-to/get-your-content-cited-by-chatgpt).
A citation means a page your brand appears on was referenced as a source for the answer; a mention means ChatGPT actually names your brand in the reply the user reads. Graphite’s October 2026 analysis found the two do not reliably go together — a brand can sit on many cited pages and still rarely be named, so citation coverage alone did not predict mentions.
The study suggests ChatGPT tends to name brands it already recognizes, and that recognition appears to follow a familiar brand across both searched and unsearched answers rather than coming from any single referenced URL. Broader 2026 research points the same way: named mentions cluster in third-party listicles, comparisons, and reviews, not a brand’s own pages.
Treat them as directional. Graphite sells answer-engine-optimization services, the coverage notes the write-up does not fully explain how it separated search from no-search responses, there were no controls for existing brand awareness, and a citation does not by itself prove a search occurred. The pattern is consistent across studies; the exact gap varies by category and prompt.
Build a recognizable, consistent presence across the surfaces answer engines read rather than optimizing one owned page. Publish on-brand content on a cadence — social, blog, newsletter, video — and work to be referenced in third-party comparisons and reviews. A tool like Kompozy generates that full spread from one idea in a single brand voice and schedules it across eight social platforms plus blog and email, which is the consistency the research says precedes a mention.