Why Isn't Your Brand Being Mentioned by ChatGPT?

What changed in ChatGPT brand mentions?

ChatGPT appears to be citing sources more often for some comparison prompts, but citations have not made brand mentions a simple retrieval problem. Graphite reported that the share of logged-out ChatGPT comparison-prompt answers containing a citation rose from 56.6% in the week of August 3, 2026 to 88.5% in the week of September 28, 2026. That is a meaningful before-and-after change in the observed answers, not proof that every ChatGPT response now uses search.

The important shift is operational. A brand can no longer treat a cited page as a direct path to a mention in the final answer. Search can supply sources, yet the model still decides which companies to name, compare, or omit. A citation is evidence that a source appeared in an answer. It is not evidence that every company on that source received a mention.

Graphite compared logged-out ChatGPT answers to the same entity-comparison prompts with and without search. Its October 16, 2026 analysis found that visibility without search tracked with visibility when search was used. The result held across groups with different rates of substantive brand mentions on cited pages.

This is why a dashboard that counts cited URLs can create false confidence. It can show that your domain, partner site, review listing, or category guide entered the answer process while missing the commercial question: did ChatGPT actually name your brand when a buyer asked?

1. Before: Track cited pages and count brand mentions on those pages.
2. After: Track cited-page coverage, ChatGPT brand mentions, and answer presence for the same prompt set.
3. What to do: Keep a no-search baseline where the testing method supports one, then repeat search-on tracking over time.

Why can cited-page coverage fail to produce a brand mention?

Cited-page coverage can fail because a source list and a final answer are different stages of the response. Graphite describes model knowledge without search as parametric knowledge, meaning information encoded in the model's weights during training. When a model uses search, retrieved content enters the response context, but that does not erase the role of the model's prior knowledge.

In Graphite's data, entities that appeared on at least 15% of cited pages still had median search visibility of 11% when they fell into the lowest no-search visibility band, from 0% to 25%. The finding does not show that a company can control model training data or force a mention. It shows that cited-page mentions did not fully account for the observed visibility outcome.

The same study measured approximately 957,000 ChatGPT fan-out queries from 22,953 non-branded entity-comparison prompts collected from August to early October 2026. Graphite reported that 59.6% contained a brand and 22.7% used a site: operator. Those queries can shape which pages become candidates for retrieval before the answer is written.

Treat this research as directional evidence, not a universal formula. Graphite is an AEO agency, its analysis concerns a particular set of logged-out comparison prompts, and its findings do not establish that adding a mention to a cited page will increase brand visibility. That limitation matters because the work of AI search visibility starts with measurement, not a claim that any single content tactic changes a model's output.

Who is most affected by lower ChatGPT brand mentions?

B2B SaaS and technology growth teams are most exposed when buyers use comparison prompts such as 'best [category] tool' or ask ChatGPT to contrast named alternatives. A brand can have conventional search visibility, third-party coverage, and a growing citation footprint while remaining absent from the short list a buyer sees.

Local and multi-location businesses face a related problem when people ask for a service in a specific city. The answer may name only a few providers. There is no page two in an AI answer, so an unmentioned business can lose consideration even if relevant pages exist elsewhere on the web.

The issue also affects established brands that assume extensive coverage guarantees inclusion. Graphite's strongest example is precisely this condition: even entities substantially mentioned on at least 15% of cited pages had low median search visibility when no-search visibility was low. Coverage remains useful, but it is not the same measurement as answer presence.

Smaller brands should not read the finding as a reason to chase volume without a prompt map. The practical unit is a relevant answer. Start with the questions a buyer, researcher, or local customer would actually ask, then record whether the brand appears, how it is described, and which competitors are named.

How should you measure ChatGPT brand mentions now?

Measure ChatGPT brand mentions with a fixed, relevant set of prompts and separate the result from source-level metrics. Capture whether ChatGPT names the brand, whether it cites sources, which competitors it names, and whether the answer responds to the user's category or comparison question.

Use Citation Share for the percentage of relevant AI answers in a category that cite your brand or content. Use Answer Presence for breadth across the question universe. Use Share of Voice to compare mentions against named competitors. These measures answer different questions, so do not collapse them into one count.

Graphite's research also supports a before-and-after testing design. Record a baseline before publishing or refreshing a content set. Run the identical prompts again after the material has had time to be discovered. Keep model, geography, account state, search setting, date, prompt wording, and response evidence attached to every observation.

A single clean answer can be misleading. ChatGPT outputs can change over time, and Graphite notes that its citation-rate observations changed over the studied period. Use repeated runs. Review the response text, not merely the link list. A page can be cited without the brand being named, and a brand can be named without a useful explanation of why it belongs in the answer.

1. Define the category, buyer question, geography, and named competitors for each prompt.
2. Record brand mention, answer position, citations, cited URLs, and competitor mentions.
3. Calculate Citation Share and Answer Presence separately for each prompt group.
4. Repeat the same prompt set on a documented schedule and preserve the response evidence.

What should you publish when ChatGPT is not mentioning your brand?

Publish authoritative coverage of the questions where your brand should be considered, then verify whether it changes the answer set. The goal is not to game a model. It is to give search systems and readers clear, current material that explains what your product or service does, where it fits, and how it differs from alternatives.

Start with gaps revealed by the prompt audit. If ChatGPT names competitors for a category question but cannot find a clear explanation of your category fit, build that explanation. If it cites broad listicles but omits the evaluation criteria that make your has relevant, publish a focused comparison or how-to page with explicit scope and dated sources.

Keep factual claims narrow. State product capabilities, audience, use cases, limits, and update dates only where your evidence supports them. Avoid invented benchmarks, unsupported market statistics, and promises of rankings or citation counts. Citation engineering is quality, coverage, and freshness at a scale AI systems can assess, not a shortcut around editorial standards.

Refresh pages that have become vague or stale. Graphite estimates a minimum two-to-five-month period between a model knowledge cutoff and release for newly published material, but it also states that this does not measure whether a page entered training data or changed an answer. Use that as a planning constraint, not a guarantee.

Should you optimize for citations or for model knowledge?

Optimize for both observed answer citations and durable, accurate brand coverage, while keeping the two measurements separate. Graphite argues that citations and what it calls parametric knowledge both shape its observed ChatGPT visibility results. The evidence supports a broader measurement program, not an instruction to abandon cited-page work.

Citation work still matters because sources make answers verifiable and can surface your material when a model searches. Strong comparison pages, definitions, original research, source-backed explainers, and clearly maintained documentation give a system something specific to cite. They also give a buyer a reason to trust the answer.

Broader brand coverage matters because models can answer without search and may enter search with candidate brands already reflected in fan-out queries. You cannot inspect a closed model's training set or prove that a particular page changed its weights. You can has a reliable public record of what your brand is, who it serves, and what evidence supports its claims.

Google's documentation provides one adjacent example of why crawler controls deserve review. Google says its common crawlers obey robots.txt rules, and its crawler documentation explains the relationship between user-agent controls and crawling preferences. That fact concerns Google, not ChatGPT. Do not assume a Google crawler rule determines ChatGPT behavior, but do audit access controls before treating visibility problems as purely editorial.

What is the next move for a brand missing from ChatGPT answers?

The next move is to build a documented baseline, identify the missing decision context, and publish the best-supported page that fills it. Do not begin by rewriting every page or buying generic mentions. Start with the few prompts where absence creates a real commercial gap.

For each prompt, inspect the answer's named brands, cited pages, omitted criteria, and language. A missing mention may reflect weak category definition, thin independent coverage, an unclear comparison page, a geographic mismatch, or a prompt that does not actually fit your offer. The evidence should determine the content brief.

Then produce a page that a buyer could use without the model. Put the direct answer first. Explain scope and constraints. Add a dated source or original data point where one is available. Link the page to the related category, comparison, and definition pages so the site does not leave the topic isolated.

Finally, rerun the same prompt set and report what changed without overclaiming causation. A higher Citation Share or Answer Presence is an observation. It becomes useful when it is connected to a stable prompt set, preserved evidence, and a clear view of which questions still leave your brand out.

Sources

Source: Graphite

Graphite's analysis reported that ChatGPT citation rates in its fixed logged-out entity-comparison prompt set rose from 56.6% in the week of August 3, 2026 to 88.5% in the week of September 28, 2026. It also reported 11% median search visibility for entities in the lowest no-search visibility band despite appearing on at least 15% of cited pages. Graphite, 2026-10-16

Source: Search Engine Journal

Search Engine Journal reported the Graphite analysis, its comparison-prompt scope, and the distinction between cited-page mentions and final brand visibility. Search Engine Journal, 2026-10-20

Source: Google for Developers

Google documents that its common crawlers obey robots.txt rules and describes crawler user-agent controls. Google for Developers, 2026-10-27