01Diagnosis
Why your brand is not mentioned in ChatGPT
Quick answer
A brand missing from AI answers is in one of three states. Absentnever mentioned, which is a coverage and authority problem. Mentioned but uncitednamed without a link, which is a third-party source problem. Mentioned and described wrongly, which is an entity problem, and usually the cheapest of the three to fix.
Run each prompt three to five times before concluding anything. The same prompt returns different sources on consecutive runs, so a single check is measuring variance rather than visibility.
Key takeaways
- Check the robots file first. OAI-SearchBot controls ChatGPT Search discovery. GPTBot is a training crawler. Many sites block both because someone copied a rule from a blog post, and then wonder why they are invisible.
- Ranking first on Google does not carry over. ChatGPT's exclusive citation rate was 73.5% in 2026 research, meaning most of what it cited was cited by nothing else in the sample.
- Misdescription is more common than absence, and it costs more. A wrong price quoted confidently loses deals that no mention at all would not have reached.
- The fix is usually off your own site. Review platforms, comparison articles and community threads are what the model is reading about you.
- No agency can guarantee this. Google states plainly that no third-party tool has access to its internal ranking or AI systems, and assistant answers are probabilistic.
- We sell a service that fixes this. The procedure below is the same one we run, published in full, so you can do it without us if you would rather.
Step one: check properly, not once
AI answers are non-deterministic. The same prompt returns different sources on consecutive attempts. A single check tells you almost nothing, and most people who conclude "we are invisible" have run the query once.
Here is a procedure you can run this afternoon without any tooling.
Write ten prompts in buyer language
Not keywords. Full questions, the way someone would actually type them. Pull them from sales calls and support tickets rather than a keyword tool. Mix problem-stage questions with comparison-stage ones.
Name your competitor set before you start
Five to eight companies you genuinely lose deals to. Writing this down first stops you adjusting the list once you see who actually appears.
Run each prompt three to five times
New conversation each time, no memory or personalisation carried over. If you use the same chat thread, earlier answers contaminate later ones.
Record four things per run
Were you cited with a link. Were you named without one. Which competitors appeared. Was anything said about you factually wrong. A spreadsheet is enough.
Read the pattern, not the worst run
If you appear in two of five runs for a prompt, you are present but unstable, which is a different finding from absent. Count across runs.
What this means for you
- Budget an afternoon, not a quarter. Ten prompts × four runs × two platforms is about two hours of clicking.
- Do it before you brief an agency. Walking into the call with your own numbers changes the conversation entirely.
- Keep the spreadsheet. It becomes your baseline, and without a baseline nothing you do afterwards can be proven.
Step two: identify which state you are in
| What you see | What it means | What fixes it |
|---|---|---|
| Never appears in any run | Coverage and authority gap. Nothing sufficiently authoritative connects you to this query family | Topical depth and ranking strength — see SEO |
| Named, but the citation goes elsewhere | A third-party source is the one being used to describe you | Map and improve that source landscape — see AEO |
| Named, but the facts are wrong | Entity resolution failure. Usually a stale third-party profile | Entity work — see entity optimisation |
| Appears in some runs, not others | Borderline. You are in the candidate pool but not a preferred source | Usually the cheapest to move. Strengthen the specific pages already being considered |
| Competitors appear consistently, you never do | They occupy the source pool the model draws on | Displacement is slow. Start where their hold is weakest rather than where they are strongest |
Step three: find out where it is getting your information
This is the step almost everyone skips, and it is the one that turns a complaint into a plan.
Ask the assistant directly what it is basing the answer on. Then ask it to compare you against a named competitor and watch which sources it reaches for. You are not asking it to be honest about its architecture. You are looking at which domains keep reappearing.
In most B2B categories the recurring sources are review platforms, comparison articles from independent publishers, community discussion, and technical documentation. Marketing pages appear least often. If the same three domains describe your category in every run, those three domains are your visibility problem, and no amount of work on your own blog changes that.
Peer-reviewed work supports the pattern. The KDD 2024 paper that introduced the term GEO found that adding quotations, statistics and cited sources measurably raised visibility in generative engines, while keyword stuffing produced little to no improvement. Being the source others cite matters more than optimising your own copy.
Why ranking well on Google may not help here
ChatGPT runs its own retrieval rather than grounding in Google's ranking. In 2026 research its exclusive citation rate was 73.5% — most of what it cited was cited by nothing else in the sample. Strong Google positions help less on this surface than on AI Overviews, where citations are drawn from Google's own index.
So the answer to "we rank first, why aren't we in ChatGPT" is usually: because ranking is not the input it is using. That is uncomfortable and it is the honest read.
The crawler check nobody does first
Before diagnosing anything else, open your robots.txt.
OAI-SearchBot handles discovery for ChatGPT Search. GPTBot is a training crawler. They are separate decisions. A surprising number of sites block both, because someone pasted a rule from a blog post about AI content scraping.
If OAI-SearchBot is disallowed, you are not invisible for any interesting reason. You asked to be. Fix that before spending anything on content.
What not to do about it
- Do not add an llms.txt file expecting a change. Google states it gets no special treatment, and adoption sits under 10% of the top 1,000 websites.
- Do not fragment your pages into chunks. There is no requirement to do this and it makes content worse for readers.
- Do not buy mentions or seeded reviews. Inauthentic mentions are named in Google's spam policies, which now apply explicitly to generative responses, and incentivised endorsements carry separate regulatory exposure for the brand.
- Do not publish fifty pages about it. Ahrefs found 1.74% of newly published pages reach the top ten within a year, down from 5.7% in 2017.
- Do not accept a guarantee. Google states directly that no third-party tool has access to its internal ranking or AI systems. A guaranteed placement is a promise about a system the seller cannot see.
What usually moves it fastest
In our experience the highest-yield fix is the third state, not the first. Brands are more often described incorrectly than absent entirely, and correcting a stale third-party profile that the model is drawing on produces visible change sooner than building new authority from nothing.
It is also the cheapest fix, which is why we check for it first. Updating a two-year-old pricing tier on a comparison site is an afternoon of work. Displacing an incumbent source is a year of it.
Observed Pattern from our own diagnostic work. Your distribution across the three states comes from an audit, not from this page.
What this means for you
- Fix accuracy before chasing presence. Cheaper, faster, and a wrong answer costs more than no answer.
- Audit the three domains that keep appearing, not a generic directory list.
- Do not judge a fix for one quarter. Noise dominates at short horizons, and academic work on citation stability found single-run metrics give a misleadingly precise picture.
When to stop diagnosing and start measuring properly
The procedure above gives you a directional read in an afternoon. It will not give you a baseline you can defend to a board, because ten prompts run manually is a small sample of one platform on one day.
If the answer matters commercially, move to a frozen prompt set of 40 to 120 prompts sampled across platforms on a fixed cadence. That is what proper measurement looks like, and you can run it yourself: the method is published. If you would rather not, the two-week baseline does it across five platforms.
02Common questions
Questions people ask next
How do I check whether ChatGPT mentions my brand?
Take ten prompts a real buyer would type, run each three to five times in fresh conversations, and record whether you were cited with a link, named without one, or absent. Repetition matters because the same prompt returns different sources on consecutive runs.
Why does ChatGPT recommend my competitors instead of me?
Usually because they occupy the third-party sources it draws on for that question — review platforms, comparison articles or community threads. It is rarely a judgement about product quality, and your own marketing pages cannot fix it directly.
Does ranking first on Google get me into ChatGPT?
Less than people expect. ChatGPT runs its own retrieval rather than grounding in Google's ranking, and in 2026 research its exclusive citation rate was 73.5%. Ranking helps far more on Google AI Overviews than it does here.
Will adding an llms.txt file help?
No evidence that it will. Google states the file receives no special treatment and adoption sits under 10% of the top 1,000 websites. Understand it, decide deliberately, and do not pay for it as a visibility service.
How long does it take to appear once we fix the cause?
Correcting a wrong fact in a source the model already uses can show within weeks. Building authority where none exists takes months, and on a new domain in a competitive category, twelve to eighteen months is a realistic planning assumption.
Is being mentioned without a link worth anything?
Yes, and it is invisible to analytics. A buyer who hears your name in an answer may search for you directly afterwards, which shows up as branded search rather than AI referral. Judge the channel on brand-search movement as well as clicks.
Can an agency guarantee I will appear in ChatGPT?
No. Google states plainly that no third-party tool has access to its internal ranking or AI systems, and answers are probabilistic. Treat a guarantee as disqualifying rather than as confidence.
Should I block GPTBot?
That is a training-data decision and it is separate from search visibility. Blocking GPTBot does not reduce your presence in ChatGPT Search. Blocking OAI-SearchBot does, and many sites block both by accident.
Sources
Every figure on this page comes from one of the following. Where a claim is ours rather than a source’s, it carries an evidence label instead.
- Google Search Central — Optimizing your website for generative AI features (published 15 May 2026, updated 10 July 2026). Source for the llms.txt, chunking and AI-schema positions, and for the statement that no third-party tool has access to Google’s internal ranking or AI systems. Primary
- Google Search Central — spam policies, now explicitly in scope for generative AI responses. Source for the scaled-content and inauthentic-mentions positions. Primary
- OpenAI — bots and crawlers documentation. Source for the OAI-SearchBot and GPTBot distinction. Primary
- Rankability — State of AI Search, 1,645 AI citation observations across 31 topics, 2026. Source for the 73.5% exclusive citation rate and the 24.1% cross-platform overlap. Independent study
- Ahrefs — new-page study, one million URLs first crawled September 2023, tracked twelve months. Source for the 1.74% top-ten figure. Independent study
- Aggarwal, P. et al. (2023). GEO: Generative Engine Optimization. arXiv:2311.09735, KDD 2024. Finds that adding quotations, statistics and cited sources raises visibility materially, while keyword stuffing does not. Peer-reviewed
- Liu, N. F., Zhang, T., Liang, P. (2023). Evaluating Verifiability in Generative Search Engines. arXiv:2304.09848, Findings of EMNLP 2023. Audited four generative search engines and found roughly half of generated sentences fully supported by their citations. Peer-reviewed
- Searlixa diagnostic work across client engagements. Any figure drawn from this carries an Observed label with its sample and date. Our own
Ask an assistant
Hand this page to an AI assistant and ask it to summarise or challenge the argument. If it gets something wrong about us, that is itself a useful finding.
03Related
Continue
How to measure AI visibility · How ChatGPT selects sources · Entity optimisation · Request an audit