01ChatGPT
ChatGPT visibility: the most exclusive source list of any platform
Quick answer
ChatGPT runs its own retrieval rather than grounding in a search engine’s ranking. In 2026 research its exclusive citation rate was 73.5%most of what it cited was cited by no other platform in the sample.
The practical consequence: your off-site footprint matters more than your own site. Earn mentions on the comparison pages, review platforms and community threads it already cites, keep your brand facts consistent, and answer directly in the first hundred words.
Key takeaways
- Check robots.txt before anything else. OAI-SearchBot controls ChatGPT Search discovery. GPTBot is a training crawler. Many sites block both by accident.
- Ranking on Google carries over poorly. One study found 37% of AI-cited domains were absent from traditional search results entirely.
- Being retrieved is not being cited. Roughly 85% of retrieved pages never appear in the final answer, so your content can be in the pool and earn nothing visible.
- Reviews carry unusual weight here. Brands with active review-platform profiles were around three times more likely to be cited in one 2026 analysis.
- Visibility is unstable. Only about 30% of brands appearing in one answer stayed visible in the next. Judge trends over three periods.
- Misdescription is more common than absence, and usually cheaper to fix.
What we observe about its behaviour
- Source selection is narrow and idiosyncratic. It frequently draws on a source no other platform touches for the same prompt. Ranking is a weaker predictor here than on Google AI Overviews.
- Third-party corroboration carries weight. Review platforms, comparison sites, community threads and industry publications often outrank a brand's own site as the source it uses.
- Attribution is inconsistent. It will often name a brand without linking. Those mentions are real and invisible to analytics.
- Answers vary between runs. The same prompt returns different sources on consecutive attempts, which is why single checks are worthless and repeated sampling is not optional.
- Retrieval and citation are different steps. A page can be retrieved, read, and left out of the answer. Reported figures put that gap at roughly 85%.
The crawler distinction that matters
OAI-SearchBot handles discovery for ChatGPT Search. GPTBot is a training crawler. They are separate decisions. Blocking GPTBot does not reduce your visibility in ChatGPT Search. Blocking OAI-SearchBot does.
A surprising number of sites block both, because someone pasted a rule from a blog post about AI scraping. If you are diagnosing absence, this is the check to run first.
Which sources it actually uses
This is the part that changes what you should spend money on. Across the categories we sample, the recurring sources are consistent, and your own domain is rarely the strongest.
| Source type | Why it gets used | How you earn a place |
|---|---|---|
| Comparison and roundup articles | Independent, structured, directly answers a shortlist question | Pitch the author with a clean factual description, not a sales pitch |
| Review platforms | Independent, current, and snippets lift almost verbatim into answers | A real request flow to real customers. Never incentivised |
| Community threads | Reads as unfiltered practitioner opinion | Genuine, disclosed participation. Seeding gets caught and costs trust |
| Technical documentation | Specific, factual, structured, frequently updated | Yours already. Make sure it is not blocked from crawlers |
| Original research | Nothing else can supply the number | Publish a real dataset with method and limitations |
| Your marketing pages | Weakest signal — the model knows it is your story about yourself | Still necessary. Just not sufficient |
The prompt classes that decide shortlists
Not all prompts are worth the same. Four classes carry most of the commercial weight, and the highest-intent one is usually the least optimised for.
- "Alternatives to [competitor]." The highest-intent class we see. A buyer asking this has already rejected someone and is actively shopping.
- "Best [category] for [use case]." Decides whether you make the shortlist at all.
- "[You] vs [competitor]." Where misdescription costs deals fastest, because the model states a comparison as fact.
- "Downsides of [you]" or "why do people leave [you]." Objection checking. Worth sampling even though it is uncomfortable reading.
The failure mode we see most often
Not absence. Misdescription. ChatGPT names the brand and then states pricing, positioning, ownership or product scope incorrectly, usually because it is drawing on a stale third-party profile rather than the company's own site.
That is an entity problem — see entity optimisation, and it is generally the highest-value fix available, because correcting a two-year-old pricing tier on a comparison site is an afternoon of work while building new authority is a year of it.
What this means for you
- Spend off-site before on-site. If the same three domains describe your category in every run, those three domains are the programme.
- Sample your own "vs" and "alternatives" prompts monthly. They move fastest and they cost the most when they are wrong.
- Do not judge a change on one week. With 30% inter-answer stability, weekly readings are noise.
Our measurement method
Repeated sampling against a frozen prompt set
Each prompt runs three to five times in fresh conversations, so non-determinism shows up as a distribution rather than as a misleading single answer.
Presence coded three ways
Cited with a link, named without a link, or absent. These need different responses and collapsing them into one number hides the finding.
Description accuracy coded from full text
We record how it characterises you, not just whether it mentions you. A drop in accuracy is more urgent than a drop in presence.
Source landscape mapped
Which third-party sources it uses for your prompts. That list, not a generic directory list, is where reputation work should go.
On referral volume
ChatGPT referral traffic remains a fraction of organic for most brands. Several 2026 datasets report AI-referred visitors converting substantially better than organic, but the absolute volume is small and a large share of influence leaves no analytics trace at all.
We report both facts rather than the flattering one. If a vendor quotes only the conversion multiple and not the volume, ask for the volume.
Observed Behavioural descriptions come from our own sampling and from the published research listed below. Method and limits: our methodology.
02Common questions
Questions people ask about ChatGPT visibility
How do I get my brand cited in ChatGPT?
Earn mentions on the third-party pages ChatGPT already cites for your category, keep your brand facts consistent everywhere, and structure your own pages so a direct answer sits in the first hundred words. Your own site is the weakest of the three signals.
Why does ChatGPT cite my competitor and not me?
Usually because a comparison article, review platform or community thread it already trusts names them and not you. It is rarely a judgement about product quality, and it is not something your own marketing pages can fix directly.
Does ranking on Google help me appear in ChatGPT?
Less than on Google's own AI surfaces. ChatGPT runs its own retrieval, and in 2026 research its exclusive citation rate was 73.5%. One study also found 37% of AI-cited domains were absent from traditional search results entirely.
Which crawler do I need to allow?
OAI-SearchBot for ChatGPT Search discovery. GPTBot is a training crawler and a separate decision. Blocking GPTBot does not reduce your visibility in ChatGPT Search; blocking OAI-SearchBot does.
Do reviews affect ChatGPT citations?
They appear to matter more here than on most surfaces. One 2026 analysis found brands with active review-platform profiles were roughly three times more likely to be cited. Review snippets also appear close to verbatim in answers.
Why does my visibility keep changing week to week?
Because answers are non-deterministic and retrieval shifts. One 2026 study found only about 30% of brands appearing in one AI answer stayed visible in the next. Judge a trend over three sampling periods, never one.
What content format gets cited most?
Comparison and alternatives pages, original research with numbers, and definition-first explainers. Corporate homepages and brand storytelling appear least often, because there is nothing in them a model can lift as a claim.
How much of what ChatGPT retrieves actually gets cited?
A minority. One 2026 analysis reported that roughly 85% of retrieved pages never appear in the final answer. Your content can be in the retrieval pool and still earn no visible credit.
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.
- Rankability — State of AI Search, 1,645 AI citation observations across 31 topics, 2026. Source for the 73.5% exclusive citation rate. Independent study
- Omnibound — Getting cited in ChatGPT (May 2026). Source for the finding that roughly 85% of retrieved pages never appear in a final answer. Industry analysis
- xSeek — citation tracking across ChatGPT, Claude, Perplexity and Gemini. Source for the figure that only about 30% of brands appearing in one answer stay visible in the next. Industry analysis
- AuthorityTech, reporting an HKUST study (April 2026). Source for the finding that 37% of AI-cited domains are absent from traditional search results. Academic, reported
- LLM Pulse, reporting SE Ranking analysis (May 2026). Source for the roughly threefold citation likelihood associated with active review-platform profiles. Industry analysis
- OpenAI — bots and crawlers documentation. Source for the OAI-SearchBot and GPTBot distinction. Primary
- Searlixa sampling across client engagements. Any figure from this carries an Observed label with its sample and date. Our own
Ask an assistant
Hand this page to an 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
Diagnose absence yourself · How the platforms differ · Answer engine work · Entity optimisation · Pricing