Platform
ChatGPT visibility
Conversational assistants are where buyers most often ask for a recommendation outright. This page explains what makes a business surface in that kind of answer, and what we can and cannot measure about it.
AI Says is not affiliated with or endorsed by OpenAI, Google, Anthropic, Perplexity or other AI/search providers. Platform names are used descriptively to identify the services being assessed.
Two very different ways you get described
A conversational assistant has two sources for anything it says about your business, and the difference between them explains most of what businesses find confusing about their own AI visibility.
The first source is the model itself — a compressed internal representation built during training. When the assistant judges that it already knows enough, it answers from that, and nothing is fetched. Descriptions produced this way are fluent, confident and potentially years out of date. They are also the hardest to influence, because changing them means changing what the public web says about you before the next training cycle, not merely changing your own site.
The second source is retrieval. When the question is specific, local, recent or commercially loaded, the assistant reaches for live sources. This is where your website can matter directly — but only if it is retrievable, resolvable as a business entity and written in a way that yields an extractable answer. Retrieval also means the assistant issues its own search queries, not your customer’s words, so the phrasing you have optimised for may never be the phrasing used.
Why specific questions favour specific businesses
Ask an assistant for “the best accountant” and it will hedge, because the question has no resolvable answer. Ask for “an accountant in Newcastle who handles construction subcontractors” and it must name somebody. Narrow, qualified questions are where small and mid-sized businesses are genuinely competitive — and they are the questions most businesses have never tested.
Integration required
Observed data needs a connected integration
AI Says reports observed ChatGPT data only where an OpenAI API key (OPENAI_API_KEY) has been configured for the account and the provider is enabled. Until then the platform panel shows Integration required rather than a number, and the components that depend on observed answers are labelled Estimated.
We would rather show you an empty state than a figure we did not measure.
What is assessed without an integration
The full twenty-four-signal website audit, your entity corroboration, your question-coverage gaps and your buyer-question library all run regardless. They are the inputs to whether an assistant can use you at all, and they are measured from your own site. See the signal list.
What we measure
Four rates, per platform
Conversational assistant. The most common place buyers ask for recommendations. Each of these is a proportion over recorded answers — never a single answer treated as a verdict.
- VisibilityWere you mentioned at all?
- The proportion of observed answers in which your business is named. Measured across the buyer-question library for your category, not against prompts containing your brand name. This is the blunt measure: you are either in the answer or you are not.
- RecommendationWere you put forward, or just listed?
- The proportion of answers where you are actively recommended rather than mentioned in passing. The distinction matters commercially: appearing in a list of nine local firms is not the same as being the one the assistant suggests the customer contact.
- CitationWas your own site the source?
- The proportion of answers that cite a URL on your domain. Where an assistant cites a directory or a competitor instead of you on a question about your own services, that is a retrievability problem rather than an awareness problem.
- AccuracyWas what it said true?
- A score for how correct and complete the description of your business was, assessed claim by claim against what is verifiably true. An inaccurate description is usually faster to fix than a missing mention, and it is more damaging while it persists.
Every rate is shown with the number of observations it rests on, because a 50% visibility rate over two answers and the same rate over forty are different claims. How observations feed the score.
Practical
What moves the needle for an assistant
Ordered by how directly it affects whether an assistant can include you in a recommendation.
- 1
Make your identity unambiguous
One consistent legal name, one canonical address, Organization schema, and sameAs links to profiles that say the same things. An assistant that cannot resolve which business you are will not name you at all — hedging is always the safer output for it.
- 2
State your qualifiers explicitly
The industries you serve, the customer types, the suburbs or regions, the things you deliberately do not do. Assistants answer qualified questions, and they can only match a qualifier that appears somewhere in words.
- 3
Answer the narrow questions in public
A page that directly answers "how much does X cost in Y" or "what should I ask before hiring a Z" is retrievable for exactly the prompts buyers type. A services page that mentions twelve offerings is retrievable for none of them.
- 4
Get corroborated somewhere other than your own site
Industry bodies, registries, reputable directories, local press, review platforms. Corroboration is what converts "a business claims this" into "this is the case" for a system that has no other way to tell.
- 5
Keep something dated and current
Visible recency gives an assistant a reason to prefer a retrieved fact over its own stale internal one. Static sites are described from memory more often than living ones.
Other surfaces
One assistant is not the whole picture
Each platform retrieves differently, cites at a different rate and therefore tells you something different.
Gemini visibility
Google-adjacent retrieval, plus the separate question of AI Overviews inside search results.
Read the pagePerplexity visibility
Cites its sources by default, which makes it the clearest read on whether your own site is being used.
Read the pageMonitoring over time
Assistant behaviour changes between model releases. One reading is a snapshot, not a state.
Read the page
Questions
Questions about ChatGPT visibility
Because it does not always retrieve. When an assistant judges that it already knows enough to answer, it answers from its internal representation of the subject rather than fetching anything. That representation reflects the state of the public web at the point the model was trained, so a rebrand, a new service line or a move to new premises can be invisible to it until either retrieval is triggered or the model is updated.
For a specific local or niche question, it usually comes down to whether the business is unambiguously resolvable as an entity and whether independent sources corroborate it. Assistants avoid naming a business they cannot confidently identify. For broader questions, the named businesses tend to be the ones written about most in the sources the assistant can reach — which is frequently a directory or a listicle rather than the businesses themselves.
Not through AI Says, and we would be sceptical of anyone claiming otherwise for organic answers. What you can do is remove the reasons an assistant would skip you: unresolvable identity, no corroborating sources, no explicit statement of what you do and where, and nothing quotable on your own pages.
Very little, unfortunately. Asking directly about a named business almost always produces a description, because the question supplies the identity. The informative test is the question a customer would actually ask — a category and a location with no brand name in it — and whether you appear in the answer at all.
It replaces estimates with observations. Visibility, recommendation, citation and accuracy rates for this platform become counts over real recorded answers, each one timestamped, with the answer text retained so accuracy can be checked line by line. Until an integration is connected, the platform panel reads "Integration required" and the dependent components are labelled Estimated.
AI Says is not affiliated with or endorsed by OpenAI, Google, Anthropic, Perplexity or other AI/search providers. Platform names are used descriptively to identify the services being assessed.
Find out what AI says about your business
Enter your website and get your AI Says Score. No credit card required.