This page explains, without hedging, what AI visibility work can do, what it cannot do, and what we will never promise you.
AI answer engines are third-party systems we do not own, control, or have any relationship with. We cannot guarantee that your business will appear in any AI answer, at any position, at any time — and neither can anyone else. What we can do is measure where you stand, fix the signals that are actually within your control, and report honestly on what changes.
Our work is measurement, diagnosis, and improvement of eligibility. Specifically:
That is the whole of it. Everything else follows from those four things or is not something we do.
Vague claims about “testing AI engines” are easy to make and impossible to check, so here is exactly what happens, in enough detail that you or your own marketer can hold us to it.
| Engine | Queried? | How, exactly |
|---|---|---|
| ChatGPT | Yes | Through the provider's API with web search enabled. Every run records the exact prompt text, the timestamp, the model version, the full response, and every source the answer cited. |
| Perplexity | Yes | |
| Claude | Yes | |
| Google Gemini | Yes | |
| Google AI Overviews | Yes | Captured from the live results page, with the overview's own cited references. This is a different surface from Gemini — different selection logic, different output — so the two are reported separately rather than blended into one “Google” number. Google does not generate an overview for every query; where none appears, we record the absence instead of calling it a failure. |
| Bing Copilot | No | Microsoft publishes no API for Copilot's answers, and Azure's “Grounding with Bing Search” tool deliberately withholds its output from developers — so there is no way to obtain a Copilot answer we could stand behind as evidence. We therefore do not test it and make no claim about it. There is a better route for your own site: Bing Webmaster Tools now reports Copilot citations for sites you control, first-party and free. We will point you at it during onboarding, which is more useful than a scraped approximation. |
| Web search index, Google Maps, directories, your own pages and structured data | Yes | Fetched and parsed directly. Your site is crawled page by page, not sampled from its homepage. |
The APIs, not the phone apps. This is the distinction that matters most, so we put it in §3 of every report as well as here. Where an engine is reached through its API, that API draws on the same retrieval substrate and exposes the same citation behaviour — which is what an audit reasons about — but it carries no session history, no personalisation, and none of the interface layer those products test on real users. Treat such results as evidence about the retrieval layer, not as a screenshot of what one specific person on one specific phone would see.
Not every engine on the list above is reached the same way. Your report states which is which, per row, before you read a single finding — and where an engine was not queried at all, it says that too, with the reason. A row marked “Not tested” is not an omission we hoped you would miss.
We could describe all of this more loosely and you would probably never know. We would rather you could check.
Every run is logged, favourable or not. A query where your business does not appear is recorded with the same care as one where it wins, and runs that fail outright are listed as errors rather than quietly dropped. The prompt set is fixed before any result is seen, so nothing can be added or removed to flatter the outcome. That log is the deliverable; the commentary is just reading it.
What a single automated cycle cannot do. Claiming and reading each third-party directory profile by hand — the work that finds four different suite numbers for one address in public circulation — is not automatable. It is part of ongoing engagements, not of a one-time audit, and every report says so in the section where it would otherwise be missed.
One cycle is a snapshot on one day. The same prompt run twice, minutes apart, can differ. The first genuinely meaningful comparison is the 90-day re-run, and it will include the results that did not move.
We have no control over, and no special access to:
We are not affiliated with, endorsed by, certified by, or a partner of OpenAI, Google, Microsoft, Perplexity, Anthropic, or any other AI or search provider. Their names appear in our materials only to describe what we test. All trademarks belong to their owners.
We will never claim, and you should be skeptical of anyone who does, that we can:
On the word "guarantee." Our website says we guarantee the work. That means one specific thing: we commit to performing the scoped services with reasonable professional skill and care, to testing honestly, and to reporting what we find whether or not it is what you hoped. It is a commitment about our conduct. It is not a warranty of any result, outcome, ranking, or performance level, and it should not be read as one.
Honesty about method matters more here than in most fields, because the underlying systems are non-deterministic.
To test what AI systems say about you, we type information about your business — name, website, service, city or service area, and sometimes a competitor name or a customer question — into third-party AI products including ChatGPT, Gemini, Bing Copilot, Perplexity, and Claude. Those submissions are governed by those providers' terms and privacy policies, which we do not control. We submit business information only, never personal information about you or your customers. See our Privacy Policy, Section 7.
Nothing we publish or deliver is legal, regulatory, medical, clinical, financial, tax, or compliance advice. We do not review your content for compliance with advertising rules, professional licensing rules, healthcare regulation, or any other legal requirement, and our engagements expressly exclude that work. You are responsible for the accuracy and legality of every claim, credential, and description we publish on your behalf, and you should have it reviewed by a qualified professional before it goes live.
AnswerFoundry's founder is a physician. He does not act in a clinical capacity for AnswerFoundry, and AnswerFoundry provides no medical services and no medical advice to anyone, including its clients and their patients. The credential describes the founder's background and the origin of the method, not the nature of the service.
Articles, guides, and blog posts on this site are general information about a fast-moving field. They describe our observations at the time of writing, may become outdated quickly, and are not tailored to your business. Do not treat them as a recommendation for your specific situation.
Where we describe our own results — including results for the founder's own medical practice — we say so plainly and we identify the relationship. Results described for any business are that business's results and are not a prediction of yours.
This disclaimer is part of our Terms of Service. The disclaimer of warranties in Section 14 of the Terms and the limitation of liability in Section 15 apply to everything on this page. If this page conflicts with Section 10 of the Terms, Section 10 controls.
If anything here is unclear, or if you think we have overclaimed somewhere on this site, tell us: hello@answerfoundry.ai. Being called on it is better than the alternative.