How our AI visibility audits work
See what we test, what we save and what the results can tell you. Website checks, AI answers and customer inquiries are measured separately.
What the free website check observes.
The scanner retrieves one public website page without running its JavaScript, then checks robots.txt and one sitemap location. It looks for ten signals: HTTPS, noindex directives, OAI-SearchBot rules, sitemap XML, business-related JSON-LD, business name, service phrase, city, contact links and a page title.
| Status | Meaning |
|---|---|
| Passed | The signal was observed in the retrieved response. This does not confirm indexing or a recommendation. |
| Review | A signal was missing, an exact phrase was not found or a restriction was observed. The evidence determines whether a change is needed. |
| Unconfirmed | Access or response content was insufficient to determine the signal. It remains unmeasured. |
A site using browser rendering, a firewall, synonyms or information on other pages may need manual review. Optional markup and exact text matches are useful clues, not mandatory AI requirements. A missing clue is different from a confirmed crawler restriction. We do not turn the ten checks into an AI visibility score.
Run the free website checkAccess and search eligibility.
OpenAI’s crawler documentation distinguishes OAI-SearchBot, used for search, from GPTBot, used for model training. Allowing a search crawler can support access; infrastructure restrictions and other eligibility conditions still apply.
Google’s AI features guidance says existing search practices apply to AI Overviews and AI Mode. Pages need search indexing, snippet eligibility and inclusion through Search Console’s generative AI control. Google ignores llms.txt for Search; no special AI schema is required. Structured data, when used, should match visible content.
An authorized owner should inspect Settings → Search generative AI in Search Console. The control can include, exclude or inherit a parent property’s choice. We record the effective setting when owner-provided evidence is available; our public scan cannot inspect it. This controls participation in specified generative AI features, separately from crawling, indexing and model-training controls. Confirm the intended effect before changing a setting.
How recommendation observations are collected.
Before research starts, the owner approves one business, one primary market, 20 buyer questions and the collection conditions. The planned surfaces are ChatGPT Search, Google AI Mode and Gemini. We confirm availability and ask before substituting a surface.
| Group | Planned attempts | How it is reported |
|---|---|---|
| Baseline | 20 questions × 3 surfaces = 60 | Successful responses and unavailable attempts are reported separately. |
| Repeats | 5 representative questions × 2 additional runs × 3 surfaces = 30 | The repeat sample has its own successful-response denominator and stability observations. |
Each record identifies the prompt, date, surface, collection method, interface or API, account conditions and location settings where controllable. We save the available response and source links, with screenshots where available. An API response is labeled as an API response; it is not presented as a consumer-interface screenshot.
A business mention, an explicit recommendation and a cited page are different observations. Failed, blocked or unavailable responses are not counted as business absences. Attempts not yet collected are recorded separately from unsuccessful attempts; neither is a negative recommendation. An answer without usable sources remains an answer with that limitation.
The findings describe the recorded sample. They are not a permanent ranking, a probability of being recommended to every buyer or proof of customer demand.
How evidence becomes a decision.
We connect observed issues to accurate business facts, useful service content, location information, profiles, technical access and sources in the tested answers. Five priorities identify the evidence, effort, dependencies, owner and completion condition. A useful draft or patch is prepared where the findings support it. Installation and future comparisons require a separate agreed scope.
The independent HVAC demonstration currently documents public website observations and a proposed draft. The complete recommendation report and human reviewer sign-off are pending. It does not establish a customer outcome.
What measurement can prove.
Search Console’s Generative AI performance report measures impressions of links to a site in supported Google Search AI features. It complements our sampled answers. It does not measure every business recommendation, ChatGPT or Gemini visibility, leads or booked jobs. We use owner-provided report evidence when available, without treating missing access or a missing report as zero visibility.
Google warns that unavailable values displayed as ~ or - can become zeros in exported data. We preserve that uncertainty and record dates, filters and dimensions before interpreting an export. Where verified inquiry or purchase records are available, we report them separately from sampled answers. Recommendation changes alone do not prove more leads or revenue. Comparable checks retain the question set and disclose changed conditions; they cannot make an uncontrollable platform constant.
Private report links expire after seven days. Contact support to request renewed access. Read the audit’s scope, timing and refund policy.
Find out who AI recommends in your market.
See the answers about your offerings, the competitors named and five priorities for your website and business information. $297 once for one business and one primary market.
No payment now. We’ll contact you when audits reopen.
See everything includedFive business days after complete onboarding and scope approval. No automatic monthly enrollment.