There is no single public list that ChatGPT consults to rank local businesses. A web-enabled answer is generated from the model, the wording and context of the prompt, and the sources retrieved for that run. Those sources can include business websites, directories, review sites, local publications, and other web pages. The named businesses can change when the prompt or run changes.
This is why asking once and taking a screenshot gives weak evidence. The screenshot is real, but it is one draw from a system that can retrieve a different source set next time. A useful audit holds the prompt and location steady, runs it several times, records the cited URLs, and reports how often each practice appears.
What the record changes
Entity consistency matters because retrieved pages need to refer to the same business. An old practice name, mismatched address, doctor's name used as the organization, duplicate location, or missing service page can split the public record. The audit identifies those conditions on sources actually present in the answers. It does not claim to know a private ranking formula or guarantee a future recommendation.
- OpenAI documents web search as a tool that can return cited web answers, while the model and tool behavior still depend on the request and run.
- Perplexity and Gemini publish their own search-grounding interfaces, which can retrieve and cite different source sets from OpenAI's.
- Sampling converts changing answers into an observable rate, but the rate remains specific to the prompts, APIs, dates, and run count in the report.
How to use this answer
An AI citation check defines buyer-intent prompts for one service and place, runs each prompt several times through the platforms' APIs, and records every named practice and cited source. The report uses mention and citation rates over a stated run count because a single answer can change on the next run. Entity checks then compare the practice's name, address, phone, service, and location pages across the sources that appeared.
- Choose the service, city, and buyer question before naming the practice you want to measure.
- Run each prompt several times on each API, save the verbatim dated answers, and calculate rates from the full run set.
- Inspect the cited directories, articles, and local pages for entity consistency, then prioritize fixes on surfaces the practice controls.
Where the service stops
Reality Contact, LLC runs prompts through platform APIs and reports dated sampled answers. API output is a proxy for consumer-app output and can differ from it. We report who appears and which sources the answer cites. We do not get a practice named, cited, ranked, or recommended, and we do not rate competitors. This is not legal advice or financial advice, and it is not advertising-compliance advice.
Sources: OpenAI web-search documentation; Perplexity Sonar documentation; Google Gemini grounding with Search.