THE METHOD
A useful snapshot.
With visible evidence.
What gets checked?
You provide your public website and the service, audience, and market your buyers ask about. The tool runs one buyer recommendation question through the OpenAI Responses API with live web search. It aims for three businesses, with at most four; insufficient evidence can produce fewer or none. It gives you a quick first look; a wider review needs more questions.
The questions do not name your business.
Your website is used to match the answers after the searches run. It is deliberately left out of the recommendation prompts, so the tool does not ask the model to include you.
What gets sent to the model?
“Prompt sent” shows the actual input text sent to the model for that saved check. The request also includes separate instructions for web search, relevant providers and structured output. It is not the whole API request. Your website is matched against the returned recommendations afterwards.
Reading the result
The shortlist shows the businesses returned by the model. “Check overview” contains the actual input prompt, a short answer overview, the model name and up to three relevant source links. Longer answers are shortened with an ellipsis; the complete answer and search sources are retained in the saved check. The result’s timestamps show when the recommendations and homepages were checked.
What this cannot tell you
This is an API visibility snapshot, not direct access to private ChatGPT conversations or a universal ChatGPT ranking. Answers can change with the question, model, location, search results, and time. A small sample does not measure your overall market share or guarantee a recommendation. ChatGPT and API answers can differ.
The short homepage comparison
The shortlist appears as soon as the AI search finishes. The submitted homepage and up to two recommended alternatives are then checked separately, so a slow website does not hold back the recommendations. This adds no second AI or search-tool request. The comparison shows each main headline or page title. Public DNS checks, short timeouts, a page-size cap and restricted redirects bound these reads.
The comparison covers readable homepage HTML only. It does not execute scripts, inspect every service page or establish the model’s ranking factors. Unavailable pages are labelled. A suggested next step can use other signals found in that same HTML, including a noindex instruction or links to client evidence. Recommendation cards show names and websites; their order refers to this answer only.
Timing, cached checks, and incomplete results
The form takes about 30 seconds to start. Live searches can take longer. A matching completed check may be reused for up to 24 hours and keeps its original recommendation timestamp. Earlier cached checks can receive a homepage comparison without another AI request; its separate time is shown. If no verifiable answer is returned, no score is generated.
From a snapshot to action
The result is a starting point. In the free guidance chat, we connect it with your offer, public information, and website to identify three actions you can take next.
For examples matched to a specific service, explore the industries. Changing an example or city does not run a check; you choose when to submit your website and question.
Start my free check ↗