Skip to main content
FullReach AI asks each prompt the way a buyer asks it: in the platform’s own app, not through the model’s API. Every answer that a counted day holds is an answer that a person in that market could see. A new project’s first hour is the one exception, and The first day describes it.

In the platform’s own app

Each day, each prompt goes to each platform through the page a person uses:
  • the chat page of ChatGPT, Gemini and Perplexity,
  • Google’s AI Mode,
  • a Google search results page, for AI Overviews.
The request is aimed at the prompt’s market, so the platform answers as it answers there. A Worldwide prompt is aimed at no country. The request is a logged-out visitor’s, in a fresh session, so no history, memory or account shapes the answer. Platforms lists each platform and where its answers come from.

Why the app and not the API

The same model can answer differently in its app and through its API. In the app, the platform decides the model, whether to search the web and how, as it decides for every visitor. An API call answers as the developer set it up. The app also shows what a buyer reads beside the answer. That is the cited sources in their order, and on some platforms the fanouts it ran and the product cards. The analysis reads those too. See Sources, Fanouts and AI Shopping. So the stored answer is the answer a buyer sees. The answer view shows it word for word, with the model when the platform names it. See The answer view.

When a request fails

A platform can block a request, or return a page with no answer. That is a technical failure, and it counts in no figure. FullReach AI asks again, up to 6 times for each prompt on each platform. Some tries go at once, and the rest go later. A prompt still without an answer gets the same tries again in three more rounds later the same day. So an outage of a few hours does not cost the day. When Google shows no AI Overview for a question, the answer is “No AI Overview”. That is a fact about Google, not a failure, and it counts against no brand.

The first day

A first collection usually takes under an hour. A new project’s dashboard does not wait for it. Five pages open at once on preview answers: the Visibility, Competitors and Sources tabs, All Prompts and All Answers. The first preview answers are the free check’s. An account with no trial shows only these, because its collection starts with the trial. A preview answer comes from a platform’s API, not from its app. ChatGPT, Gemini and Perplexity give them, to at most 25 of the project’s prompts. The same model can answer differently in its app, so the top bar of each of those pages carries the word Preview. So does each brand figure on the Visibility tab, and each preview answer, in a list and in the answer view. A banner on those pages says that the answers are still being collected, and that the dashboard updates until all of them are collected. It names the platforms that show preview answers for now, and the platforms still in collection. When a platform’s own answers are collected and analyzed, they replace its preview answers, and its figures change. When the first collection is complete, every preview answer is deleted. A platform whose collection gave no answer that day then shows none. No counted day holds a preview answer. Until the first collection is complete, those pages offer no CSV export. Their cards offer neither Save as image nor Copy to clipboard. Nothing marked Preview leaves the page. The Digest, the Monthly report, the Actions, the MCP tools and the Data Studio connector never read one. A project that was added to an organization later has no preview answers. Its pages open when its first platform is collected. A project paused on its first day hides the answers it holds so far, preview answers included. They show again when the project resumes.

Where the API is used

  • The free check asks its prompts through the platforms’ APIs. Its answers can differ from the daily collection’s. They are the first of the new project’s preview answers, and no counted day holds one.
  • Custom plans can also measure over the model vendors’ own APIs, as a source beside the apps and never instead of them. The pricing page lists those models.