Loading the page
Opening the page…
Opening the page…
PromptFace collects AI answer data for SEO and GEO analysis. We keep the full prompt, recorded response, model identity, measurement date and available source context connected, so each finding can be traced back to its record.
We collect answers from ChatGPT, Google AI Overview and Google AI Model. Each recorded answer keeps its model identity, response date and available sources. Answers can differ between products and collection methods.
Prompts are written or imported by our editor. Each question has its own configured models, covering ChatGPT, Google AI Overview and Google AI Model. Available models are listed with each question. We run a weekly measurement with two independent responses per model. The public page shows only the most recent available successful response, chosen deterministically from that measurement.
A record connects the full prompt and AI response to the model, response date, company mentions and recorded sources. For cited sources, we fetch and save the full page text, including the cited fragment and its surrounding context. The source URL, domain, title and capture date remain attached to that record.
The date next to an answer is the actual response date. If a more recent measurement has no successful response, the last available answer retains its original date. Different models may have different response dates. Past responses and technical execution details are not shown in the public library.
Compare repeated measurements of the same prompt and model to identify changes in answers, company mentions and source URLs or domains. Keep the wording, model identity and collection context in view; a changed prompt or model can affect the comparison. A source appearing or disappearing is an observed difference, not proof of a change in a model’s preferences or the effect of an SEO action. We confirm historical coverage and the comparison period for each report.
A cited source is supported by the provider’s citation annotations. A link in the answer is not automatically a confirmed citation. A retrieved search result or mentioned domain does not prove that a page was used. Source labels preserve these distinctions.
When an AI answer cites a source, we follow the source URL, fetch the whole page and save its full text. We keep the cited fragment together with the surrounding paragraphs and the rest of the page, so the source can be analyzed as a complete document. The saved page is connected to the prompt, AI response, model and measurement; its capture date is recorded separately from the answer date.
For example, if AI cites a CRM product page for its pipeline features, the source record keeps the cited passage along with the page’s introduction, feature descriptions and other surrounding text. Your team or AI agent can inspect the whole document rather than work from an isolated excerpt.
A page can change between the AI response and our capture. The saved text represents the page we fetched at the recorded capture time; it is not a guarantee of the exact page version the model saw. Source captures and provider citation metadata keep their distinct roles.
The catalog contains English prompts. Search matches the meaning of your question across languages and combines it with text matching. It finds existing questions; it does not generate an answer on demand. If semantic search is unavailable, we clearly label text-only results.
Reports package the agreed prompt scope and available measurements as CSV/JSON for analysis in your own tools or by an AI agent. We confirm the fields, source content, historical coverage and access date before an order. The public library shows the latest available answer; it is not the complete historical dataset. Direct MCP access and an analysis dashboard are planned, not currently available.
Responses may vary, omit relevant companies or contain errors. Citations are not a complete list of training data or an explanation of a model’s reasoning. A mention is not an endorsement, a ranking or proof of market share. Demonstration responses are explicitly labeled.