What Answer Radar Does Not Do

It does not tell you how many people asked

No engine publishes prompt volume for a private assistant. Anyone quoting one is estimating, and Answer Radar will not attach an estimate to your prompts or rank proposals by one. Fan-out shows what engines searched on your prompts, and Search Console shows what real buyers typed into Google, which are both observed. Volume inside ChatGPT, Claude, Gemini or Perplexity is not.

It does not attribute visits or revenue to answers

Assistants pass little or no referrer, so a claim that an answer sent a visit is inference. Answer Radar measures presence, recommendation, position, sentiment and citation, which are observable in the answer, and reports how they change over time. It does not report AI traffic, and it does not connect a visibility number to pipeline.

It does not say why an engine chose a source, or that your fix caused a change

Observed facts, mentioned, cited, retrieved, searched, are kept apart from hypotheses about what may contribute and from testable actions. No result states that a page was cited because of anything. Re-verification reports an observed delta between two measurements on the same instrument after an indexing delay, with control prompts, and has no field in which a cause could be recorded. When the controls moved as much as the target, it says the change was not isolated.

It does not reproduce the consumer apps

Every metric set carries the same note: measured through official provider APIs with grounded web search, a controlled measurement proxy and not an exact reproduction of consumer ChatGPT, Claude, Gemini or Perplexity, specific to the provider, model, tool version, locale, prompt and sampling configuration disclosed alongside. A consumer session with memory, history and personalisation is a different instrument, and no vendor can measure it at scale.

It does not write your brief, your fix or your card text

The measurement layer generates no prose. Gaps carry observations, citations, hypotheses and actions as data. Cards carry a label and evidence, no written brief. A fix draft is an evidence skeleton with no model involved. The words are written by you or by your MCP client from that evidence, and the faithfulness guard keeps them sourced. This is the doctrine the product is built on: expose the data and let the client do what it already does well.

It does not publish, and it does not approve for you

An approved fix is text you publish where you choose. Nothing in Answer Radar writes to your site. Approve and publish are human actions in the Fixes page and cannot be set from any tool.

It does not show model ids or currency

Customers see engine names and an opaque instrument token, and credits where one credit is one measured sample. The model snapshot is stored so re-verification can pair instruments correctly, and redacted at the presentation boundary. No spend figure is shown.

What is not yet shipped

Google AI Overviews is approved as the fifth engine and its pages say that run data populates when the engine ships; it is not measured today. Only two action card types are generated, content gap and competitor won, because they are the two claims a measured run can support; untracked page, community opportunity, setup gap, offsite mention and sentiment cards each wait for their evidence source. Daily audit schedules are restricted to internal accounts. Schedules are set through MCP rather than the web app.

• Google AI Overviews: approved, populates when the engine ships.
• Five action types: defined, not generated until their evidence exists.
• Daily schedules: internal accounts only.

FAQ