Evidence you can inspect

Set the questions and conditions

The standard audit covers 25 questions we select independently from your business context, four repeats on each of five specified API engines: ChatGPT, Gemini, Claude, Perplexity and Grok. The separate consumer-surface check has its own questions and conditions. API output is not a substitute for observing the consumer application.

Keep every attempt

We save prompts, answer text, source URLs, engine and model labels, dates and location settings where supported. Failed attempts remain visible. Unsupported geography is labeled unset; a city in a prompt is not evidence of the user's location.

Count different outcomes separately

A mention, an explicit recommendation and a cited domain are different observations. Every rate carries its denominator and condition. A repeated question is not a representative sample of all buyers.

Match claims to evidence

We distinguish observed defects, external evidence, hypotheses and things not measured. A page title correlated with citations is a hypothesis to test. It does not prove that changing a title will cause inclusion.

Measure changes carefully

We record what changed, preserve comparison questions, and disclose changes in models or retrieval methods. A before-and-after difference can have several causes. Controls help only if they have not also been affected by the work.

Keep business results separate

The 90-day recheck repeats five signature questions four times on the same five API engines. Inquiry records and sales follow-up are needed to assess customer acquisition. Citation changes alone cannot establish revenue or return on investment.

Read the audit scope · What an audit should contain · Showing up across engines