How to Plan an AI Search Visibility Audit
An AI-search visibility audit should distinguish observed assistant responses from a review of the website. A crawl or HTML check cannot establish whether a brand was mentioned, cited, or recommended.
Define a reproducible question set
Select questions from documented customer research, sales objections, site search, and support topics. Include non-branded discovery, comparison, qualification, and factual questions. Freeze the wording for the baseline rather than rewriting prompts after seeing results.
- Assign each question an ID and purpose.
- Document region, language, account state, and any enabled search mode.
- Choose a collection date and repeat policy.
- Avoid presenting a convenient sample as representative of all users.
Preserve observations
For each run, retain the exact response, source links shown, timestamp, platform, prompt, and collection conditions. Mark whether the brand appears without treating tone or order as an objective rank unless a written coding rule defines it.
Review the underlying public material
- Fetch important pages and confirm useful text is present in rendered or server HTML.
- Check canonical, robots, status, internal links, titles, and structured data.
- Compare product, organization, location, and policy facts across owned pages.
- Trace inaccurate assistant statements to possible public sources, but label that diagnosis as a hypothesis.
Prioritize by evidence and control
Separate confirmed defects, observed response patterns, and proposed experiments. Give priority to inaccurate first-party facts, blocked or broken pages, and high-value questions. Assign an owner, acceptance test, and review date to each action.
Limits and next step
A sample describes only the recorded runs. Responses may change, and technical readiness is not recommendation evidence. Repeat the same protocol after meaningful changes and report both favorable and unfavorable observations.