An AI visibility audit checks how accurately and often AI assistants find, describe, cite and recommend a business for questions its buyers actually ask. A useful audit records a repeatable prompt set, the competing businesses and sources in each answer, the gaps in the business’s own information, and a prioritised plan. Crawler requests alone do not show that an assistant recommended the business to a customer.
Start with buyer questions, not a generic score
The first job is to define the business, its customers, service area and the decisions buyers are trying to make. An audit for a local service firm should not use the same questions as one for a national software provider. The prompt set should reflect genuine discovery, comparison and provider-selection questions.
For each prompt, record the exact wording, location assumptions, assistant, date and whether web search was available. Run important prompts more than once in fresh sessions. Answers can change between runs, so a single mention or omission is an observation, not a reliable trend.
- Discovery: which businesses can help with this problem?
- Comparison: which providers fit a specific budget, location or requirement?
- Evaluation: what are the strengths, limitations and evidence for each option?
- Action: how would a buyer enquire, book or purchase?
Measure more than mentions
A mention is not the same as a recommendation. The audit should distinguish being named in passing, appearing in a shortlist, being recommended for a particular need and being cited as a source. It should also check whether the assistant gets essential facts right.
Record the competitors shown alongside the business and the pages or third-party sources the assistant cites. This helps explain whether the gap is discoverability, unclear positioning, thin evidence or a stronger competitor rather than treating every missing mention as a technical SEO problem.
- Mention and recommendation rate across repeated prompt tests
- Accuracy of services, customers, locations, pricing and next steps
- Citations and other sources that support the answer
- Competitor presence and the reasons given for choosing them
- Relevant visits and enquiries, where tracking can identify them
Check the evidence assistants can find
The audit should inspect the pages an assistant or search crawler can reach, but not stop at crawlability. A clear service page should say what the business does, who it is for, where it operates, why it is credible and what to do next. Important claims should be supported by examples or independent evidence where possible.
Also review relevant listings, reviews, publications and partner pages. If those sources contradict the site, the problem may be inconsistent evidence rather than missing keywords. A crawler fetching a page proves access to that URL; it does not prove the page was used in an answer or shown to a buyer.
What should the report deliver?
A useful report lets someone reproduce the baseline and act on the findings. It should show the prompt panel and raw observations, a concise summary of where the business appears, the competing sources, and a short list of changes ranked by likely impact and effort.
Each recommendation should name the affected buyer question and the evidence behind it. For example, “clarify service area on the main service page because assistants repeatedly assume nationwide coverage” is more useful than “add more AI-friendly content.” Separate quick fixes from changes that need customer proof, partnerships or new case studies.
- Prompt set, test dates, assistants and repeat counts
- Mention, recommendation, citation and accuracy findings
- Competitor and source map with links to supporting evidence
- Prioritised fixes with owners and measurable hypotheses
- A retest plan using the same prompts and unchanged controls
How do you judge whether it worked?
Take a baseline before publishing changes. Retest the affected prompts after the relevant pages or independent sources have been discovered, while keeping some prompts unchanged as controls. Look for a consistent directional improvement across repeated runs, not a single favourable screenshot.
Keep outcome measures separate. AI crawler requests show discovery; assistant mentions and citations show answer visibility; referral visits and qualified enquiries show possible business value. None of those measures alone proves that an audit produced revenue, but together they make the work testable.
Questions about what does an ai visibility audit actually measure?
Is an AI visibility audit the same as an SEO audit?
No. They share checks for accessible, useful pages, but an AI visibility audit also tests actual assistant answers, recommendations, citations, accuracy and competing sources for buyer questions.
Can an audit tell me whether AI agents visited my website?
It can identify some self-declared or verified crawlers in server traffic. That is different from showing that a person used an AI assistant to view or choose your business. Referral and answer testing provide additional, imperfect evidence.
How many prompts should an audit test?
Enough to cover the business’s main buyer intents without padding the panel with irrelevant variations. The exact number matters less than clear selection, repeated runs and a stable set for retesting.
Can an audit guarantee more AI recommendations?
No. Assistants control their own answers and may change their sources or models. The audit should identify evidence-backed improvements and a way to test their effect, not promise a fixed position.