What should an AI visibility audit actually check?

What to check first

Use these checks to identify the problem before buying work or changing the setup.

• ✓A defined set of buying questions
• ✓Provider, date and retrieval conditions
• ✓Mentions, recommendations and citations recorded separately
• ✓Specific access, content and evidence fixes

Use three kinds of question

Try a branded factual lookup, an unbranded buying question and a comparison with a meaningful constraint. For illustration: “What does this business offer?”, “Who builds websites for small local services?” and “Which option fits a focused £250 website brief?”. These are test examples, not measured high-volume prompts.

Keep a compact observation sheet

For each test record the question, platform, date, location wording, whether retrieval was used, the businesses named, the recommendation wording and source URLs. Keep unknowns marked unknown. Do not quietly change the question until a favourable answer appears.

• ✓Exact question and date
• ✓Provider and visible search use
• ✓Mention / recommendation / citation
• ✓Incorrect or missing facts
• ✓The next verifiable fix

Inspect the page behind the answer

Does it load, describe the right service, state real coverage and support its claims? Can someone inspect the work and find the next action? A cited page can still be a poor sales destination; the audit should cover the customer journey too.

Connect observations to useful work

Prioritise a broken public response, a missing service fact, an unsupported claim or a confusing enquiry route. Keep speculative ranking theories separate from proven defects. Review identifiable referral visits and customer-reported discovery alongside the sample of answers.

Apply this to your business

Tell Graeme what you sell, where customers are and where the journey breaks. The next step is a scoped conversation about ai search visibility and chatgpt discoverability, not an automatic package of unrelated services.