The AI visibility audit checklist for agencies.

An audit is only useful if you can run it again later and compare. Lock the scope before collecting any answers.

• Write down the brand factsLegal and trading name, services, locations, website and official profiles — the facts the engines should get right.
• Name three to five competitorsThe businesses the client actually loses work to, not just the biggest names in the category.
• Choose the answer enginesTypically ChatGPT, Google AI Overviews, Gemini, Perplexity and Microsoft Copilot — agree which matter for this market.
• Build the prompt setRealistic buyer questions across discovery, comparison and decision, sourced from sales calls, enquiries and reviews.
• Record the run conditionsDate, engine, signed-in state and location, so later runs can be compared fairly.

• Run every prompt on every chosen engineSave the full answer text and any cited links, not just a yes or no.
• Mark how the brand appearsRecommended, mentioned, cited as a source, or absent.
• Check how the brand is describedNote anything inaccurate or out of date: services, locations, prices, opening hours.
• Repeat a sample of promptsAnswers vary between runs. Re-running a sample shows which results are stable.

• Compare business details across listingsName, address, phone, hours and categories on the website, map platforms, directories and review sites.
• Look for duplicate or abandoned profilesOld addresses, previous brand names and duplicate map listings all muddy the entity.
• Check the site states who the business isA clear About page: what the business does, for whom, where, and since when.
• Validate Organization or LocalBusiness structured dataFacts in the markup should match the visible page and the listings, with sameAs links to official profiles.

• One clear page per core serviceA single catch-all services page is a weak match for specific questions.
• Pages open with a direct answerWhat it is, who it's for and how it works — before the detail.
• FAQs reflect questions buyers really askCompare against the prompt set. Missing questions are content gaps.
• Claims are backed by visible evidenceCredentials, process, examples of work and reviews — not unsupported superlatives.

• Check robots.txt rules for AI crawlersLook for rules affecting OpenAI's GPTBot (training) and OAI-SearchBot (ChatGPT search), Anthropic's ClaudeBot (training) and Claude-SearchBot (search), and PerplexityBot (search). Google-Extended is a robots.txt token that controls whether Google may use content for Gemini — not a separate crawler.
• Main content is in the page HTMLView the source or fetch the page without JavaScript. Content that only appears after scripts run may not be read.
• Key pages return 200 and aren't noindexedCheck redirects, canonical tags and robots meta on every priority page.
• The XML sitemap lists the priority pagesAnd only pages that should be indexed.
• Pages load quickly on mobileSlow, script-heavy pages are harder for any crawler to fetch and render.

• List the sources the engines citeGroup them: directories, review platforms, publications, forums, competitor sites.
• Check the client's presence on eachMissing, incomplete or out-of-date profiles on frequently cited sources are high-priority fixes.
• Review volume, recency and detailOn the platforms that matter for the category.

• Count mentions per competitorAcross the whole prompt set, per engine.
• Note why winners are chosenWhich sources back them, and which attributes the answers highlight.
• Identify the prompts competitors ownThese are usually the clearest opportunities.

• Keep a dated evidence logEvery prompt, engine, answer and citation — the record that makes the findings credible.
• Prioritise fixes by impact and effortQuick wins such as blocked crawlers or inconsistent listings first; content and citation work as a program.
• Frame outcomes honestlyNo one can guarantee placement in third-party AI answers. Report what changed and how you know.
• Set the date of the next runSame prompts, same engines, same conditions.