The AI visibility audit checklist we run in week one

Every engagement we take on starts with the same audit. It isn't glamorous. It's a checklist, and its job is to answer one question before anyone writes a word of new content: why are you showing up in AI answers today, or why aren't you?

Here's what's on it, in the order we run it.

Prompt baseline

• Build a prompt set from real buying questions, split by persona and funnel stage
• Include branded prompts ("is [brand] good for...") as well as unbranded ones
• Add location variants if you sell in more than one country or city
• Run the set across ChatGPT, Gemini, Perplexity, Claude, and Google AI Overviews
• Record which brands each answer names, in what order, and how consistently across runs

The output is a baseline. For each prompt: are you named, and by which engines? Most teams have never seen this laid out in one place, and it tends to be the page of the audit that gets forwarded around.

Competitor share of voice

• List the brands that appear most often across the whole prompt set
• Note which competitors you expected and which ones surprise you
• Compare how each brand is described, and whether competitors get clearer reasons attached

The surprise competitors deserve a close look. They can be smaller brands that happen to be well covered on the sources the engines read, which tells you something about where to focus.

Citation sources

• For each prompt where you're missing, list the pages and domains the answer cited
• Sort them into your own domain, competitors, publishers, review sites, and community platforms
• Flag domains that show up across many prompts, since those shape your whole category
• Look at Reddit and YouTube on their own, because they behave differently from publisher sites

Crawler access

• Review robots.txt rules for GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot, and Google-Extended
• Check CDN and firewall bot rules, which can block crawlers that robots.txt allows
• Pull crawler logs to see which AI bots visit, how often, and which pages they fetch
• Look for error codes and redirect chains in those logs
• Confirm core content is present in the raw HTML and not only after JavaScript runs

A note on Google-Extended: it's a robots.txt token that controls whether Google may use your content for Gemini apps and Vertex AI. Google says it doesn't affect inclusion in Google Search, and AI Overviews are part of Search. Blocking it won't remove you from AI Overviews, and allowing it won't get you in.

Content gaps

• Map each missing prompt to the page on your site that should answer it
• Mark the prompts that have no matching page at all
• For pages that exist, check whether they answer the question directly and state facts a model can quote
• Review comparison and alternatives pages, which commercial prompts lean on heavily

AI-referred traffic

• Check whether analytics separates visits from chatgpt.com, perplexity.ai, gemini.google.com, and similar referrers
• Find out whether those visits convert, and on which landing pages
• Note the blind spots, such as AI visits that arrive without a referrer and land in direct

What you get at the end

A ranked list of fixes. Each one is tied to evidence from the checklist: which prompts it affects, which pages or outside sources are involved, and whether the cause is access, content, or offsite coverage. Ranking them by effort against the number of prompts they touch usually makes the first month's plan obvious.

The tooling behind it

We run this audit in Promptwatch, and it's the reason the audit takes a week rather than a month. Prompt tracking with personas and country, state, and city targeting builds the baseline. Competitive benchmarking and share of voice cover the competitor section. Citation analytics, including the Reddit and YouTube views, does the source mapping. Agent Analytics pulls AI crawler logs through integrations with Cloudflare, Vercel, Fastly, and other CDNs. Content gap analysis lines missing prompts up against your pages, and visitor analytics answers the traffic questions with conversions attached.

Keeping it all in one workspace also stops the audit from going stale. The same prompts keep running after week one, so the baseline becomes the first data point of the program instead of a PDF nobody opens again.