Running your own AI visibility audit

This is the full method, written so you can run it yourself. It takes a focused day for a small site and it is worth doing before you buy anything in this category, including from us.

Prerequisites: the glossary, GEO basics and entity SEO.

Part 1: Access (30 minutes)

Rule out the boring explanation before investigating the interesting one.

Check robots.txt. Visit yourdomain.com/robots.txt and look for Disallow rules affecting GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot or Google-Extended. A blanket User-agent: * / Disallow: / left over from staging is more common than you would hope.

Check rendering without JavaScript. Disable JS and reload your key pages. If your services, pricing or contact details vanish, many AI crawlers see the same nothing. This single check explains a surprising share of “we’re invisible to AI” cases.

Check response codes. Crawl your sitemap and confirm every URL returns 200. Redirect chains, soft 404s and 5xx errors under crawler load all suppress you.

Check the sitemap is honest. It should list canonical, indexable URLs only, no redirects, no noindex pages, no duplicates and be referenced in robots.txt.

Record anything broken. Fix it before measuring, or you will measure the wrong problem.

Part 2: Prompt testing (3 to 4 hours)

Write 15 to 25 prompts your buyers would actually type, across five groups:

• Category discovery, “Who are the best providers of X?”
• Comparison, “X vs Y for a company of our size”
• Problem-aware, “How do I fix [the problem you solve]?”
• Brand-specific, “What is [your brand]?”, “Is [your brand] good?”
• Segment or local, however your buyers qualify the question

Run each prompt three times per assistant, in a clean logged-out session on each of the platforms your buyers use. Memory and history contaminate results badly, do not run these in the account you have been researching yourself from.

Record for each run: were you named, in what position, which competitors appeared, how you were described, whether you were linked and the date.

Report as rates. “Named in 5 of 12 runs” is data. “ChatGPT mentioned us” is an anecdote.

Part 3: Accuracy review (1 hour)

Go through the brand-specific answers and mark every factual error, wrong services, wrong location, outdated pricing, confusion with a similarly-named company.

For each error, find the source. It is usually one of:

• Your own site says it ambiguously, or no longer says it at all
• An outdated directory listing or old profile
• Genuine confusion with another entity sharing your name

This list is frequently the highest-value output of the whole audit, because these are correctable and the correction compounds.

Part 4: Entity review (1 hour)

Pull up your site, LinkedIn, Google Business Profile and every directory listing you can find. Put the company name, category, description and service area side by side in a spreadsheet.

Mark every inconsistency. Then check your Organization schema: does it exist, does sameAs list only profiles you control, does description match what you say elsewhere?

Reconciling these is dull work with outsized returns.

Part 5: Content review (2 hours)

Take your five most commercially important pages. For each, ask:

• Is the main question answered in the first paragraph, or does the page build to it?
• Do the headings match how people phrase the question, or are they internal labels like “Our Approach”?
• Pick three sentences at random. Could any be quoted verbatim as an answer? If they are all brand atmosphere, that page cannot be cited.
• Are comparisons in tables and processes in numbered lists?
• Is there anything here that exists nowhere else? Original data, a documented method, a real result. If not, a model has no reason to prefer you over a larger competitor saying the same thing.

Part 6: Competitor read (1 hour)

Take the two or three competitors who appeared most in your prompt runs. Look at the pages being cited.

You are looking for specifics: do they publish original research, do they have comparison pages, are they described by independent sources, is their entity data cleaner than yours? Do not copy their content. Identify the category of advantage and decide whether you can build a better version of it.

Part 7: Prioritise (30 minutes)

Score every finding on impact, confidence, effort and risk. Then order the work:

1. Access and technical blockers. Nothing else matters while these are live.
2. Factual errors about your business. Cheap, high value, compounding.
3. Entity reconciliation. Dull, foundational, enables everything after.
4. Content restructuring on commercial pages. Where the traffic already is.
5. New original material. Slowest, most durable advantage.

Resist starting at five. It is the most enjoyable and the least useful while one is unresolved.

Re-run on a schedule

Monthly is right for most businesses. Weekly generates noise you will overreact to. Read trends over three cycles before concluding anything.

Keep every run. The value is in the series, not any single measurement and the baseline you captured before you started changing things is the only thing that will ever let you prove the work mattered.

---

If you would rather have this run for you, with the prompt set handed over so you can verify it, that is our AI visibility audit.