Skip to content

What an AI visibility audit actually covers

Five stages, one document, no retainer attached to it. Plenty of clients take the report and do the work themselves, which is a perfectly good outcome.

01

Baseline prompts

We agree a set of 20 to 40 prompts a real buyer might type, in their language rather than yours, covering the ways someone arrives at your category. Each is run across ChatGPT, Perplexity, Google AI Overviews and Claude. We record every business named and every source cited.

You get: A table of who currently owns the answer, and which pages the assistants are reading to decide.

02

Crawler access

We request your pages as GPTBot, ClaudeBot, PerplexityBot, Google-Extended and a few others, and compare what comes back with what a browser gets. This catches robots.txt rules, WAF and bot-fight settings, geo-blocking, and content that only exists after JavaScript runs.

You get: A pass or fail per crawler per template, with the exact rule or setting responsible for any block.

03

Quotability review

We read your key pages the way a model does: looking for sentences that can be extracted and stood behind. Vague claims, facts buried in images or PDFs, prices that exist only on request, and service boundaries that are implied rather than stated all surface here.

You get: A marked-up list of the specific sentences to rewrite, with suggested replacements.

04

Corroboration map

We look for where else your core facts appear: directories, association registers, licensing bodies, supplier lists, news, client sites. Then where your competitors appear that you do not.

You get: A gap list of places worth being, ordered by how often assistants actually cited them in stage one.

05

Structured data and machine-readable pages

Whether your schema markup is present, valid and true. Whether an llms.txt would help. Whether your key facts are in markup as well as prose, and whether the two agree with each other.

You get: Corrected markup ready to paste, and a note on anything currently asserting something false.

What the report looks like

Somewhere between fifteen and thirty pages depending on the size of the site, written to be read by whoever has to do the work rather than by a board.

In it

  • The prompt set and the raw responses, so you can check our reading of them.
  • Every issue found, ordered by how much it is costing you.
  • Rewritten sentences you can paste straight in.
  • Corrected schema markup.
  • A re-test date and the same prompts to run again.

Not in it

  • A visibility score out of 100. There is nothing to calibrate it against.
  • Traffic or revenue projections. We cannot honestly model those yet.
  • Competitor benchmarks presented as fact when they are one snapshot.
  • A recommendation to buy something else from us on every page.

One thing to be clear about

An audit tells you why you are not being named and what to change. It does not change it. If nobody at your end is going to rewrite the pages or chase the directory listings, the report will sit in a drive and nothing will move.

Worth deciding who owns the follow-up before you commission it, even if that person is not us.

Talk about scope