AI search optimisation

AI visibility audit

For UK businesses that want to know what ChatGPT, Perplexity, Gemini, Copilot and Google AI Overviews say about them. I sample the answers your buyers would see, check what stops assistants reading your site, and label every finding by how it was measured.

An AI visibility audit records what ChatGPT, Perplexity, Gemini, Microsoft Copilot and Google’s AI Overviews say when your buyers ask about the kind of thing you sell. It shows whether your business is named, how it is described, which competitors appear beside it and which web pages the answers lean on. It is a measurement, not a fix: you come away knowing where you stand and what is worth working on, with every finding labelled by how it was obtained.

The audit is the diagnostic part of my AI search optimisation work, and it looks at AI answers only. If you want your website’s crawling, content and links reviewed for ordinary Google rankings, a full SEO audit with a written action plan is the right commission.

What an AI visibility audit tells you

Business owners often notice AI answers by accident. A new client mentions that ChatGPT suggested you, or a colleague pastes in an answer that lists three competitors and leaves you out, or gives an address you left years ago. One answer on one day proves very little, because assistants write every reply fresh, and the result shifts with the wording, the user’s location and their chat history.

The audit replaces those anecdotes with a dated record that can be repeated. For an agreed set of questions it shows:

  • how often each assistant names you, and the questions where it never does;
  • whether the description is right: your services, where you work, who you work with and your price level;
  • which competitors are named in your place, and how often;
  • which pages and third-party sites the assistants cite, the clearest clue to where their information comes from;
  • whether anything technical stops the assistants’ crawlers from reading your site.

Who this is for, and who it is not for

It suits UK businesses whose buyers research before getting in touch: software and B2B suppliers, accountants, solicitors and other professional practices, private clinics, and online shops in crowded categories. It is also worth doing before you commission generative engine optimisation, from me or anyone else, because it gives you a baseline to judge that work against.

It is less useful for a business that Google barely indexes yet, or a local trade that wins most of its work from its Google Business Profile and word of mouth. There the money does more good on the basics first. And if you already know the assistants describe you wrongly and simply want it put right, there is no need to buy the audit separately; the same measurement becomes the first step of the fix.

What the audit measures

AI visibility is easy to overstate. A single score out of 100 can hide how few prompts were run, on which day and from where. I sort every finding into one of three groups, so you can see how much weight it will bear.

Measured directly

  • Crawler access. Your robots.txt file checked line by line against the crawlers each AI company documents, plus the bot settings on any firewall or CDN in front of the site, which can block AI crawlers without anyone deciding to.
  • What crawlers can read. The raw HTML of your key pages, because text that only appears once JavaScript runs may never reach an AI crawler.
  • Bing indexing. Copilot answers from Bing’s index, so I check how much of your site Bing holds in Bing Webmaster Tools, where you can give me access.
  • Referral visits. Sessions in GA4 referred by chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com. Some visits arrive without a referrer and land as direct traffic, so this is a floor, not a total.
  • Your own facts. Business name, address, services and contact details on your site, Google Business Profile, Companies House record and main directory listings, compared side by side.

Sampled, with the method stated

Everything an assistant says is a sample. I write the prompt set with you, in your buyers’ own words, across four kinds of question: describing a problem, asking for options, comparing named suppliers and checking your business by name. Each prompt is run in each assistant in fresh sessions with memory switched off, more than once, and the log records the date, the assistant, the mode where one is shown and a UK location. The size of the prompt set, the number of runs and the assistants covered are agreed with you before work starts.

From those runs I report how often you were named, where you sat in any list, whether the description was accurate, which competitors appeared and which sources were cited. Google AI Overviews are checked for the same questions typed as ordinary searches from a UK location, since they do not appear for every query and their content changes from week to week.

Not reported

  • A single visibility score. A count from a sample is reported as a count from a sample, with the sample size beside it.
  • Search volumes for prompts. No assistant publishes how often a question is asked, so any such figure would be made up.
  • Traffic lost to AI Overviews. Google’s documentation on AI features says appearances in AI Overviews and AI Mode are counted within the overall web search figures in Search Console, so they cannot be separated out there.
  • Why a model said something. I can show the sources an answer cited. I cannot see inside the model, and I will not pretend to.

How the audit runs

  1. Scope and access. A short call to agree the services, places and competitors that matter. You give read-only access to GA4, Search Console and Bing Webmaster Tools if you use them.
  2. Prompt set. I draft the questions and you correct the wording to match what buyers really ask. You approve the set before sampling starts, so the baseline is one you trust.
  3. Technical checks. Crawler access, what the raw HTML shows, Bing indexing and the comparison of your facts across profiles.
  4. Sampling. Every prompt run in every assistant by the stated method, with each answer saved alongside its date and the sources it cited.
  5. Analysis. Answers coded for presence, accuracy and competitors named. Cited sources are grouped by type: your site, competitors’ sites, directories, review platforms, trade press and forums.
  6. Walk-through. A call to go through the findings and agree what to act on first.

Problems that often show up

  • A firewall or bot-protection setting blocking AI crawlers that nobody chose to block.
  • An assistant repeating an old address, phone number or discontinued service taken from an outdated directory listing.
  • A business the assistants know by name but never suggest for “who is good at X” questions, because no independent page links it to that need.
  • A hallucination: a service, price or accreditation the business never had, filled in where the sources were silent.
  • Competitors cited from comparison articles and trade directories in which the business has never appeared.

What you receive

  • The approved prompt set and the full answer log in a spreadsheet you keep, with dates and every cited URL.
  • A written summary for each assistant: how often you were named, how accurately, and your share of voice against the competitors we agreed.
  • A source map of the sites cited for your questions, marking which of them mention you.
  • A list of factual errors, with the likely source of each and how to get it corrected.
  • Crawler access findings, with the exact robots.txt or firewall changes needed.
  • A prioritised list of next steps, each labelled with the kind of work it needs.

Those labels matter, because the fixes sit in different places. Changing how assistants describe and recommend you is GEO work. Getting cited in Google’s own summaries is AI Overviews optimisation. Restructuring pages so one clear answer can be quoted is answer engine optimisation. You can act on the list yourself, hand it to your agency or ask me to quote.

The fee is quoted in GBP after a first call and agreed in writing before any work starts. If you re-check later, using the same prompts means each round compares like with like.

Next step

Send me your website, the two or three competitors you most often lose work to, and a few questions you think your buyers put to ChatGPT or Perplexity. Book a consultation through my contact page and I will tell you plainly whether an audit is worth it for you, or whether your money would do more elsewhere first.

Frequently asked questions

How is this different from an SEO audit?

An SEO audit examines your website: crawling, indexing, content, links and how it ranks in Google's ordinary results. An AI visibility audit examines the answers assistants give about your market, and only looks at your site where that affects those answers. The two overlap on crawler access and clear business facts, but neither replaces the other.

Can I check what ChatGPT says about my business myself?

Yes, and it is worth doing. Use a temporary or logged-out chat so your history does not colour the answer, ask the same question in a few different wordings, run each more than once and note which sources are cited. The limit is scale and consistency: a handful of checks shows you a pattern, not a baseline you can measure against later.

Should I block AI crawlers to protect my content?

Decide crawler by crawler. Most AI companies run separate crawlers for collecting training data and for fetching pages to answer questions; OpenAI's GPTBot, for example, is its training crawler, not the one ChatGPT search relies on. Blocking a search crawler usually means the assistant cannot cite your pages, so block training crawlers only if that is the trade-off you want.

What if an assistant says something false or damaging about us?

Assistants offer businesses no way to edit their answers, so the practical route is to find the source being repeated and correct it there: a directory listing, an old news story, a review profile. Corrections can take a while to reach the answers, depending on when the assistant next retrieves those pages. If a statement is defamatory or concerns a person's data, speak to a solicitor as well.

How often should AI visibility be re-checked?

It depends on how fast your market moves and whether you are changing anything. After fixes, leave enough time for the changed pages and listings to be crawled again before re-sampling, otherwise you are measuring the old picture. Re-checks should use the same prompts and method as the baseline, or the comparison means nothing. This is my judgement rather than a published standard.

Do you need access to our analytics accounts?

Not for the sampling, which runs entirely in the public assistants. Read-only access to GA4, Search Console and Bing Webmaster Tools adds the referral and indexing evidence, and you can remove it once the audit is finished. My guide to how AI answers choose their sources explains the background if you would like to read it first.

Ready to talk about your project?

A straight answer about what would move the numbers, and a written proposal if we are a fit.