The AI visibility audit

How the test is run

We write the questions a real customer would ask — not brand-name searches, which almost always return something, but the open questions people actually type, such as asking for a recommendation in your town or for help with a specific problem. We run them across ChatGPT, Claude, Gemini, Perplexity, Copilot and Google’s AI Overviews, and record the answers verbatim with the date. Then we work backwards from those answers to find what is causing them. A person does this. It is not a script scraping your homepage and colouring in a gauge.

Choosing the prompts

This is the part that decides whether an audit is useful. Brand-name prompts flatter you. Prompts written in agency language — "leading provider of" — do not match how anyone speaks. We use the phrasing customers actually use, including the unglamorous version of your service, and we include the questions asked at the research stage rather than only the ready-to-buy ones. Where you have a specialism, we test it, because that is usually where a small firm can genuinely win.

Why six platforms and not one

Because they disagree, and the disagreement is the diagnosis. A business named by Perplexity but ignored by ChatGPT usually has a live-web presence that is fine and a broader footprint that is thin. A business named by Gemini and nowhere else is often carried entirely by its Google Business Profile. Averaging six platforms into one score hides exactly the information you need, so we report each one separately.

The ten checks, in more detail

Crawler access looks at robots.txt and any blocking at the server or CDN. Assistant answers is the live test described above. Entity clarity compares your name, address, phone number, categories and description across every source we can find. Structured data checks what markup exists, whether it validates, and whether it matches the visible page. Answer readiness examines whether your key pages state their answer early and cleanly. Local presence covers the Google Business Profile in full. Reputation looks at review volume, recency, distribution and whether you reply. Off-site footprint maps who corroborates you. Technical health covers rendering, speed and status codes. Competitive picture records who is being named in your place and what they have that you do not.

What the report contains

A score for each of the ten areas with the evidence attached, the assistants’ answers quoted directly, the competitors being named in your place, and a prioritised list of fixes ordered by likely impact rather than by ease. Each fix says what to change, where, and roughly what it should achieve. Where something cannot be tested reliably, the report says so rather than filling the gap with a guess.

How to read your score

The score exists to prioritise, not to compare. It has no meaning outside our own report — it is not an industry standard, nobody else calculates it the same way, and a competitor’s score from a different tool is not comparable. A low score on crawler access is a five-minute fix. A low score on off-site footprint is six months of work. Read the sections, not the total, and start with whatever is cheap and broken.

Findings we see most often

Four recur constantly. A Google Business Profile in a category that does not match what the business actually sells. A homepage that never plainly states the service and the location in the same sentence. Structured data that was added once, by a plugin, and now describes a version of the business that no longer exists. And a trading name that appears nowhere in the official record, so the model cannot connect the brand to the company. None is glamorous. All are fixable.

What it does not do

It does not guarantee an outcome, predict a timescale we cannot support, or produce a score that means anything outside our own report. It does not measure traffic from assistants, because that data does not reliably exist. Model behaviour shifts, and an audit is a snapshot — which is why we date every report and re-test rather than treating one run as the truth.

What happens afterwards

Nothing, unless you want it to. Plenty of businesses take the report and act on it themselves, and the fixes are written so that is genuinely possible — specific enough for a developer, plain enough for an owner. If you would rather we did the work, we will scope it and price it. There is no pressure attached, partly on principle and partly because a report that only makes sense if we implement it is not a report, it is a sales document.

Who reads the report inside your business

We write one document for two readers, because it usually gets forwarded. The opening summary is for whoever signs things off — what is wrong, what it is costing, what to do first, in plain English with no vocabulary you would have to look up. The detail underneath is for whoever implements: file locations, specific markup, the exact wording to change, and how to check it worked. If a finding cannot be explained clearly to both, we take it as a sign we do not understand it well enough yet.

Why we run it by hand

Automated AI-visibility tools mostly fetch your homepage, look for a few tags, and produce a gauge. That is useful as a first look and we offer one for exactly that reason. What it cannot do is ask an assistant a question and read the answer, notice that the competitor being recommended has three accreditations you also hold but never mention, or spot that your trading name appears nowhere in any official record. Those findings come from a person reading, and they are consistently the ones that change something.

Confidentiality

Your report is yours. We do not publish client names or audit findings as case-study material without written permission, and where we describe patterns on this site they are generalised across many businesses rather than drawn from one. If you want an anonymous audit before deciding whether to talk to us at all, say so and we will keep it that way.

The free checker versus this audit

There are two things on this site and they are not the same. The instant checker runs in seconds, tests the technical signals on your own site, and is useful as a first look. This audit is done by hand over a couple of days, includes the actual assistant answers and the competitor names, covers the off-site picture the checker cannot see, and goes considerably deeper. Start with the checker if you want a number today. Ask for the audit if you want to know what is actually happening.