What an AI Visibility Report Should Actually Show

Most AI visibility reports are screenshots with a number attached. A percentage, a coloured dial, a chart that goes up and to the right, and no instruction anywhere about what to do on Monday.

That is not a report. It is a mood.

A report earns its place in a client relationship when it changes a decision. It has to answer four questions: what did the engines say, which sources produced that, what is the gap worth, and what are we doing next. Anything that does not help answer one of those is decoration.

The six things a real report contains

1. The prompt set, stated in full. Not a sample, not a summary. The actual questions, written the way buyers write them, mapped to the pages that carry revenue. Include the comparison and objection phrasings, because that is where decisions happen. If the client cannot read the list and recognise their own customers in it, the rest of the report is measuring the wrong thing.

2. Presence per engine, and per prompt. "Visible in AI" is not a finding. Visible for which prompt, on which engine, in what position in the answer, next to whom. Gemini, ChatGPT, Perplexity and Google's AI Overviews retrieve and cite differently, and averaging them hides the one that matters to that client.

3. The sources actually cited. This is the most valuable page in the document and the one most often missing. List the domains the engine cited for each prompt. That list is the client's real competitive set, and it is regularly different from the list they name in a pitch.

4. The classification of each gap. For every prompt, say which of these happened:

• you were not in the sources at all
• you were in the sources but not cited
• you were cited but described badly
• you were cited correctly

Those are four different problems with four different owners. Collapsing them into "not visible" guarantees the fix will be aimed at the wrong stage.

5. A ranked action queue. Each item with what to change, which page or third party it lives on, who owns it, and roughly what effort it takes. Ranked by commercial value of the prompt, not by ease. A report without an ordered list of next actions is a status update pretending to be a plan.

6. Trend and method, stated plainly. The same prompt set, run the same way, over time. Plus the working: which engines, which market, which dates, how many runs, and what changed in the method between reports. Engines are non-deterministic, so an honest report shows the noise rather than smoothing it away.

"Most agencies automate writing. We automate strategy. That's the difference between 'content at scale' and 'AI fluff at scale.'"

John Russell, Director, Digital Womble

Three things that should not be in it

A single health score. One number invites optimisation of the number. If it goes up because the prompt set got easier, the client has been misled by their own report.

A screenshot collage. Screenshots are evidence, not analysis. A page of them with no classification tells the client something happened and nothing about why.

Competitor percentages with no method. "You have 12 per cent share of voice, your rival has 31" is meaningless without the prompt set, the run count and the engine list. It is also the line most likely to be quoted back at you when the numbers move for innocent reasons.

Why this matters commercially

A report that classifies the gap is the difference between selling a retainer and selling a retainer conversation.

When the client can see, in their own words, that they are in the sources but never in the answer, the next step is obvious and specific: passage work. When they can see they are absent from comparison prompts entirely, the next step is architecture. Both are scoped, both are priced, and neither requires the client to trust a dashboard.

That is also how an AI visibility audit stops being a one-off diagnostic. The audit produces the ranked queue. The queue is the work. The work is the retainer.

What it looks like when it is done well

One page the client can actually read, followed by the detail they can interrogate:

1. What we asked, and why those questions carry revenue
2. What came back, engine by engine
3. Who was cited, and what those sources said
4. Where the gap is, classified
5. What we are doing next, ranked, with owners
6. What changed since last time, with the method stated

Six sections, no dials.

"13 interconnected articles with real quotes, automated in one hour — that's AI content at scale done right."

John Russell, Director, Digital Womble

Reporting is the product

If you are an agency building this line, the report is not the deliverable at the end of the work. It is the thing that defines the work, prices it, and renews it. Get the classification right and the sales conversation becomes arithmetic.

We run this as a white-label GEO service for agencies: the prompt set, the citation evidence, the classified gap and the ranked queue, delivered under your brand. Details are on the white label SEO page, with commercial terms on the pricing page.

Next in this series: the 15-minute GEO pitch and why "we do GEO too" is not an offer.