How to report AI search visibility to clients

Below are the eight sections, the five metrics to use, how many answers a number needs before it means anything, and how to connect your work to the results without claiming more than the data shows. You can see all of it filled in on our example client report.

A rank report reads a results page and records a position. It is close to the same for everyone who searches at that moment, so one check is a fair record.

AI answers do not work like that. Ask ChatGPT the same question twice and you can get two different lists of firms. Answers change with the user, the place, the day and the model version. So there is no position to look up. The only honest measure is to ask the same questions many times, on a schedule, and count how often the client is named.

That changes what the report must do. It has to show a share (named in 30% of answers), the number of answers behind it, and the period it covers. Leave out any of the three and the client cannot tell a real change from noise.

This is the order a client reads in: the answer first, then the proof, then the method. It is the same order as our example report.

1. 01Summary. Three sentences: where the client stands, what changed, and what you will do next. Many clients read nothing else.
2. 02AI visibility. The share of answers that named the client, as a line over time, with the number of answers under it. Split by AI tool if the tools differ.
3. 03Competitors. Share of voice against the three to five rivals the client cares about.
4. 04Key questions. Which buyer questions the client is named in, which they are missing from, and who is named instead.
5. 05Recommended actions. Three fixes, in order, each with the evidence that led to it. Three is enough to act on; ten is a list nobody reads.
6. 06This month’s progress. What your team finished, with dates. This is the section that renews the contract.
7. 07Evidence. A few real answers, labelled with the AI tool and date, so the client can see what a buyer sees.
8. 08How we measured. The questions, the AI tools, the period, the number of answers and the region. Short, but always there.

Most confusion in AI reports comes from mixing these up. Define them once, in the report, and use them the same way every month.

This is the section most reports skip, and it is the one that protects you at renewal. A visibility figure is a share of a sample. Small samples swing. The table shows how far a 30% visibility figure could be from the true share, by sample size, at the usual 95% confidence level.

What this means in practice

• A client with 25 questions checked daily on 3 AI tools collects about 2,250 answers a month. On the table above, that is a margin of about ±2 points at the least; the real figure is wider, because the same questions repeat.
• Split that by AI tool, or by single question, and each slice is far smaller. A question asked 30 times in a month on one tool can swing by 15 points or more. Report single questions as “named often” or “rarely named”, not as exact percentages.
• A move of two or three points in a month is usually noise. A move that holds for two or three months, in the same direction, is a change worth reporting.

The client is paying for the work, so the report has to connect the two. The trap is claiming the work caused every rise. A rival’s bad month or an update to the AI tool can move the numbers just as much.

• Mark the work on the timeline. Put each finished task on the visibility chart on the date it went live.
• Compare before and after. Look at the two weeks before a change and the two weeks after, on the questions that change was meant to help.
• Rule out the usual suspects. Did the rivals move the same way? Then the AI tool probably changed, not the client. Did the question list change? Then the trend broke.
• Choose your words. “Visibility on payroll questions rose 10 points in the two weeks after we rebuilt the payroll page” is a fair claim if the data shows it. “Our work raised visibility 10 points” is a claim you cannot prove.

AI answers are half the picture. Put the client’s own Google data next to them, so the report covers search as a whole rather than one channel.

Google’s guide to AI search points site owners to a report in Search Console that shows how their pages appear in Google’s AI features. Check what it shows for each client’s account. It is first-party data from Google, so use it alongside an AI tracker’s numbers, not instead of them: Search Console covers Google only, and it cannot tell you what ChatGPT or Perplexity say.

Expect clicks to fall even when visibility rises. When an AI summary answers the question, people tend to click less. In Pew Research’s 2025 study, people clicked a normal result on 8% of visits when Google showed an AI summary, against 15% when it did not. Tell the client this before it shows up in their analytics, not after.

• A single blended “AI score”. Averaging AI tools with very different sample sizes hides more than it shows. Show each tool, and the total.
• “Number one in ChatGPT”. No such fixed spot exists, and the client will check.
• A screenshot as the headline. One good answer is a nice moment, not a result. Use it as evidence beside the trend.
• Fifty recommendations. Three ranked actions with evidence get done. Fifty get filed.

Embrisk builds this report from each client’s live data and shares it by link, with your agency’s logo and colours from the Growth plan up (beta). Whatever tool you use, keep the same sections every month: clients learn where to look, and a change stands out.