Anatomy of a good AI visibility report for clients: sections, charts and what to leave out
The monthly report is the product. The client never sees your hours of analysis or your source corrections: they see one document a month, and from that document they decide whether the service is worth what they pay. With something as new as AI visibility — GEO (Generative Engine Optimisation): measuring and improving how a brand shows up in the answers given by ChatGPT, Gemini, Perplexity and Claude — where the client has no prior reference, the report matters even more.
There's a principle worth taping to the monitor of whoever builds it: the report is read by a manager, not an SEO. It's read by someone who decides renewals between two meetings, who doesn't know what a citation is and who only wants three answers: how am I doing?, better or worse than before?, and what are you doing about it? Anything that doesn't help answer those is surplus.
Here are the six sections, in this order, each with its chart.
Section 1: the executive summary (half a page, written last)
Three or four sentences in plain language: where visibility stands this month, what has changed since last month, the most relevant finding (good or bad) and what's going to be done about it. Not a single bit of jargon. If the manager only reads this — and many months they will only read this — they need to come away with the right picture.
Visual: two or three big figures with a trend arrow (↑ ↓ =). Nothing else. The summary carries no charts; it carries conclusions.
Section 2: how visibility is evolving
The "better or worse than before?" question answered with data: in how many of the monitored prompts the client appears, by week or by month, and how that compares to the previous period.
Chart: a time-series line, one series per AI (or the aggregate, with the per-AI breakdown below). The line is the chart for showing change over time — it lets you read the trend at a glance, which is the only thing the manager wants here. Pair it with a reading sentence: "the week 3 rise lines up with the publication of the corrected listings." A chart without a reading is an exam for the client; with a reading, it's your argument.
Section 3: what the AIs say about the client
The section the manager reads in full, because it's about them. Literal quotes from the answers: how they're described, what's highlighted, what data is given — and whether it's right. The accuracy findings go here: the wrong opening hours, the service no longer offered, the outdated price an AI keeps repeating.
Visual: highlighted block quotes, two or three, with the AI and the date. A real ChatGPT answer talking about their company says more than any metric. The bad stuff goes here too, alongside the correction already under way — not in a footnote.
Section 4: the picture against competitors
In how many prompts each competitor appears versus the client, and in which ones a competitor is mentioned and the client isn't. It's the section that holds the most renewals together, because it turns an abstract figure into a concrete feeling: "your competitor is showing up where you aren't."
Chart: horizontal bars, one per brand, sorted. Bars are the chart for comparing a small number of things — and they let you place last month's alongside this month's to see the movement. Avoid the pie: it shows no change over time and becomes unreadable with small percentages.
Section 5: the work done
A concrete, verifiable list of what was done in the month: data corrected in which sources, content published (with links), listings updated, reviews obtained. No padding: five real things beat fifteen vague ones. This is the section that justifies the invoice in the months when results take their time — which in this channel is several months, the early ones especially.
Visual: none. A list with links. Not everything needs a chart.
Section 6: next steps
Three to five actions for the coming month, tied to what was found: "the AI doesn't cite us in the [service] prompts → we publish X and Y." It closes the report's loop: we measure, we find, we act. And it plants the conversation for the next meeting — which part of the report to talk through live and how, you have in what to show in the AI reporting meeting.
The mistakes that ruin a correct report
Dumping data. The full table of 40 or 75 prompts × 4 AIs isn't a report: it's a punishment. It goes in an appendix for whoever asks, or it doesn't go at all. Every figure you add without a conclusion dilutes the ones that have one — the report doesn't prove you work hard, it proves you think.
Hiding the bad news. The false fact still sitting there, the competitor who has moved in, the lost prompt. The client will end up seeing it — all they have to do is ask ChatGPT — and what they won't forgive isn't the fact: it's that your report didn't tell them. The bad news is always told, in section 3 or 4, glued to its action plan.
Untranslated jargon. "Share of model", "citations", "positive sentiment in transactional prompts". Every term the manager doesn't understand is a point where they disconnect from the document. Translate it all: "out of every 10 times someone asks about your sector, you show up in 3." If you want to use an aggregate metric like share of model, explain it once and carefully — here's how to do it without losing the client.
Data without an opinion. A report that describes but doesn't interpret turns your service into a data subscription, and data subscriptions are easy to cancel. Every section carries at least one "this means that" or "so we're going to" sentence. The interpretation is what the client can't get anywhere else.
The final test before sending: could your client explain this report to their business partner in two minutes, without you in the room? If not, half of it is surplus.
Frequently asked questions
How many pages should it have?
Between 5 and 8 with the six sections. Fewer than 4 looks like little for what you charge; more than 10 won't get read. The appendix with the full table doesn't count — and remember that thickness doesn't communicate value: it communicates that you didn't filter.
Monthly or weekly report?
Measure weekly — the AIs' answers move week to week and you need the fine-grained data. Report monthly: the manager doesn't want four documents a month, they want one good one. If something urgent happens between reports (a serious false fact, a collapse), a short email with the screenshot and the action taken beats bringing the report forward.
Should I include screenshots of the AIs?
Two or three, chosen and annotated, in section 3: a real answer with a date is the most eloquent thing you can show. What doesn't work is the album-report of forty screenshots with no structure — screenshots as illustration, structured data as the skeleton.
What do I do in the month when there's nothing good to report?
Report it: flat trend, the diagnosed reason and the plan adjustment. Sections 5 and 6 hold the document up (work done, next move) and honesty holds the relationship up. If the lack of progress drags on, it's not time to dress up reports — it's time for diagnosis and a conversation.
Building this by hand every month is hours of layout that add no analysis. In Surfeo for agencies each client has their report as a PDF with the weekly evolution across the 4 AIs ready to deliver — you add the interpretation and the next steps, which is your part of the value. What a real one looks like you can see in an AI visibility report in minutes, and the pricing on the pricing page.
