The difference between a GEO service that gets cancelled after three months and one that runs for years is rarely in the work. It's in the report. A client who sees in two minutes every month what changed, why, and what you're doing now, stays. A client who gets a dashboard with thirty numbers drops off.
What a client wants from a report
A business owner or marketing manager wants four questions answered, in this order: Are we in? More or less than last month? Why? And what are you doing now? Everything in the report has to answer one of those four questions. Anything that doesn't belongs in the appendix or the dashboard, not on the first page.
A second requirement: every number has to be traceable. If the report says 'visibility 29%', the client must be able to click through to the prompts and answers that number comes from. GEO has a trust problem (it feels like magic) and traceability solves it.
The six metrics that matter
1. Visibility (win rate) per AI model. The share of non-branded prompts in which the client is mentioned, separately for ChatGPT, Perplexity, Gemini and optionally Claude. This is the headline number. The difference between models is often the most instructive number in the report: 'on Perplexity 41%, on ChatGPT 12%' immediately says where the work is.
2. Share of voice. The share of all business mentions in the answers that belong to the client, next to the competitors. Visibility says 'how often am I in'; share of voice says 'how big is my slice of the answer'. Show the top 5 including the client as bars.
3. Position. When the client is mentioned, in which place? An AI answer usually names two to five businesses; the first gets most of the attention. Going from an average of 2.8 to 1.9 is a real improvement that visibility doesn't show.
4. Sentiment. The split of positive, neutral and negative across mentions, with the literal sentence as evidence for negatives. Only show it when there's something to report: 'two negative claims, both from the same review source, action started'.
5. The GeoReady score. One number from 0 to 100 that weights visibility, share of voice, position and sentiment. This is the number for leadership: one figure, with an arrow up or down. How the score is calculated is in tracking and metrics.
6. Measured effect per action. For every completed action: in how many prompts the client has been mentioned since, compared with before. 'Listing on heatpumpguide.com: since September 12 mentioned in 9 of 14 prompts, before that in 4.' This number proves the work works, and it's the reason the client is still a client next year.
Order: from number to cause to action
A report that works always has the same three layers. At the top the number (score and visibility per model, with the change versus last month). Below that the cause (who does get mentioned, on which sources, what AI literally said). At the bottom the action (what was done, what it did, what's planned for next month). The client reads top to bottom and after every layer has an answer to one of their four questions.
Write three sentences yourself with every report. The automatic report delivers the numbers; your three sentences deliver the interpretation: 'The drop on Gemini is due to a new competitor listed on two comparison sites; we submitted the listing there this week.' Those three sentences are what the client pays for.
What you leave out
Not on the first page: the number of runs and data points (that's your cost picture, not their result), model versions (gpt-5-mini versus gpt-5 means nothing to a client), the full prompt list (appendix), citation share per URL (dashboard), and any number without a comparison to last month. A number without a change isn't information.
Also resist the temptation to celebrate every rise. AI answers vary per run; a rise of 2 points in a week is noise. Report trends over four weeks or longer and say when something falls within the noise. That costs you a small cheer and earns you credibility the moment there's a real jump.
Week, month and leadership
Three reports for three readers. The weekly report (automatic, email, Friday or Monday) for the marketer: score, changes since last week, signals (new competitor, new source, drop), completed actions, work list. Short and operational. The monthly report (PDF, in your branding) for the client contact: the six metrics with the change versus last month, the actions with their effect, and your three sentences of interpretation. The live view (a link with a password, always current) for leadership: score, visibility per model, competitors, no login. GeoReady produces all three automatically; see reporting and dashboards.
The report as a sales tool
Client A's monthly report is the best pitch for client B. Not by sharing client data, but by showing the format: 'This is what you get every month: this number, these causes, these actions, this effect.' A prospect who sees how concrete GEO becomes no longer needs convincing of the concept; they only need to see their own baseline.
Also keep the effect numbers per industry. After a year you know which sources work in HVAC, which in professional services, and which actions deliver results fastest. That's knowledge no tool gives you and what sets your agency apart. Also read how to measure AI visibility for the background of every metric.
Show visibility per AI model, share of voice against competitors, the GeoReady score as a summary, the completed actions with their effect, and the work list. Leave out raw runs, model versions and prompt details.