A Weekly AI Visibility Report Template for Agency Clients
A client-ready AI visibility report is one page, five sections: the prompt-set summary (how many buyer-intent prompts checked, on which engines — the methodology line that makes every number reproducible), presence movement (cited/mentioned/absent counts vs last week, with the delta explained or explicitly unexplained), the evidence rows (specific prompts where the client appeared or lost ground, quoted), work shipped this week (dated changes, linked to the approval log), and next week's focus. The discipline that makes it trustworthy: shipped work and observed movement stay in separate sections — the report never claims the first caused the second, and clients learn to trust it precisely because it refuses the claim.
Clients don't ask for AI visibility reports because they understand the channel — they ask because they saw a competitor in a ChatGPT answer and want to know where they stand. The report's job is to convert that anxiety into a legible weekly rhythm: what we measure, where you stand, what moved, what we did. One page. The template below survives contact with real clients because every number on it can be re-derived.
The five sections, with a filled example
| Section | Fields | Filled example (Client B, week 38) |
|---|---|---|
| 1. Methodology line | prompt count, engines, cadence | 20 buyer-intent prompts, 4 engines (ChatGPT, Perplexity, Gemini, Claude), checked Tue AM |
| 2. Presence movement | cited / mentioned / absent, vs last week | Cited 6 (+2) · Mentioned 5 (−1) · Absent 9 (−1) |
| 3. Evidence rows | prompt, engine, what appeared | "best invoicing tool for contractors" — Perplexity now cites /guides/contractor-invoicing (new) |
| 4. Work shipped | dated changes from the approval log | Sept 16: comparison-page refresh approved + shipped (3 URLs) |
| 5. Next focus | one line | Close the 'pricing' prompt gap — brief in review |
Why the sections are separated
Section 2 moves for reasons you don't control — engines resample, models update, competitors publish. Section 4 is fully in your control. A report that fuses them ('our work drove +2 citations') claims an attribution no one can verify, and the first week the numbers dip despite good work, the client rereads every previous claim as spin. Keeping them adjacent-but-separate lets correlation accumulate honestly: after a quarter, the pattern of shipped-then-moved rows IS the case, made by dates instead of adjectives.
The failure example
Screenshot theater: a report built on hand-picked screenshots of good answers — impressive week one, unfalsifiable always, and shattered the week the client runs the prompt themselves and gets a different answer (stochastic engines guarantee they will). The methodology line is the vaccine: a fixed prompt set, a stated cadence, counts the client could re-derive. If your report can't survive the client re-running a prompt, it isn't reporting — it's marketing to your own customer.
Producing it without a Thursday-night ritual
Manually: the prompt-set spreadsheet (our buyer-intent guide) plus fifteen minutes per client per week — genuinely fine at three clients, genuinely not at fifteen. A Fenn report is built from one measurement — how often each of five assistants names the client, who it names instead and which sources it reads — and it can be copied as text, printed or saved as PDF, or exported to CSV. It isn't white-labelled or emailed on a schedule, so the weekly cadence, the shipped-work column and your branding stay in this template. The template stands either way: the sections are the deliverable, the tooling just changes who fills them.
FAQ
Weekly or monthly? — Weekly during active engagements (the cadence is the point — clients stop asking 'anything happening?' when Thursday answers it), monthly for maintenance retainers. Whatever you pick, never skip a bad week; the skipped bad week is the credibility event.
What if nothing moved? — Report it plainly: 'no presence change; engines resampled; work shipped continues below.' Flat weeks reported honestly are what make the up weeks believable. The report's asset is its refusal to perform.
Should the client see the raw prompt set? — Yes, on request and at onboarding — it's their market being measured. Clients who've seen the prompt set trust the counts; clients who haven't eventually ask why these twenty prompts and not others, which is a better conversation to have early.
Put this to work on your own website.
Fenn finds what your customers ask, drafts the articles and site fixes, and measures what ChatGPT, Claude, Gemini, Perplexity and Grok say about you — with every change waiting for your approval.