An AI visibility report clients and executives actually read
The deliverable that retains clients is a report a VP reads in 90 seconds. Here’s the structure that works — every number computable from stored monitor results.
Page 1: the three lines
Three sparkline charts, week over week:
- Mention rate — share of monitored prompts naming the brand
- Citation share — share of all
sources[]citations going to the brand’s domains - Answer rank — average position in
entities[]on recommendation prompts
Each line split by engine (ChatGPT, Perplexity, Gemini) — per-engine divergence is itself the story: “Gemini cites us; ChatGPT doesn’t” is an actionable finding, not a data problem.
Page 2: what changed
Transitions only — nobody reads stable numbers:
- Prompts where the brand entered or left the answer this month
- Citations won or lost to specific competitors, with the competing URL
- New fan-out queries (
searchQueries[]) the engines started running — new searches mean new content targets
Page 3: the action list
Each finding gets a verb: “publish a page targeting this fan-out query”, “update the cited page — it’s outdated”, “this Reddit thread drives the citation — engage it.” AI-visibility reporting without actions is astrology.
What to leave out
Screenshots of single answers (cherry-pickable, unverifiable), blended scores that hide per-engine splits, and anything without a time axis. The credibility of the report is that every number traces to a stored API response — keep the raw JSON behind it for the client who asks.
The share-of-voice framework defines the metrics; the agency service post covers packaging it.