A real monthly AEO report shows five things: a prompt-by-prompt citation table (which buyer-intent questions you were and weren’t cited on, per AI engine), month-over-month deltas against a fixed baseline, the competitors cited instead of you and where, the sentiment and accuracy of how AI describes your business, and the specific work shipped that month tied to the gaps it targets. If your report shows keyword rankings, traffic graphs, or “AI visibility scores” with no prompt-level detail, you’re getting SEO reporting with new labels — and you have no way to know whether the AEO work is real.
This post is the checklist we wish every business had before hiring anyone (including us). It’s also, frankly, a description of our own monthly deliverable — but the point stands independently: these five sections are what makes AEO work verifiable, and their absence is what makes it vapor.
Section 1: The prompt-by-prompt citation table
The core of the report. A fixed set of 40–60 buyer-intent prompts specific to your trade and service area — “best HVAC in Mesa,” “emergency plumber near Chandler open now,” “who installs tankless water heaters in Gilbert” — each tested against ChatGPT, Claude, Perplexity, Gemini, and Google AI Overview.
For each prompt × engine cell: were you cited (yes/no), at what position if multiple businesses were named, and with what description. This is the ground truth everything else derives from.
What to check: the prompt set should be visible, stable month to month, and recognizable — you should read the prompts and think “yes, that’s what my customers would ask.” If the prompts are hidden, change every month, or read like keyword lists (“HVAC Mesa AZ best 2026”), the table is measuring something other than your buyers.
Section 2: Deltas against a fixed baseline
Citation share this month means little without the trajectory. The report should compare against a Month 0 baseline that never moves: cited on 4 of 52 prompts at baseline, 9 of 52 this month, 7 last month — that shape. Single-month numbers are noisy (models update, retrieval indexes refresh, answers vary run to run), so the honest report shows the trend and says which moves are signal and which are noise.
What to check: ask to see the baseline report. If there wasn’t one — if the engagement started without a documented starting measurement — there is nothing to attribute progress against, and every future claim of improvement is unfalsifiable.
Section 3: Who’s being cited instead of you
Every prompt you’re not cited on, someone else is. The report should name them. Over a few months this becomes the most strategically useful section: you see which competitors dominate which prompt categories, which are gaining, and — critically — where they’re being cited from (a directory, a listicle, a news mention), because that source list is the roadmap for your own corroboration work.
What to check: competitor names should be specific and local, not generic. If the “competitive analysis” is a table of domain-authority scores, that’s an SEO artifact — domain authority is not a thing AI answer engines expose or that predicts citation.
Section 4: Sentiment and accuracy of your descriptions
Being cited wrongly can be worse than not being cited. The report should quote how each engine actually describes your business when it names you — and flag inaccuracies: wrong service area, outdated offerings, a years-old fact presented as current. These errors are fixable (they usually trace to a stale directory listing or an old page contradicting a new one), but only if someone is reading the actual answer text rather than counting mentions.
What to check: the report should contain quoted answer text, not just checkmarks. If nobody at the agency is reading what the AI says about you, nobody is catching the errors.
Section 5: Work shipped, tied to gaps
The report should close the loop: here are the gaps the citation table shows, here is what we shipped this month targeting them, here is when we expect the effect to show up (with honest lag times — directory corrections take weeks to propagate; corroboration work takes months). This is what makes the retainer auditable. You can look back over a quarter and see whether the work tracked the gaps or whether the same deliverables shipped regardless of what the data said.
What to check: the deliverables should change as the data changes. A report where “work completed” reads identically every month describes a content mill, not a practice.
The tells that you’re getting SEO reporting in an AEO wrapper
Five red flags, any one of which is worth a direct question to your agency:
- Keyword rankings as the headline metric. Rankings measure a results page. AEO’s surface — the synthesized answer — has no rankings.
- A proprietary “AI visibility score” with no prompt-level backup. Composite scores aren’t inherently bad, but if you can’t drill into which prompts produced the number, the score is unfalsifiable.
- Traffic as the primary success measure. AI citation often drives calls and branded searches rather than clicks — the answer is the visit. A citation report that leans on sessions is measuring the wrong pipe.
- No engine-level breakdown. ChatGPT, Perplexity, Gemini, and Google AI Overview retrieve differently and move independently. One blended number hides which surface moved and why.
- No mention of what didn’t work. Real citation data is lumpy — some months a model update erases a gain. A report that only ever shows improvement is a report that’s being curated.
Why we publish this
Partly because it’s genuinely useful for anyone evaluating an AEO agency, including against us — every question above is one we’re happy to answer on a sales call. And partly because the discipline is young enough that reporting standards don’t exist yet, and the vacuum is filling with rebadged SEO dashboards. Buyers who know what a citation report should contain make the whole category more honest. That’s good for us, and it’s good for you regardless of who you hire.