Which KPIs belong in an AEO report, and what should it look like each month?

Last updated October 14, 2026. Written by the Third Coast AI team.

A monthly AEO report should contain five things: the visibility and citation rates by engine over repeated runs, with the run counts shown; the fact-check log and what changed in it; the work shipped against the plan (pages published, corrections made, entity records updated, mentions earned); the leading indicators that predict the rates (crawler access, pages indexed, decision factors proven, cited-page mix); and the business outcomes the work is meant to produce (AI-referred sessions, leads that mention an AI answer, and conversions from those sessions). Leading metrics tell you whether the work is being done and whether the engines can see it; lagging metrics tell you whether it worked, and they move months later.

This page is the report format we use and the reasoning behind it, written so that you can ask any vendor for the same thing or build it yourself from a PromptWatch or Semrush export. The measurement guide covers how the rates are computed; this page covers what to put around them so the report means something to an owner.

Which metrics are leading and which are lagging?

The leading rows are the ones to watch in the first quarter, because the lagging rows will not have moved yet and a report that shows only them will look like failure. If the crawlers can reach the pages, the pages prove the factors, the entity records agree, and the fact-check errors are closing, the visibility rate is very likely to follow, and the honest report in month two says "leading indicators green, lagging indicators unchanged, as expected." The page on how long AEO takes explains the timing in more detail.

What should the monthly report contain?

1. A one-paragraph summary in plain language: what moved, what did not, what was shipped, what is next. An owner should be able to stop here.
2. The rates by engine with the run count behind each, this month against last month and against the baseline, and a note on whether the change is outside the variance. SparkToro and Gumshoe.ai found less than a 1 in 100 chance of the same brand list repeating on a single run, so a report that does not show run counts cannot be trusted on trend.
3. Prompt-level detail for the prompts that moved, with the answer text and citations, so the reader can see the engine's own words.
4. The fact-check log, listing each wrong fact, where the engine got it, what was corrected, and whether the correction has shown up.
5. Work shipped against the plan, meaning pages published with their gate results, corrections made, entity records touched, and mentions earned and where.
6. Leading indicators, meaning crawler access, bot visits, index status and factors proven, with anything red called out first.
7. Business outcomes where measurable: sessions from AI referrers, leads attributed to AI answers in the CRM, conversion rate of those sessions.
8. Next month's plan in a short list, with what you need to supply (a reviewer's time, a subject-matter answer, a profile login).

How do you connect visibility to revenue?

Imperfectly, and it is better to say so than to invent an attribution model. Three measures are available to most companies. Analytics can isolate sessions whose referrer is chatgpt.com, perplexity.ai, copilot.microsoft.com or gemini.google.com; Google's AI Overviews and AI Mode are not separated out, since Google's documentation on AI features says they are folded into Web search traffic in Search Console. A "how did you hear about us" field on forms and in the CRM, with an AI option, catches the buyers who do not click but do remember. And conversion rate of the AI-referred sessions is worth reporting on its own, because Semrush's AI search traffic study found those visitors convert at about 4.4 times the rate of classic organic, and Pew Research measured users clicking a link in only 8% of visits with an AI summary against 15% without, so fewer visits at a higher conversion rate is the expected shape and a report that looks only at traffic will read it as decline.

Should the report include a single score?

We would avoid it, or publish it only alongside the rates it summarizes. A single visibility score across engines hides the fact that only about 11% of cited domains overlap between ChatGPT and Perplexity, and a weighted blend invites arguments about the weights. The better single line is the one in the summary paragraph: "named in 38 of 120 prompts in ChatGPT across four runs, up from 22 at baseline, with the gain in the service pages we rewrote in August." If the board wants a chart, chart the rate per engine over time with the run count in the footnote, and keep share of voice on a second chart against named competitors.

Which mistakes make a report useless?

• Reporting a position or rank in an AI answer, which the engines do not produce consistently.
• Reporting a rate without the run count, so nobody can judge whether the change is real.
• Blending engines into one score.
• Reporting traffic alone and calling a drop a failure, when Pew measured click-through falling from 15% to 8% with an AI summary present and the recommendation itself is the outcome.
• Leaving out the fact-check log, which is where most of the early wins are.
• Leaving out what was shipped, which turns the report into a weather forecast.
• Changing the prompt set between months without versioning, which breaks the trend.

A report built this way is short, and the first one is the hardest because the baseline has to be established with enough runs to be stable. After that it takes a few hours a month from a tool export, and it is the document we would want to read if we were the buyer. The question list includes the questions to ask a vendor about their version, and the first 90 days page shows how the report evolves over an engagement's first quarter.

Frequently asked questions