practice

Reporting AEO and GEO to an executive

A one-page AEO and GEO report: readiness trend, crawler access, fixes shipped, AI referrals as a range, and the claims to leave out of it.

An executive report on AEO and GEO fits on one page with four sections: the readiness trend for priority pages, the access status for the AI crawlers you allow, the fixes shipped since the last report, and the AI referral and impression figures, given as a range with their limits stated. Everything in the first three sections is measured from your own site and can be re-checked by anyone. The fourth is reported from systems you do not run, and it is labelled that way. The report is credible because it keeps those two kinds of number apart.

This post is the template: what goes in each section, where each number comes from, a worked example, and the claims that should never appear on the page. It belongs to How to audit a site for AEO and GEO, and it is the monthly counterpart to the internal business case, which is where the measures in this report should first have been agreed.

Why one page, and why this order?

One page because an executive reads the first screen and decides whether to read further. This order because it runs from what you control to what you do not. A reader who stops after section two has still read only facts. A reader who reaches section four reaches the soft numbers last, after the context that keeps them in proportion.

The order also protects the work. If the report leads with AI referrals, the work is judged by AI referrals, and a flat month looks like failure even when every priority page became readable. If it leads with readiness, a flat referral month is context, not the headline.

Section 1: Readiness trend

Report readiness as a trend over the same set of priority pages, measured the same way each month. Three numbers are enough:

  • Priority pages with no Critical findings, as a count and a share of the set.
  • Median AEO and GEO score across the set, or whichever readiness scores your crawler produces, with last month’s figure beside each.
  • Templates where every section opens with a text answer, as a count of templates checked.

Keep the page set fixed for a quarter. If pages are added or removed, say so on the report, because a changed denominator moves the figures without any page changing. Keep the crawl settings fixed too: the same scan mode, the same thresholds. A comparison across different settings measures the settings.

Get the figures by hand if you must: the answer-readiness audit works from a crawler export and a spreadsheet, and a pivot table gives the counts. With a crawler that saves history, the comparison is already computed. In GEOBee, comparing a baseline crawl with the current one gives delta chips for AEO, GEO, SEO, Pages, and Critical across the crawl, plus a per-page list of Improved, Regressed, Unchanged, Added, and Removed URLs, and it warns when two crawls are not comparable, such as a quick scan against a deep scan (projects & compare). Crawl, fix, re-crawl covers how to read that comparison honestly.

Section 2: Access status

Access is the one section that can be reported as a binary per crawler. List the AI crawlers your organisation has chosen to allow, and for each one state whether it can reach the priority pages: allowed in robots.txt, and not turned away at the CDN or firewall. Add one line for any crawler your organisation has deliberately blocked, with the date and owner of the decision, so a deliberate block is never read as a fault.

This section is short when things are working, and it should stay on the report anyway. Access regresses silently: a robots.txt deploy, a new bot-management rule, a migrated subdomain. A line that reads “unchanged” every month is the line that catches the month it changes. How to check whether AI crawlers can actually reach you is the method.

Section 3: Fixes shipped

List what changed on the site since the last report, grouped by template rather than by URL. Each line has three parts: the fix, the number of pages it touched, and the date of the crawl that confirmed it. A fictional example:

Product template: section openings moved above images. 7,200 pages. Verified by crawl on 14 November.

Template-level lines are what an executive can follow. A list of four hundred URLs is not a report, and the URLs belong in an appendix or a linked sheet. Include one line for anything that regressed and was fixed, because a report that never shows a regression is not believed.

Section 4: AI referrals and impressions, as a range

This section reports outcome indicators from analytics and Search Console. Each figure is labelled with its source and its limits, and none is presented as caused by section 3.

AI-surface impressions. Search Console’s generative AI performance report counts impressions, defined as “how many times links to your site were shown to a user in a generative AI feature on Google Search.” It covers AI Overviews and AI Mode, excludes Search Labs experiments, and can be grouped by page, country, date, or device. The documentation describes impressions only, not clicks or queries, and notes the report may not appear for a site that “hasn’t received enough impressions in generative AI features” (Search Console Help: generative AI performance report). Clicks from those features are not separated out: Google’s documentation states that traffic from AI features is “included in the overall search traffic in Search Console,” under the Web search type (Google Search Central: AI features and your website). The Search Console generative AI report, read honestly covers how far to trust these figures.

Assistant referrals. Sessions in analytics whose source is an assistant’s domain. Tracking ChatGPT and Perplexity referrals in GA4 covers how to set up the segment and why it undercounts: not every assistant passes a referrer, and some visits arrive as direct traffic.

A manual question log, if you keep one. A small set of questions run on a schedule, recorded as how many runs out of three cited the site. Report it as a range, never as a single result. The same question can return a different answer on the next run, so one run is not a measurement.

Present each as a range or a trend over three months, not a single month-on-month percentage. Small absolute numbers produce large percentages, and a 40% rise from a small base reads as a result when it may be noise. Measuring AI visibility without overstating it covers the full set of collection rules.

What does the finished page look like?

The example below is fictional. The site, the figures, and the dates are invented to show the format.

SectionThis monthLast monthNote
Priority pages with no Critical findings212 of 240168 of 240Same 240 pages since August
Median AEO / GEO, priority pages78 / 7169 / 66Quick scan, default thresholds
Templates with text-first sections4 of 62 of 6Category and location templates outstanding
Allowed AI crawlers reaching priority pagesAll allowedAll allowedTraining crawler blocked by policy, March
Fixes shipped3 template fixes, 41 rewrites1 template fix1 regression found and fixed
AI-surface impressions (Search Console)9,000–11,000 / month, 3-month range—Impressions only; no clicks reported
Assistant referral sessions (analytics)300–420 / month, 3-month range—Undercounts; some arrive as direct

Below the table, two or three sentences: what was fixed, what is next, and what is blocked and by whom. That is the page.

What should never appear on the page?

Some statements are tempting in an executive report and cannot be defended. Leave them out, even when the numbers seem to support them.

  • “Our fixes increased AI citations.” Citation is decided by the engine, the question, and the competing sources. A rise in impressions after a fix is a correlation in one month of one report.
  • “We rank in ChatGPT.” Answers vary by run and by user. There is no rank to report.
  • “Our share of voice is X%.” Unless the method, the question set, and the run count are stated alongside it, the figure cannot be checked.
  • A single blended score. Readiness and outcomes measure different things. Averaging a crawl score with a referral count produces a number that means neither.
  • Forecasts. “At this rate we will reach N referrals by Q2” projects a trend in a figure you do not control.

A useful test for any sentence on the report: could someone else re-derive it from the stated source next month? If not, it is opinion, and it belongs in a conversation rather than on the page.


To produce the readiness trend and the fixes-shipped section from saved crawls and comparisons, download GEOBee. The 7-day trial includes projects, re-crawl, and compare.

On this page