Skip to content

GEO courseModule 10 of 10: Measure and PlanLesson 10.1

Measuring Honestly: The Monthly Report, and What Can’t Be Measured

How to turn your monthly prompt runs, Google's and Bing's AI reports, Analytics and your enquiry sources into one honest page each month, without fooling yourself about what moved.

4 min read

01What it is

Weighing yourself once, after a big lunch, tells you nothing. Weighing yourself at the same time every week, on the same scales, tells you a lot.

GEO measurement done badly is worse than none, because it feels like knowledge. AI answers change from run to run, so the honest unit is never "we're number 2". It's "we were named in 11 of 48 answers".

What to measure

MeasureWhat it means
Mention rateAnswers that name you ÷ answers run. Always show the count: 11 of 48
Citation rateAnswers that link to or show a page from your site ÷ answers run
Share of voiceYour mentions ÷ all agency mentions across the same prompts. Keep your own definition fixed
AccuracyCorrect facts ÷ facts checked (fees, areas, services, hours, contact, team), with an error list
Source shareHow often each website is cited. Feeds your source map (lesson 7.1)
SentimentHow you're described: positive, neutral or negative, judged by you, not a tool
AI visitsSessions in Analytics from AI assistants. Expect an undercount
EnquiriesValuations, viewings and calls by source, including people who say AI sent them
PositionNot recommended. Order changes every run

The free reports from Google and Bing

1Search Console: Generative AI performance report

Free

Impressions of your pages in AI Overviews, AI Mode and Discover's AI features, by page, country, device and date. No clicks and no searches. Rolled out to all sites on 31 August 2026; sites with very few appearances may not see it.

Illustration
Illustration of Search Console's Generative AI performance report showing 3,030 impressions over three months and the pages that appeared most. Search Console's Generative AI performance report: how often your pages appeared in AI Overviews and AI Mode. Impressions only: no clicks, no searches. Example numbers.
Open Search Console: Generative AI performance report (opens in a new tab)

2Bing Webmaster Tools: AI Performance

Free

How often Copilot and Bing's AI answers cite your pages, which pages, and a sample of the searches they ran. It counts citations, not clicks, and not recommendations of you by name.

Open Bing Webmaster Tools: AI Performance (opens in a new tab)

3Google Analytics: AI Assistant channel

Free

Visits from ChatGPT, Gemini, Claude and others (lesson 9.1). Google's own AI Overviews and AI Mode visits stay in Organic Search.

Open Google Analytics: AI Assistant channel (opens in a new tab)

4Crawler logs from your host

Free (ask your host)

Which AI crawlers fetched which pages and whether they got errors. A fetch by ChatGPT-User or Perplexity-User suggests a person's question led to your page. It doesn't show whether you were recommended.

Open Crawler logs from your host (opens in a new tab)

The one-page monthly report

  1. 1Headline in one sentence, with the number of answers behind it.
  2. 2The measures above: this month, last month and baseline, each with its count.
  3. 3Results by audience: sellers, landlords, buyers, tenants.
  4. 4Who else was named, and which sources were cited.
  5. 5Accuracy: errors found, fixed and still open.
  6. 6AI visits, Search Console and Bing figures.
  7. 7Enquiries by source, including self-reported AI.
  8. 8What you changed this month.
  9. 9The three actions for next month.
  10. 10Caveats: sample size, and anything that changed in how you tested.

How not to fool yourself

  • Never judge from one run, one platform or one day.
  • Never reword a prompt to get a better result.
  • Record the conditions every time: platform, logged in or out, location, date.
  • Show the count next to every percentage, and don't compare a month of 30 answers with a month of 150.
  • With under about 100 answers a month, treat moves of less than about 10 points as noise, and look at three months, not one.
  • Don't claim a cause. If you made six changes, you don't know which one mattered.
  • Track accuracy and enquiries, not just visibility. They're what matter.

A worked example (invented numbers)

MeasureBaselineMonth 3Note
Mention rate18 of 144 (12.5%)29 of 144 (20.1%)Up 7.6 points
Citation rate9 of 144 (6.3%)17 of 144 (11.8%)Up 5.5 points
Branded answers with an error3 of 81 of 8Old fee and hours fixed
AI assistant visits14 of 4,10023 of 4,300Undercounts
Enquiries that said AI13Self-reported
Fenbury Lets (fictional) removed a crawler block, fixed its name and phone number everywhere, replied to every review, and published a plain-text fees page between the two runs. With about 144 answers, a 7.6-point rise is within the noise you'd expect even if nothing had changed. The honest verdict: encouraging, not proven. The accuracy fix is the convincing result, because it came from specific fixes to specific pages.

What you can’t measure

  • How many people saw an answer naming you. No AI company provides it, and tools' "prompt volumes" are estimates.
  • Clicks from Google's AI features separately from ordinary search.
  • Visits that lose their label and land in Direct.
  • Someone told your name who then rings or searches, unless they tell you.
  • What one particular person, with their own history and location, was shown.