GEO KPIs and AI Search Metrics to Report to Clients

Sort GEO KPIs and AI search metrics by where each number comes from and what it can't show, then build a one page client report with counts over a base and a rule for explaining change.

Angelique Swain · · 7 min read

Cover reading GEO KPIs and AI Search Metrics to Report to Clients, beside a 40% share shown as 23% to 59% on 25 answers and 33% to 47% on 200
On this page

The client call is on Thursday, and you can already guess how it goes. They'll have Search Console open on one screen and your tracker on the other, the numbers won't match, and somebody will ask which one to believe.

Neither is wrong, and that's the heart of a good GEO KPIs report. Each source counts something different: collected answers, visits in analytics or impressions in Search Console. Sort every number by where it comes from, say what it can't show, and the mismatch stops being a surprise.

Let's look at how to lay that out on one page, with a count and its base beside every percentage.

What should an AI search report answer?

An AI search report answers three client questions: is the brand named when buyers ask, is the site linked, and which competitors appear instead.

Rankings and clicks answer none of the three. A page can rank well and still go unnamed in the answer, and a name can appear with no link beside it. Firemark's guide to named, linked and recommended mentions separates those cases, and the report should keep them apart too.

Behind a request for KPIs usually sit two plain questions: "are we in the answers?" and "is it getting better?" Every number in the report either answers one of those or belongs in a working file.

That test cuts most of a long metric list. A dashboard with forty tiles is a report nobody has finished reading. Crawler hits, answer position and sentiment can all help the team, and none of them replaces the three questions.

Which GEO KPIs can you count?

Four GEO KPIs can be counted from collected answers: the share of answers that name the brand, the share that link the site, the competitors named instead, and the sources cited.

Which GEO KPIs can you count?
KPIQuestion it answersRead fromWhat it can't show
Share of answers naming the brandDo buyers' questions return our name?Collected answers, counted over a baseWhy the answer changed
Share of answers linking the siteDoes the answer point readers to us?The same answersWhether anyone clicked
Competitors named insteadWho takes our place?The same answersWhether a rival earned the spot
Sources citedWhich sites shape the answer?The same answersWhich source caused a mention

Each KPI is a count over a base: named in 80 of 200 answers, not "40% visibility" on its own. Firemark's guide to AI share of voice shows the formula and why the denominator changes the number.

A count also needs a range, because a sample moves. NIST's handbook describes a method for a confidence interval around a proportion from the count and the sample size. It treats answers as independent draws, so it covers sampling noise and not the choice of questions.

Firemark's method page shows the arithmetic, labeled as not data from a real brand. Named in 10 of 25 answers gives a 95% range of 23% to 59%, and named in 80 of 200 gives 33% to 47%. The same 40% share comes with a much wider range on the smaller base: 36 points wide at 25 answers against 14 at 200.

Two cards: named in 10 of 25 answers gives a range 36 points wide, 23% to 59%; named in 80 of 200 gives one 14 points wide, 33% to 47%.
Firemark method page sample size arithmetic, 95% range. Not data from a real brand

A single blended score hides all of this. If a client wants one number, write the formula on the report, show the base beside it, and keep the four counts underneath so anyone can trace it back.

What do Search Console and analytics add?

Search Console's Generative AI performance reports give impressions for a site's URLs in Google's AI Overviews and AI Mode, and analytics shows visits whose referrer identifies an AI assistant.

Google announced the Search reports on June 3, 2026. They show impressions within AI Overviews and AI Mode, with pages, countries, devices and dates, and Discover has a dedicated report of its own. Google's help page says the insights reached all websites worldwide on August 31, 2026.

Search Console gives a different number from a collected-answer count. It reports impressions of your URLs in those Google features, so a brand named without a link to your site doesn't register, and ChatGPT or Perplexity answers still need a collected sample. Firemark's guide to tracking Google's AI answers puts the two views side by side.

When the two disagree, check what each one counts before anyone panics. Search Console counts impressions for your URLs. A collected count records whether the answer text names your brand, and an answer can name a brand without a link to its site, so the figures can move apart without either being wrong.

Search Console supplies Google impressions, and analytics supplies visits, so use each for the question it can answer. Neither says what ChatGPT told a buyer, which is the gap the collected counts fill. Add a line under each source naming the question it answers, and the client can see why the numbers differ.

The Generative AI report shows impressions only, and analytics adds visits, including visits from assistants Search Console doesn't cover. Filter the traffic report by referrer and label the result as visits from the sources you filtered. Call it a floor, not a total, because nothing in the data says how many readers saw a brand and never clicked.

What can't you report?

A count of collected answers can't tell the client what readers did next, because it records what the answers say and what people do after is outside the data.

Firemark's method page says the same of its own tracker: it doesn't cover clicks, visits or sales that follow an answer, personalized or logged in answers, or countries and languages outside US English and the markets it adds on request. Four other items belong in the "leave out" list:

  • A score with no written definition. If the formula isn't on the report, the client can't check it.
  • A single run read as a trend. One run is a sample of one, and a line needs two points at the very least.
  • A cause for a change. Counts show that the answers moved, not why.
  • Revenue claimed from visibility alone. Show visits and sales next to the work, and label each as measured or correlated.

Say the limit once, in plain words, where the client will see it and not in a footnote: "These counts describe the questions we track, asked once each per run, from a US location in English. They don't describe every question anyone could ask."

Quote: say the limit once, in plain words, where the client will see it and not in a footnote.

How do you lay out the report?

A one page AI search report fits five blocks: the question set, the counts with their bases, the movement, the cited sources, and what the report can't show.

Five report blocks: question set, counts with bases, movement, cited sources and competitors, and what the report can't show.
Report layout from this article. Call a change only after repeat runs
  1. The question set. List the buyer questions tracked and the engines asked, so the client sees what the counts describe. Firemark's guide to how to measure AI visibility covers how to fix the questions.
  2. Counts with their bases. For each engine, name the answers collected, the answers that name the brand and the answers that link the site.
  3. Movement. Compare with the last comparable run, and call a change only by the rule in the next section.
  4. Cited sources and competitors. Show which sites the answers cited and which competitors appeared where the brand didn't.
  5. What this can't show. Copy the one-sentence limit from above, plus the Search Console and analytics notes if you include them.

A worked row looks like this (arithmetic, not a real brand):

How do you lay out the report?
Answers collectedNamedShare95% range
251040%23% to 59%
2008040%33% to 47%

Firemark sells a tracker, so its CSV export (prompts, engines, competitors and sources) and its 13 months of answers, with counts and verdicts kept for the life of the account, are one way to build this. Any export that keeps the counts and the base works.

The first report has no movement block. State the baseline, the date of the run and the dates of the next two runs, so the client knows when a change can first be called. A report that promises movement before the third comparable run has promised something the method can't deliver.

How do you explain a change?

Explain a change only after repeat runs agree, with the base and range beside it. Firemark's method page calls a movement after three comparable runs and a change of 5 points or more.

That threshold is Firemark's own convention, not a finding from any study. The case against trusting one run comes from SparkToro. In a study published in January 2026, 600 volunteers ran 12 prompts through ChatGPT, Claude and Google's AI a combined 2,961 times.

SparkToro found a chance of less than 1 in 100 that ChatGPT or Google's AI gives the same list of brands in any two responses. Its read was that visibility percentages across many runs are likely decent proxies for how brands appear. Perplexity and Gemini weren't in the study, so read it as a caution about single runs, not as a figure for every engine.

A sentence the client can read: "Named in 80 of 200 answers, range 33% to 47%, the third comparable run at this level." A sentence they shouldn't get: "Visibility is up 3 points." At 200 answers the range is 14 points wide, so three points sits well inside it, and the client will remember the number, not the caveat.

Frequently asked questions

What are GEO KPIs?

GEO KPIs are the counts used to judge whether AI answers name a brand, link its site, name competitors instead and cite sources. They work best as counts over a base with a range beside them. A GEO KPI with no base is a number nobody can check.

How often should you report AI search metrics?

Report on a schedule you can compare, and judge movement over several runs rather than one. Firemark's method page calls a movement only after three comparable runs and a change of 5 points or more. How often that is depends on how often you collect answers.

Can Search Console show AI search visibility?

Search Console shows impressions for your URLs in Google's AI Overviews and AI Mode through its Generative AI performance reports, which Google says reached all websites worldwide on August 31, 2026. It counts impressions of your URLs in those Google features, so it can't tell you whether ChatGPT or Perplexity names your brand.

Is AI share of voice a KPI?

AI share of voice is a KPI when it's a count over a base: the answers that name you divided by the answers collected. It needs the base and a range beside it, or it can mislead. The formula and the choice of denominator decide the number.

Should you report AI traffic from analytics?

Report AI traffic as visits from the referrers you filtered, labeled a floor and not a total. Analytics shows who clicked, which counts of collected answers can't, and it can't show who read an answer and never clicked. Keep it next to the counts and don't merge the two into one score.

Report counts, not scores

A client report is only as good as the base under each number, so the first fix is to put the base next to every percentage. Firemark counts named and linked answers over a base and shows the range, which keeps the report to what the answers said. The demo report shows the layout on sample data, and the method page lists what the counts leave out.

Angelique Swain AI-first SEO Strategist

Written by

Angelique Swain · AI-First SEO Strategist

Angelique Swain is an AI-first SEO strategist with over a decade in organic search across retail, medical and B2B software. Most of her work now is on how AI engines choose which brands to name. She builds fixed prompt sets, re-runs them to see how much the answers vary, and traces what the cited pages have in common. At Firemark she writes about measuring AI visibility and what to do once you can see it.

  • Cover reading How to Track AI Traffic in GA4 and What It Misses, in cream type on a dark Firemark ground

    Measure AI visibility

    How to Track AI Traffic in GA4 and What It Leaves Out

    How to track AI referral traffic in GA4 with one custom AI channel that combines Google's AI Assistant channel and a tested regex, plus what GA4's count leaves out and why.

    · 7 min read

  • Cover reading Unlinked Brand Mentions in AI Answers: What They Mean, in cream type on a dark Firemark ground

    Measure AI visibility

    What Do Unlinked Brand Mentions Mean in AI Answers?

    An unlinked brand mention in an AI answer is an answer that names your brand and cites none of your pages. It happens when the answer cites nothing or cites other sites, and only the second points at pages you can work on.

    · 6 min read

Find out if AI recommends you

Start tracking your brand. New accounts get one free run.