OPTIMIZE · Chapter 6

Measuring AI citations and GEO performance.

GEO measurement is harder than SEO measurement because most of the value is zero-click — the AI engine cites the page, the user reads the answer, no session lands in GA4. The four-metric scorecard below is the smallest set of measurements that gives a founder honest weekly insight without buying a 5-figure enterprise tool.

KEY TAKEAWAYS

GA4 referrals from AI surfaces are systematically undercounted — half land in Direct without a custom channel group.
Prompt-panel citation rate is the leading indicator. Sessions are the lagging confirmation.
The smallest viable measurement stack is a 30-query monthly prompt panel + a GA4 AI channel group + a quarterly citability audit.
Tie GEO to the same downstream events as the rest of marketing — qualified inbound, demo bookings, pipeline created.

The four-metric GEO scorecard

Metric	What it measures	Source	What good looks like
Prompt-test citation rate	Of N priority queries run across ChatGPT, Perplexity, Claude, and Copilot, what percentage cite the brand site.	Manual prompt panel (10 to 30 queries) run monthly, or a tool like Profound, Otterly, or Peec AI.	Track monthly, not weekly. Aim for citation rate trending up quarter over quarter on the queries the business actually cares about.
AI-referral sessions (GA4)	Sessions where the referrer matches a known AI surface — chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, copilot.microsoft.com.	GA4 custom channel group with regex matching on source/medium. Many AI referrers land in Direct without it.	Trending up week over week once the channel group is configured. Absolute numbers stay small for most SMBs in 2026.
Brand-mention share inside answers	When the brand topic comes up in a generative answer, how often the brand is named — cited or otherwise mentioned.	Prompt panel transcripts + brand-mention monitoring (Brand24, Mention, or AI-specific tools).	Climbing on the priority topics. For most SMBs the first goal is being mentioned at all.
Indexed page citability score	Of pages indexed in Google, what percentage have an Article + Person + Organization schema stack, an authoritative TL;DR, and pass the 'one-paragraph quote test'.	Internal audit (run quarterly). One-paragraph quote test: can a model lift any paragraph from the page and have it stand on its own?	100 percent of pillar and money pages within a quarter. Long-tail pages can lag.

Building the prompt panel

A prompt panel is the closest thing GEO has to a rank-tracking tool. The process is manual but fast — one disciplined hour per month produces the data.

List 10 to 30 priority queries. Mix informational ("what is X"), comparison ("X vs Y"), and commercial ("best X for Y") prompts. Anchor on what real buyers ask before contracting.
Run each query in each engine. ChatGPT, Perplexity, Claude, Copilot, and Google AI Mode. Use a fresh session per engine to avoid personalisation bias.
Record cited, mentioned, or absent. Cited means the brand site appears in the citation list. Mentioned means the brand name appears in the answer but is not cited. Absent means neither.
Tally citation rate per engine. Of N queries, what percentage cite the brand on ChatGPT, on Perplexity, on Claude, on Copilot, on AI Mode.
Re-run monthly. Track trend, not absolute numbers. The line moving up is the signal.

The GA4 channel group regex

Set up a custom channel group in GA4 → Admin → Channel groups. The regex below covers the major AI surfaces as of mid-2026. Re-check quarterly — new surfaces appear and old ones change their referer behaviour.

Channel name: AI Search
Match: Source matches regex

^(chatgpt\.com|chat\.openai\.com|perplexity\.ai|www\.perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com|bing\.com/chat)

Best for

Founder-led SMBs where one person reviews marketing data weekly and needs a simple, honest scorecard.
Teams currently treating AI search as invisible — the prompt panel makes the channel measurable.
Sites investing in GEO and needing to defend the spend with non-vanity metrics.

Fails when

GA4 referral sessions are treated as the only metric — most AI value is zero-click.
Citation share is tracked without tying it to downstream pipeline events.
The prompt panel changes every month — trend data needs a stable query set to compare against.

Verify before

GA4 has a custom AI Search channel group live and matching real sessions.
A stable 10 to 30 query prompt panel is documented and run monthly.
GEO citation metrics are reviewed alongside pipeline metrics, not in isolation.

Frequently asked

How do I track AI search referrals in GA4?

Create a custom channel group in GA4 with regex rules matching known AI referrers: chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, copilot.microsoft.com, and bing.com/chat. Without it, most AI traffic lands in Direct because the referrer is dropped or stripped. The custom group recovers the visible portion — true volume is always higher because zero-click answers leave no session at all.

What is a prompt panel and how often should I run it?

A prompt panel is a fixed set of 10 to 30 priority queries run across ChatGPT, Perplexity, Claude, Copilot, and Google AI Mode each month. The output is a citation-rate scorecard: of N queries, how many cite the brand on each engine. The panel surfaces movement faster than GA4 referrals do, because citation share shifts visibly before traffic does.

Do brand-mention tools actually pick up AI citations?

Some do, imperfectly. Profound, Otterly, and Peec AI are GEO-specific tools that monitor AI citation share. Brand24 and Mention pick up brand mentions in transcripts indirectly. For most SMBs in 2026, a 30-query manual prompt panel plus a GA4 AI channel group covers 80 percent of the visibility for $0 in tooling cost.

How long before GEO investment shows up in measurement?

Prompt-panel citation rate can shift inside 30 to 90 days on focused queries — AI engines re-crawl and re-rank citations faster than Google ranks new pages. GA4 referral sessions take longer to climb because much of the AI value is zero-click. Brand-mention share inside answers is the leading indicator most worth watching; sessions are the lagging confirmation.

How do I know whether GEO is paying back commercially?

Tie GEO measurement to the same downstream events as the rest of marketing — qualified inbound, demo bookings, pipeline created. The clearest signal is a step-change in branded search and direct traffic that lines up with citation-share growth: AI engines name the brand, prospects then search the brand directly. If citation share is climbing but pipeline is flat for two quarters, the GEO-cited pages are not the ones a buyer is reading on the way to a purchase.