Generative Engine Optimisation · Measurement

How to Measure Your Generative Engine Optimisation Performance

How to tell whether your GEO work is paying off when the old metrics no longer apply. A plain guide to the handful of measures that matter in AI search, what share of model means plus a simple monthly method any business can run without expensive tools.

Updated: May 2026
Written by: Andrew Odgers, MD
Reading time: 8 minutes
The short answer

You measure GEO performance by tracking how often and how accurately AI engines name your business for the questions your customers ask. Clicks plus rankings no longer tell the whole story, because so many AI answers end without a click. Instead you watch a small set of measures: your share of model, how often you are cited or mentioned, the sentiment of those mentions plus any traffic AI tools send you. Test a fixed set of prompts across ChatGPT, Gemini plus Perplexity each month, then watch the trend rather than any single number.

Why the old metrics fall short

Clicks and rankings no longer tell the story

If most AI answers never produce a click, then counting visits captures less than half of what is happening. A business can be named in the answer a customer reads plus get no visit at all, so traffic alone hides the real picture.

The shift is from position to presence

In traditional SEO you measured position: where you ranked. In AI search you measure presence: whether you appear in the answer at all plus how you are described. That is a different question, so it needs different measures.

Getting those measures right is part of the reporting we build into our Generative Engine Optimisation Agency service, because work you cannot measure is work you cannot improve.

The measures that matter

Five GEO metrics worth tracking

Five measures cover most of what a business needs. None requires expensive software to start.

MetricWhat it measuresHow to track it
Share of modelHow often you appear across your tracked promptsTest a fixed prompt set monthly plus count appearances
Citation rateHow often you are used as a cited sourceRecord answers that cite you, with or without a link
Mention rateHow often your brand is named in the answerNote appearances even when no link is given
SentimentHow positively the AI describes youScore each mention positive, neutral or negative
AI referral trafficVisits arriving from AI toolsSet up a GA4 channel for chatgpt.com plus the rest

The first three measure visibility, sentiment measures quality plus referral traffic measures the spillover. Together they show whether AI search sees you, trusts you plus sends you anyone.

The headline metric

Share of model, in plain terms

What share of model is

Share of model is the AI version of market share. You define a set of questions your customers actually ask, then measure how often your business appears in the answers. If you appear in eight of ten tracked prompts, your share of model is 80 per cent. It is the single clearest measure of whether AI search sees you.

The trend beats the absolute number

A snapshot is less useful than a direction. Three consecutive months of rising share of model tells you the work is landing. A flat or falling number after months of effort tells you something needs to change. Watch the trend, not just today’s figure.

What it looks like in practice

A worked share of model example

Test the same 20 customer questions in each engine, then record how often you are named. The result is your share of model per engine.

Appears in ChatGPT (14 of 20 prompts)70%
Appears in Gemini (11 of 20 prompts)55%
Appears in Perplexity (9 of 20 prompts)45%

Read across the engines, not just one. This example business is strong in ChatGPT yet has room to grow in Perplexity, which points to where the next work should go.

Illustrative example based on a 20-prompt monthly test across three engines.

A method you can run yourself

Pick the prompts, then repeat monthly

Choose around 20 buyer-intent prompts

List the real questions a customer would ask before choosing a business like yours, phrased naturally. Around twenty is enough to start. These are your golden prompts, plus they matter more than any keyword list because they mirror how people actually ask an AI.

Test them across the engines each month

Run the same prompts through ChatGPT, Gemini plus Perplexity once a month plus record whether you are named, cited or ignored, plus whether the details are right. It takes under an hour plus needs no paid tools to begin.

Watch the trend, then connect it to leads

Track your share of model over time, then tie it to enquiries. Adding a simple how did you hear about us question to your contact form is a cheap way to see whether rising AI visibility is turning into real business.

Measure what actually matters

Want to know if your AI visibility is improving?

Our Generative Engine Optimisation Agency service tracks your share of model, citations plus sentiment across ChatGPT, Gemini plus Perplexity, then reports the trend in plain terms. See exactly what is included plus how we connect AI visibility to real enquiries.

Part of our guide

Generative Engine Optimisation Guides

This article sits inside our complete GEO hub: a connected set of guides covering how AI search works, how each engine chooses businesses plus what a GEO strategy should include.

Visit the hub

Measurement makes sense alongside the work it measures, which is why our Generative Engine Optimisation Guides hub sets them together. It indexes every question a business owner tends to ask before, during plus after starting GEO, from how each engine picks businesses through to cost, timescales plus what a proper service should include. Working through it in order is the quickest way to see how measurement fits the plan.

Keep reading

Where to go from here

To build on this, these reads help. GEO Results sets out what good performance actually looks like. GEO Strategy shows the work the measurement is checking. AI Search and Website Traffic explains why clicks alone no longer capture your visibility.

Frequently asked

Measuring GEO questions

How do I measure GEO performance?
You track how often plus how accurately AI engines name your business for the questions your customers ask. Watch a small set of measures: your share of model, your citation plus mention rates, the sentiment of those mentions plus any traffic AI tools send you. Test a fixed set of prompts across ChatGPT, Gemini plus Perplexity each month, then follow the trend rather than any single number. Clicks plus rankings alone no longer capture the picture.
What is share of model?
Share of model is the AI version of market share. You define a set of questions your customers actually ask, then measure how often your business appears in the answers across the engines. If you appear in eight of ten tracked prompts, your share of model is 80 per cent. It is the clearest single measure of whether AI search sees you, plus the trend over time matters more than any one reading.
Which GEO metrics matter most?
Five cover most needs. Share of model shows how often you appear. Citation rate shows how often you are used as a source. Mention rate shows how often your brand is named, even without a link. Sentiment shows how positively you are described. AI referral traffic shows the visits AI tools send you. The first three measure visibility, sentiment measures quality plus traffic measures the spillover into your site.
Do I need expensive tools to measure GEO?
No. You can start with a free manual method: pick around twenty buyer-intent prompts, test them across ChatGPT, Gemini plus Perplexity once a month plus record whether you are named, cited or ignored. It takes under an hour. Paid tools add scale plus daily tracking once you are monitoring fifty or more prompts. They are not needed to begin measuring meaningfully.
Why are clicks and rankings not enough for GEO?
Because so many AI answers end without a click. A business can be named in the answer a customer reads plus receive no visit at all, so counting visits misses most of what is happening. Rankings tell you where you sit in the ordinary results, not whether you appear in the AI answer above them. GEO measurement shifts from position to presence: whether you show up in the answer plus how you are described.
How often should I measure GEO performance?
Monthly is a sensible rhythm for most businesses. AI answers can shift from one run to the next, so a single test is noisy. A monthly reading across a fixed prompt set smooths that out plus reveals the trend. Review the direction each month plus take a fuller look quarterly, connecting your share of model to actual enquiries so you know the visibility is turning into business.