How to Measure GEO Campaign Success Without Vanity Metrics
How to measure GEO: count mentions, recommendations and citations separately, check accuracy, then link them to branded search and enquiries.
Quick answer
There is no single GEO number worth trusting. Run the same 20–40 buyer questions every month, record for each answer whether you are mentioned, recommended or cited and whether the description is accurate, and compare the result with your own first run rather than an industry benchmark. Six measures or fewer, reviewed monthly, is enough for most businesses.
To measure GEO, track three things separately: whether AI answers mention your brand, whether they recommend it, and whether they cite your pages as a source.
Then connect them to business signals you can already measure, such as branded search and enquiries.
Adding everything into one 'AI visibility score' hides more than it shows.
GEO, generative engine optimisation, does not behave like a paid campaign.
AI tools do not give you impression reports for your brand the way an ad platform does, and the same question can get a different answer on another day or for another user.
So the measurement has to be sampled, repeated and honest about its limits.
This article is the measurement part of our GEO series.
For the groundwork see what GEO is; for the actions see how to appear in ChatGPT answers.
The same logic of influence and assisted results underpins our guide to attribution models.
Start With the Decision GEO Should Influence
Before any dashboard, define the decision: are you building authority in your category, supporting long sales cycles, or helping buyers who are comparing options?
Each goal points to different early signals.
If you skip this step, you will optimise for screenshots of AI answers instead of for the health of your sales pipeline.
Mention, Recommendation, Citation: Why Count Them Separately?
These are three different outcomes, and they move for different reasons.
Report them side by side and never add them up.
More mentions without recommendations can mean you are known but not preferred.
Citations without mentions can mean a page is useful as a source while the brand behind it stays invisible.
The table shows what each measure means and how to check it.
| Measure | What it means | How to check it |
|---|---|---|
| Mention | The answer names your brand anywhere, even in a list or in passing. | Run your fixed question list and record whether the brand appears. |
| Recommendation | The answer suggests your business as a good choice for the question asked. | Record the wording: one option among several, the first suggestion, or named with a caveat. |
| Citation | The answer links to or names one of your pages as a source. | Record which URL is cited. A citation can appear without a recommendation. |
How to Build a Question Sample You Can Repeat
Use the same list every time, so a change means something.
Answers vary from run to run, so one result proves little: look at the share across the whole list and across repeated runs.
Monitoring tools can automate the runs.
They do not remove the need to read the answers.
- Write 20–40 questions your buyers actually ask, in their words, including comparisons and 'best provider in my city' questions.
- Run them in the AI tools your buyers use, for example ChatGPT, Perplexity and Google's AI Overviews, the same way each time: same wording, a fresh session, the same language and country where you can.
- For each answer, record mention, recommendation and citation, and whether the description of your business is accurate.
- Repeat on a fixed rhythm (monthly is usually enough) and compare against your own first run rather than an industry benchmark.
Is the Answer Right About You?
Being mentioned inaccurately can be worse than not being mentioned.
Check whether answers get your services, area, prices or audience wrong, and which sources they seem to rely on.
These checks should lead to fixes: unclear definitions on your own pages, missing evidence, weak differentiation, or facts about the business that differ between your site and other listings.
- Topic coverage: which questions should you be part of, and are you?
- Sources: which of your pages get cited, and which never do?
- Accuracy: does the answer misstate your offer, area or prices?
Business Signals: Traffic, Assisted Enquiries and Branded Search
Some visits from AI assistants arrive as referrals from the assistant's domain; others arrive without a referrer and are counted as direct traffic.
Treat referral numbers as a floor, not the full picture.
In Google, traffic from AI Overviews and AI Mode is counted in the Search Console Performance report under the Web search type, together with other web results, so it is not reported as a separate line.
GEO often works indirectly, so add assisted conversions, an AI option in your 'how did you hear about us' field, and notes from sales calls.
A rising level of branded search can reflect growing awareness, although seasonality and offline campaigns move it too.
| Signal | What it tells you | Watch out for |
|---|---|---|
| Branded search volume | Growing awareness of the brand | Seasonality and offline campaigns |
| Referrals from AI tools and niche publications | Discovery through answers and citations | Small numbers and missing referrers |
| Pages per visit on landing pages | Whether the message fits the visitor | Short visits that still convert |
Reporting Cadence
Weekly: what was shipped and published.
Monthly: the question sample, accuracy and traffic quality.
Quarterly: the link to the sales pipeline and a reset of priorities.
Avoid daily panic.
GEO results improve when your pages and reputation improve, not when a dashboard refreshes.
Define the decision
Pick six measures at most
Review monthly
Anti-Patterns in GEO Reporting
Scorecards that only count mentions per week encourage noisy tactics.
Dashboards without a link to your customer records overclaim revenue.
Treat each measure as a driver of a decision: if the number moves, which page, fact or sales asset should change next?
- Adding mentions, recommendations and citations into one score.
- Reporting citations without checking whether the answers misstate your offer.
- Changing the question list every month, so no trend can be compared.
- Letting brand campaigns or PR spikes distort simple before-and-after comparisons.
Minimum Data Hygiene Before You Scale GEO Spend
Check that your enquiry tracking works (in Google Analytics 4, GA4, these are called key events), that links you earn carry consistent tracking tags (UTM parameters), and that your enquiry form or sales process records where the customer heard about you.
Without this, every GEO discussion becomes opinion.
Freeze your definition of a qualified lead for at least a quarter so trend lines stay comparable.
You measure GEO success by keeping mentions, recommendations and citations apart, checking that answers describe you accurately, and linking all of it to enquiries, not by treating AI visibility as an end in itself. If you want help setting up a scorecard, see our AI & GEO service or start with a consultation; for the organic side, read how to measure SEO ROI.
Want a GEO scorecard built around your buyers' questions and your sales process?
Book a strategy consultationFrequently asked questions
What is the single best measure for GEO?
There is not one. Track mentions, recommendations and citations separately on a fixed question list, then add assisted conversions, branded search and feedback from sales.
Can GEO be measured in Google Analytics 4?
Partly. GA4 shows visits and conversions, including referrals from AI assistants when the referrer is passed. It cannot see answers that mention you without a click, so you still need a repeated question sample.
How long until GEO work shows in AI answers?
Fixes to your own pages can be checked as soon as they ship. How long it takes for AI answers to change is not publicly documented, so measure against your own starting point rather than a fixed timeline.
Should we attribute revenue to GEO directly?
Use assisted models and a 'how did you hear about us' question. Last-click attribution will undercount channels that work through influence.
What data do we need before starting?
Working enquiry tracking (key events in GA4), clear stages in your customer records and a written picture of your ideal customer. Without those, GEO reporting becomes anecdote.
How do we avoid vanity reporting?
Tie every chart to a decision it should change: a content update, an internal link, a corrected fact about the business or a new sales asset.
