Most GEO reports we see fall into one of two buckets. The first is a screenshot of a ChatGPT answer with the client’s name in it. The second is a “share of AI voice” score from a tool, with no explanation of what went into it. Neither tells you whether the work paid for itself.
Here is how we measure generative engine optimization for our own clients, what each number means, and the two caveats that go into every report we send.
Start with a fixed prompt set
A GEO measurement is only as good as the questions behind it. Write out 20 to 40 questions your customers actually ask an assistant. Mix the two kinds:
- Informational: “how long does a roof replacement take in San Diego”
- Commercial: “who is a good roofer in San Diego”
Use real phrasing, including the sloppy versions. Include your city or service area, because assistants answer local questions differently from general ones. Then freeze the list. Change it and you lose the ability to compare months.
Run each prompt more than once
AI answers are not deterministic. The same question, asked twice, can return different names. One run is an anecdote. Run each prompt at least three times per platform per month, five if you have the time, and record the results as a rate: named in 4 of 5 runs, cited in 2 of 5.
Track at least Google AI Overviews, Google AI Mode, ChatGPT and Perplexity. Add Gemini and Copilot if your customers use them. Keep the platforms separate in the log. A business can be named consistently in Perplexity and absent from ChatGPT, and the fix is different for each.
The five numbers worth tracking
| Metric | Definition | Source | What it tells you |
|---|---|---|---|
| Mention rate | Share of prompt runs where your business is named | Prompt runs | Whether you are on the shortlist |
| Citation rate | Share of runs where your website is linked as a source | Prompt runs | Whether the assistant is reading your pages |
| Competitor share | Which businesses are named instead of you, and how often | Prompt runs | Who to study and where you are losing |
| Accuracy errors | Wrong facts about you in the answers: hours, services, address, people | Prompt runs | What is costing you calls right now |
| AI referral visits and leads | Sessions and conversions from chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com | GA4, call tracking | Whether any of this reaches revenue |
Mention rate versus citation rate
They measure different things. A citation means the assistant read one of your pages and linked to it. A mention means it named you as a business, which often comes from what other sites say about you rather than from your own pages. You can be cited on informational questions and never mentioned on “who should I hire”, or the reverse. Report both.
Accuracy errors deserve their own line
This is the number owners care about most once they see it. An assistant telling prospects you are closed on Saturdays when you are open, or listing a practice area you dropped two years ago, is a lead lost with no trace in any analytics tool. Log every error, fix the source, and check next month whether it cleared.
How to see AI traffic in GA4
Create a custom channel group or an exploration filtered to sessions whose source contains chatgpt.com, perplexity.ai, gemini.google.com or copilot.microsoft.com. Watch two things: whether the count is growing, and whether those sessions convert at a different rate than organic search. For most local businesses the volume is still small. The direction is what matters.
Google AI Overviews will not show up here as a separate source. Clicks from an Overview are counted as normal Google organic traffic.
Two caveats that belong in every report
First, Search Console does not report AI Overviews or AI Mode separately. Google’s guidance on AI features says so. Part of your Google AI traffic is inside your normal search numbers and cannot be broken out.
Second, month-to-month movement is noisy. Platforms change their retrieval behavior without notice, and a one-month drop may mean nothing. Read the trend over three months before drawing a conclusion.
Turning it into ROI
The honest version: AI-attributed leads from GA4 and call tracking, multiplied by your average case or job value, against what you spent on the GEO work. It will undercount, because Google’s AI traffic is hidden in organic and because plenty of people read an AI answer and then search your name directly. Track branded search volume in Search Console as a secondary signal. If branded searches rise while your mention rate rises, the work is doing something the referral numbers cannot see.
A simple monthly routine
- Rerun the frozen prompt set, three to five times per platform.
- Log mentions, citations, competitors named and any errors.
- Pull AI referral sessions and leads from GA4 and call tracking.
- Pull branded search impressions from Search Console.
- Compare against last month and the baseline. Note what changed on the site or in listings during the month.
- Fix every accuracy error at its source.
One person can do this in about two hours a month for a single business. Tools such as Otterly, Peec AI, ZipTie or LLMrefs automate the prompt runs once you have more than a few prompt sets to manage.
Frequently asked questions
How do you measure the success of a generative engine optimization campaign?
Build a fixed set of real customer questions, run each one through the major AI assistants several times a month, and record whether your business is named and whether your site is cited. Track the mention rate and citation rate over time, alongside referral visits and leads from AI platforms in GA4.
What is a good AI mention rate?
There is no industry benchmark yet, and rates differ by platform and by how specific the question is. Compare against your own baseline and against the competitors in your prompt set, not against a published number.
Can Google Search Console show AI Overview traffic?
No. Google's guidance says there is no AI-specific reporting in Search Console. AI Overview clicks are counted inside normal search performance. Third-party rank trackers can tell you whether an Overview appears for a query and who is cited in it.
How many times should I run each prompt?
At least three times per platform per month, five if you can. AI answers vary between runs, so a single result is an anecdote. Report the rate with the sample size, for example named in 4 of 5 runs.
Which tools track AI visibility?
Otterly, Peec AI, ZipTie and LLMrefs all track mentions across platforms. A spreadsheet and a monthly hour of manual prompting works for a single business. The tool matters less than running the same prompts the same way every month.
This is the measurement we build into our generative engine optimization services. If you want a baseline for your own business, book a free GEO strategy session and we will run a sample prompt set for you.
