Generative Engine Optimization has a measurement problem: half the industry reports vanity screenshots of one good ChatGPT answer, the other half drowns in dashboards nobody reads. Seven numbers, reviewed weekly, cover what actually matters. Here they are, with how to get each one.
1. Presence rate on category prompts
The share of your money prompts, "best [category] for [audience]" and its variants, where your brand appears in the answer. This is the headline metric because it maps directly to shortlist inclusion. Measure daily across models, report weekly. A presence rate of 10 percent with stable prompts and rising competitors is a fire alarm, whatever your traffic says.
2. Share of Voice
Your mentions as a percentage of all brand mentions in your tracked prompt set: (your mentions \u00f7 total brand mentions) \u00d7 100. It contextualizes everything else. Growth in mentions during a market that grows faster is decline, and only this metric shows it.
3. Average position in recommendation lists
When models list brands, where you sit. Position one gets read; position six gets scrolled past even when it technically counts as a mention. Track the average and the distribution, because a stable average can hide a worrying shift from first places to last places.
4. Citation rate of your domain
How often models link or reference your site as a source, as distinct from mentioning your name. Citations mean the model reads and trusts your content, and they are the metric most directly moved by publishing structured, dated, question-answering pages. Watching which specific URLs earn citations tells you what to write more of.
5. Sentiment and accuracy of mentions
A mention that calls you outdated, or describes you wrong, is not a win. Classify mentions as positive, neutral or negative, and separately flag factual errors. As a reference point from our own tracked data: one brand\u2019s recent 535 classified citations split 314 positive, 121 neutral, 100 negative. The negative fifth is a work queue, not a statistic.
6. AI referral traffic and conversions
The on-site half of the picture: sessions arriving from chatgpt.com, perplexity.ai, gemini.google.com and friends, and what they convert to. Volumes are typically small and intent unusually high, which makes conversion rate the number to watch rather than sessions. Set up channel groups in GA4 once and this reports itself.
7. AI crawler activity on your site
Which AI bots visit, which pages they read, how often they return. Crawling precedes citing, which makes this the leading indicator in the set: movement here shows up in citations weeks later. It requires server-side measurement, since JavaScript analytics do not see bots. Trackbase does this through a WordPress plugin or a Cloudflare integration; raw log analysis works too if you have the access and the patience.
Putting the seven together
Metric | Cadence | Leading or lagging |
|---|---|---|
Presence rate | Daily data, weekly review | Core outcome |
Share of Voice | Daily data, weekly review | Core outcome |
Average position | Weekly | Outcome refinement |
Citation rate | Weekly | Driver |
Sentiment / accuracy | Weekly | Quality guard |
AI referral conversions | Monthly | Business proof |
Crawler activity | Weekly | Leading indicator |
One page, seven numbers, each with a trend arrow. If your GEO reporting does not fit on that page, it is measuring effort instead of progress.
FAQ
What are common GEO tracking mistakes?
Testing only brand-name prompts, checking models while logged in (personalization flatters you), treating one answer as truth instead of sampling daily, and reporting mentions without competitive context.
Can GA4 measure GEO on its own?
It covers metric six only. The other six live at the models and in your server logs, outside what site analytics can see.
Which of the seven should a small team start with?
Presence rate and Share of Voice. They answer the two questions that matter first: are we in the answers, and are we winning or losing ground.
How do I report GEO to leadership?
Lead with Share of Voice against named competitors and AI referral conversions. Both survive contact with a skeptical CFO better than screenshots of good answers.
