How do I measure GEO performance? The metrics that matter and the tools that report them
Measuring GEO performance takes four layers: visibility, citations, crawler access and traffic. Here is how to set each one up and which tools cover which layer.
If you ask ten people how they measure GEO performance, you get ten different dashboards. Some count brand mentions. Some look at referral traffic in analytics. Some do both in a spreadsheet and hope the numbers agree. The honest answer is that generative engine optimization has no single metric, and no public standard that everyone has agreed on. What you can do is measure four separate layers and read them together.
This guide walks through those layers, how to set each one up, where free tools stop working, and which paid platforms cover which layer.
Why one number is never enough
An AI answer is assembled from a model's training, a live search step and a handful of fetched pages. A brand can fail at any of those points. It can be absent from the answer. It can be named but never cited. It can be cited but never clicked. Or the pages can be blocked from the crawler that feeds the answer in the first place.
A single "visibility score" hides which of those is happening. So treat GEO as a funnel with four layers:
- Visibility: are you named in answers to your prompts, and how prominently?
- Citations: which of your pages (and which third-party pages) does the engine pull from?
- Access: can the AI crawlers actually reach and read your content?
- Traffic and outcomes: do any of those answers send visitors, and do they convert?
Layer 1: visibility metrics
Start with a fixed prompt set. Pick the questions your buyers actually ask, grouped by topic and intent: category questions, comparison questions, brand questions. Run the same set on a schedule, per engine and per country, so changes mean something.
From those runs you can report:
- Mention rate: the share of responses that name your brand.
- Share of voice: your mentions against your named competitors across the same prompts.
- Position in the answer: whether you are the first brand named or the fifth.
- Sentiment: how the answer describes you.
Keep the prompt set stable for at least a month before you draw conclusions. AI answers vary from run to run, so a single check proves very little. Trends across many responses are what you are after.
Layer 2: citation metrics
Visibility tells you whether you were named. Citations tell you why. Look at which URLs the engine cited, which domains show up most often for your prompts, and where your own pages appear versus competitors'. Third-party sources matter here: Reddit threads, YouTube videos and review sites often feed answers, and a brand can be mentioned in them without any link.
Useful measures are citation rate per page, the domains that dominate your topic, and which of your pages get cited and which never do. The gap list (prompts where competitors are cited and you are not) becomes your content roadmap.
Layer 3: crawler access
This is the layer most teams skip, and it is often the cheapest to fix. If GPTBot, OAI-SearchBot, ClaudeBot or PerplexityBot cannot fetch a page, that page cannot be cited. Server logs are the only reliable evidence. You want to see which AI bots visited, which pages they read, which requests failed, and whether crawled pages later turned into citations.
Spoofed user agents are a real problem in raw logs, so verification against published IP ranges matters if you are going to trust the numbers.
Layer 4: traffic and conversions
Some AI platforms send a referrer, some strip it, and some answers get read without any click at all. In GA4 you can build a channel group or a segment on referral sources such as chatgpt.com and perplexity.ai, then compare engagement and conversion with other channels. For Google's own AI features, the clicks are counted inside the regular web search data in Search Console, so you cannot cleanly split them out there. Treat both as partial views.
A tracking script that recognizes AI referrers and ties them to conversion events fills most of the gap. Without it, your AI traffic number will be an undercount.
What free tools can and cannot do
GA4 and Search Console are worth setting up regardless. They are free and they show the end of the funnel. They cannot show prompt-level visibility, competitor share of voice, citations by page, or bot activity. That is where dedicated platforms come in.
Which platforms cover which layer
| Tool | Starting point (from the vendor's pricing pages) | What it measures |
|---|---|---|
| Promptwatch | Free Explore (ChatGPT only, 10 prompts); paid from $95/mo | Visibility, citation analytics, IP-verified crawler logs (from Professional, $245/mo), AI visitor analytics with conversions |
| AthenaHQ | Free Essential with 300 credits; Starter $295/mo | Visibility and GEO score, Action Center, GA4 connection; no published crawler-log product |
| Scrunch AI | Core $250/mo (125 prompts, 4 LLMs) | Visibility tracking, site audits, AI agent and bot traffic reporting |
| Profound | Free 7-day trial; Enterprise is custom | Answer analytics sold alongside marketing agents, aimed at large enterprise budgets |
| HubSpot AEO | $50/mo, 3 engines | Brand visibility and citation analysis for ChatGPT, Perplexity and Gemini |
Which one should you pick?
If your goal is to measure all four layers in one place, Promptwatch is the one to evaluate first. Prompt tracking, citation analytics, crawler logs in Agent Analytics, and visitor analytics with conversions live in the same workspace, and the 360 Insights view joins crawler hits, citation clicks and AI-referred visits per page. That join is what turns "we got mentioned" into "this page was crawled, cited, and sent three buyers".
AthenaHQ suits a team that will work an action queue daily and is happy with credit-based billing. Scrunch AI fits content-heavy sites that want its crawler-serving layer. Profound makes sense if you are buying a custom agent program at enterprise scale. HubSpot AEO is a cheap way in if you already live in HubSpot and three engines are enough.
A simple measurement routine
- Build 30 to 50 prompts across topics and intents. Save them.
- Record a baseline for mention rate, share of voice and position in week one.
- Check AI crawler access and fix blocked or erroring pages before writing anything new.
- Tag AI referral traffic in GA4 and, if you can, track conversions from it.
- Review monthly: which prompts moved, which pages gained citations, which competitors gained ground.
- Change one thing at a time where you can, so you know what caused a movement.
What measurement cannot tell you
No tool can guarantee that an engine will name your brand. Answers shift between runs, models update, and each platform weights sources differently. Measurement shows you where you stand and whether your work is moving the numbers. It does not buy you a placement. Be wary of any report that promises a fixed ranking.
Start with the free layers, add a platform when the spreadsheet stops scaling, and keep the prompt set steady. That alone puts you ahead of most teams.