How do I measure if my GEO or AI visibility efforts are working?
How do I measure if my GEO or AI visibility efforts are working? The metrics that matter, how to baseline them, and which tools cover each layer, Promptwatch first.
You have published content, fixed some technical issues and maybe pitched a few third-party sites. Now someone asks whether GEO is working. If your only evidence is "I asked ChatGPT and we showed up", you cannot answer that.
Measuring generative engine optimization takes four layers of evidence, taken in order: whether you appear, whether the engines can reach you, whether they cite you, and whether anyone arrives and does something. Most tools cover the first layer. The useful ones cover all four. Promptwatch is the one we would start with, for reasons that follow.
Layer one: are you appearing in answers?
Start with a fixed panel of prompts that reflect how buyers actually ask. Track them daily, per engine, per country. From that panel you can compute:
- Visibility Score, the share of answers in which your brand appears.
- Share of Voice, your appearances against those of named competitors for the same prompts.
- Position in Answer, where in the answer you are named.
- Sentiment, how the answer describes you.
The panel matters more than any single metric. If you change the prompts every month, your trend line is meaningless. Fix the set, add new prompts in a separate group, and compare periods against the same baseline.
Promptwatch runs prompt panels daily per model, country and language, and reports Visibility Score, Share of Voice, Sentiment (scored 1 to 100 per response, with a weekly AI summary) and Position in Answer. An A/B period comparison lets you set the weeks before and after a change side by side, and brand aliases plus a Brand Book control what counts as a mention.
Layer two: can the engines reach your content?
A page that AI crawlers cannot fetch will not be cited reliably, however good it is. Check crawler logs for bots like GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot and Google-Agent. Look for errors and blocked paths, and for pages that never get crawled.
Promptwatch's Agent Analytics logs these in real time with IP verification, so spoofed user agents are filtered out, and it shows errors with failure explanations. The TreeHouse case study (Pararius and Huurwoningen.nl) is a good example of why this layer matters: by Promptwatch's own account, AI search became about 25% of their organic conversions, 99% of it from ChatGPT, after they fixed crawler access and automated weekly reporting. That is one customer's result, not a promise, but it shows the order of operations.
Layer three: are you being cited?
Appearing in text and being cited as a source are different signals. Track which of your URLs are cited, in which position, and how the set of cited domains changes over time. Citation Trends in Promptwatch shows first-cited date, ramp-up time, peak rate and decay half-life per page, which is the closest thing the category has to a content performance curve. Page Tracker lets you follow specific URLs, including competitors'.
Layer four: does it produce visits and conversions?
This is where most measurement plans stop short. Referral traffic from AI engines can be seen in a general analytics tool, but only if the referrer is passed, and not all engines pass it. A dedicated AI traffic view helps. Promptwatch's visitor analytics attributes AI-referred traffic and conversions (ChatGPT, Perplexity, Gemini, Claude, Copilot, DeepSeek and more) through a lightweight script, a GTM template or the WordPress plugin. Crisp's case study reports that AI-referred traffic converted at twice the rate of traditional channels. Again, that belongs to Crisp, not to every user.
Search Insights adds Google Search Console and Bing Webmaster data on top, so you can set classic search traffic next to AI crawls and citations.
How the tools compare on measurement depth
| Tool | Prompt visibility | Crawler logs | Citation analytics | AI visits and conversions |
|---|---|---|---|---|
| Promptwatch | Yes, daily per model | Yes, from Professional | Page and domain level | Yes, with conversions |
| AthenaHQ | GEO score, 11 models on Starter ($295/mo) | Not published | Citation Engine on Enterprise | GA4, Shopify and Webflow attribution |
| Scrunch AI | 4 LLMs on Core ($250/mo) | Bot traffic reporting | Insights | Not published |
| Rankscale | Yes, from $20/mo (yearly) | No | Citation analysis | No revenue attribution |
| Ahrefs Brand Radar | Yes | AI crawler activity | Cited sources | Not published |
AthenaHQ is the strongest alternative if you want an action workflow tied to GA4 and Shopify attribution. Scrunch AI fits teams interested in crawler-side delivery. Rankscale is a cheap way to trend visibility, though its own documentation concedes that visibility leads business results by weeks and only proxy linkage exists. Ahrefs Brand Radar suits teams already inside Ahrefs.
A simple measurement plan
- Weeks one and two: build a prompt panel of 30 to 100 questions grouped by topic and buyer persona. Record the baseline for visibility, share of voice and citations.
- Week three: check crawler access and fix blocks before publishing anything new.
- Weeks four to eight: ship changes in small batches, one topic at a time, and annotate the date of each change.
- Monthly: compare the same prompt panel period over period, then read citations and AI-referred visits for the topics you changed.
Expect lag. Citations and visits tend to follow publishing by weeks, not days, and answers vary run to run, so judge trends over several weeks rather than single days.
Who should pick which
If you need the full chain, from appearance to conversion, in one workspace, Promptwatch is the better fit. If you only need a visibility trend and the budget is tight, a tracker such as Rankscale will do. If your team needs Google, Bing and AI data in one view, look at Promptwatch's Search Insights or at Ahrefs if you already pay for it.
Whatever you choose, do not report one number. Report four: appearance, access, citation and visits. That is the honest way to say whether GEO is working.