Measure five separate things, not one score. That is the core shift in the PRCA AI Innovation Group's "How to Measure GEO" guide, published in June 2026 as a practical companion to the Association for Measurement and Evaluation of Communication's GEO Principles. Most companies that have started tracking generative engine optimization, whether and how often ChatGPT, Gemini, Perplexity and Google AI Mode mention them, have settled on a single number: a visibility percentage, a mention count, a composite index score from a vendor tool. The guide's authors, a group of six measurement and AI practitioners from agencies including Golin Ketchum, Burson and Purposeful Relations, argue that a single number cannot carry the weight being put on it, and lay out what to track instead.
Who actually wrote this, and what is it attached to?
The guide, subtitled "Beyond Visibility," was published by the PRCA's AI Innovation Group with five named authors: Jonny Bentwood of Golin Ketchum, Anna Wilson of Tangerine, Alison Spray of Burson, Sanjiv Winayak of Milk & Honey, Stuart Bruce of Purposeful Relations, and Nathalie Agnew of Muckle Media. It is built as the practitioner-facing companion to the AMEC GEO Principles, which AMEC launched at its Global Summit in Dublin on May 20, 2026, after more than six months of work with agency, academic and vendor input, including Converseon. Bentwood's own framing of the project is blunt: "GEO is too important to hand to vanity metrics."
What are the five dimensions it says you should track?
The guide organizes measurement into five connected areas, moving from awareness down to business impact.
Presence covers visibility share by topic area and how often your brand gets named without a prompt specifically asking about it.
Reach covers how that presence spreads across AI engines, since a brand that shows up well in ChatGPT but nowhere in Gemini or Perplexity has a concentration problem, not broad visibility.
Quality covers sentiment, how your brand is framed against competitors, the authority of the sources an AI system is citing when it mentions you, and whether what it says about you is actually accurate.
Behaviour covers what happens after someone sees an AI answer that mentions you: AI-referred traffic to your site, whether a user's follow-up prompts continue engaging with your brand, and any resulting lift in branded search volume.
Commercial covers whether your brand is actually recommended or preferred in high-intent prompts, and whether that recommendation leads to a measurable outcome, an assisted conversion or its equivalent for your organization.
What are the seven specific metrics it names?
Rather than leaving the five dimensions abstract, the guide proposes seven starting metrics it calls the baseline set: GenAI Visibility percentage by topic, GenAI Share of Voice against named competitors, GenAI Net Sentiment, cited URL count checked against your own server logs rather than taken on a tool's word, branded search volume trend, key event completions, and accuracy of the answers AI systems give about you. That log-verification detail matters more than it sounds. A visibility tool can report that a model cited your page; your server logs are what confirm a crawler or a referred user actually reached it.
Why does the guide call citation volume a "vanity metric"?
Because an AI answer is not a stable, repeatable rank position the way a page-one Google result used to be. The same prompt, asked twice, by two different users, in two different sessions, can return two different answers with two different sets of sources. The guide states plainly: "User-facing citation volume is a vanity metric." Its broader instruction is to "treat AI visibility as a directional indicator" rather than a precise score, and to never rely on any single platform, tool or number as proof of anything. That instruction echoes AMEC's own GEO Principles, which set baseline evidence standards of repeatable prompts, documented methods and stated limitations before any number gets reported upward. A citation-volume count without that context is the same mistake covered in our look at who dominates AI citations: the raw number tells you less than the pattern behind it.
How often should you actually retest, and is anyone checking whether AI crawlers can even reach your site?
Monthly is the guide's default cadence for an active GEO program. Alongside that, it recommends a quarterly check of your own robots.txt and server logs against the specific crawler user agents that matter: GPTBot, ClaudeBot, PerplexityBot and Google-Extended. That check matters because a default configuration, a CDN setting, or a security tool can block one of those crawlers without anyone noticing, which silently caps your Presence and Reach numbers regardless of how good your content is. The guide treats llms.txt as optional and unofficial, noting that no major AI provider has confirmed it actually changes what gets served, a conclusion that matches our own look at the file.
How should this actually get reported, and to whom?
In three tiers, the guide says, not one slide reused for every audience. A boardroom report should carry the Commercial-dimension numbers and a short directional read. A program team report should carry all five dimensions with month-over-month movement. An analyst-level report should carry the full methodology, the raw prompts used, and the log-verified citation data behind the summary numbers above it. The guide maps this structure onto AMEC's existing Integrated Evaluation Framework, outputs, out-takes, outcomes and impact, so a communications team already using that framework for traditional PR measurement does not have to build a second, incompatible one for AI visibility. Google's own Search Console AI performance report and third-party AI visibility tools can both feed data into this structure, but neither replaces it on its own.
Does any of this work if nobody at your company has actually been assigned to it?
No, and that is the more common starting problem than a bad measurement framework. Muck Rack's 2026 State of PR report found 29% of PR professionals say no one at their organization owns GEO at all, which means the five-dimension structure above has nowhere to live. A measurement framework assumes someone is checking the crawler logs monthly, someone is writing the analyst-tier report, and someone is deciding what the boardroom actually needs to see. Fix the ownership gap first; the framework in this guide is what that person should be running once they exist.
The direct answer, for anyone who read this far
Stop reporting a single AI visibility score and start tracking Presence, Reach, Quality, Behaviour and Commercial outcomes separately, retested monthly, with citation claims checked against your own server logs rather than a vendor dashboard alone. The PRCA's guide exists because the industry kept treating a directional signal as a precise ranking, and that habit produces reports that look confident and mean very little. If you want help building a GEO measurement program that survives a skeptical finance team's questions, talk to us.
