I’ve Been Measuring AI Visibility Wrong. So Has Most of the Industry

Introduction: My friend and long-time collaborator Greg Jarboe has been my go-to guy on SEO for at least a decade, regularly teaching our Measurement BAse Camp session on the topic. Naturally, we we added GEO to the course, we turned to our guru Greg. So, when AMEC released their GEO Principles today, I was dying to find out what he thought. The following post is his response:

I owe my followers a correction. Not a factual one, but a conceptual one.

Over the past several months, I have written about generative engine optimization (GEO), AI visibility, citation share of voice, and winning in AI search. The framing I’ve used, borrowed from SEO’s competitive instincts, treats AI citations as the new rankings. More citations, better position, higher visibility. Win. That framing is understandable. It is also, I now think, wrong.

On May 20, 2026, AMEC (the International Association for the Measurement and Evaluation of Communication) launched its GEO Principles and a companion Practitioner’s Guide to GEO Measurement at the AMEC Global Summit in Dublin.

The Seven Principles are:

1. AI-led discovery should be measured against
   communication objectives and stakeholder information
   needs.
2. GEO measurement must assess the upstream information environment before interpreting AI generated answers
3. Search and content readiness should be evaluated as evidence of whether reliable information can be found, understood and cited.
4. Observed AI outputs are directional indicators and should be tested transparently across tools, prompts, markets, languages and time.
5. GEO measurement should distinguish visibility from outcomes and connect AI discovery to awareness, trust, behavior and impact.
6. Reliable, trustworthy and current sources matter more than volume, promotion or short-term visibility.
7. Ethical GEO improves the public information environment and should not manipulate, disguise or flood it

The document was developed over six months by practitioners from FleishmanHillard, Ketchum, Hotwire Global, Converseon, Big Valley Marketing, and PR Agency One, with academic scrutiny from researchers across nine countries. It is the most rigorous professional standards document yet produced for this space, and it directly challenges the way most SEO professionals, including me, have been talking about AI visibility.

Like I’ve said before, sometime you find the news. Sometime the news finds you.

What AMEC Got Right That I Glossed Over

The principles open with a framing that sounds simple but isn’t. GEO measurement focuses on whether stakeholders encounter information that is accurate, useful, current, credible, and trustworthy, rather than simply increasing visibility.

That word “rather” is doing significant work. AMEC is not saying visibility doesn’t matter. It is saying that visibility is the wrong primary objective, and that treating it as the primary objective leads practitioners toward exactly the behaviors that degrade the information environment. Flooding the web with content designed to manipulate citation frequency. Disguising promotion as independent evidence. Treating AI outputs as factual without verification.

I have argued in this series that the commodity/non-commodity divide is where the real strategic question lives. That content produced without genuine expertise, direct experience, and editorial accountability will lose to content that carries those things. That ground-truthing matters. All of that is consistent with what AMEC is saying.

What I failed to make explicit is the reason those things matter. Not because they will get you more citations, but because they improve the accuracy and usefulness of the information environment that stakeholders actually rely on. The competitive framing and the ethical framing point at the same behaviors, but they point at them for different reasons, and the reason matters.

The Three Components I’ve Been Conflating

AMEC structures GEO measurement around three “Evidence Domains” — i.e. pillars of information — that need to be assessed in sequence, and reading them helped me see that I’ve been jumping between them without acknowledging the differences. They are:

1. Upstream information and reputation

The first is upstream information and reputation – earned media, shared media, owned content, public records, reviews, and stakeholder discussion. This is the information environment that AI models draw on when generating answers. Before you can interpret what an AI says about your organization, product, or issue, you must understand what sources it has available to learn from. I’ve touched this domain in pieces about content quality and Experience, Expertise, Authoritativeness, and Trustworthiness, (E-E-A-T) the framework Google uses to evaluate the quality and credibility of web content when deciding how to rank it in search. I’ve written about it before, but never named it as the necessary starting point.

The four characteristics of E-E-A-T break down like this:

• Experience — Does the author have first-hand, real-world experience with the topic? A product review written by someone who actually used it outranks one written by someone who didn’t.
• Expertise — Does the author have the knowledge and skills to write authoritatively on the subject? This is particularly important in what Google calls “Your Money or Your Life” topics — health, finance, legal, and safety content.
• Authoritativeness — Is the author and the publication recognized as a go-to source in their field? This is built through citations, backlinks, mentions by other credible sources, and reputation over time.
• Trustworthiness — Is the content accurate, transparent, and honest? This includes things like clear authorship, cited sources, factual accuracy, and editorial standards.

2. Search and content readiness

The second domain is search and content readiness – whether reliable information is discoverable, structured, current, accessible, and supported by credible sources. This is the technical and editorial layer between what you have produced and what an AI system can find, parse, and cite. I’ve discussed structured data, entity recognition, and content architecture in passing, but AMEC’s framing makes clear this is a distinct discipline from the upstream reputation work.

3. Downstream AI output tracking

The third is what stakeholders actually see, including presence, framing, citations, omissions, source quality, accuracy, and risk. This is the domain I’ve discussed most, and where the competitive citation-counting metrics live.

Observed AI outputs

AMEC’s fourth principle– that observed AI outputs should be evidence of reliable information can be found, is the one that should change how practitioners talk about it: observed AI outputs are directional indicators and should be tested transparently across tools, prompts, markets, languages, and time.

That last sentence matters more than it might appear. Citing what a single AI tool returns in response to a single prompt, at a single moment, in a single market, is methodologically weak evidence. It is illustrative at best. Meaningful GEO measurement requires a governed query library linked to stakeholder questions, documented tools and prompts, repeat testing with disclosed variation, and saved outputs as evidence. Most of what gets presented as GEO analysis in the trade press, does not meet that standard.

The Principle That Changes Everything

AMEC’s fifth principle is the one the SEO industry needs to analyze and examine: GEO measurement should distinguish visibility from outcomes and connect AI discovery to awareness, trust, behavior, and impact.

This is the ground-truthing argument applied to GEO specifically. Appearing in an AI-generated answer is an output metric. Whether that appearance changed what a stakeholder understood, believed, or did is an outcome metric. The industry has spent years learning, painfully, that search rankings are outputs and conversions are outcomes. GEO measurement is going to have to learn the same lesson.

The practical implication for SEO professionals and digital marketers is that the question “are we getting cited in AI answers?” is the beginning of the measurement conversation, not the end of it. The questions that follow are harder: is the citation accurate? Is it framing us the way we intend? Is it reaching the stakeholders who matter? Is it changing anything they do? Those are the questions that require more rigorous methodology, and the ones AMEC’s seven principles are designed to answer.

What This Doesn’t Change

AMEC’s ethical guardrail is direct. Do not flood the web with low quality content, disguise promotion as independent evidence, manipulate reviews or forums, or treat AI outputs as factual without verification. No single score, tool, or prompt set can prove total AI visibility or communication impact.

That last sentence should be printed on every GEO dashboard currently being sold. The proliferation of tools offering an “AI visibility score” is producing the same dynamic that domain authority scores produced in traditional SEO. A single number that feels like an answer but isn’t, and that creates incentives to optimize the score rather than the underlying reality the score was meant to approximate.

The argument I’ve made many times in the past is that reliable, trustworthy, current sources matter more than volume or short-term visibility, is now AMEC’s sixth principle. The argument that ethical practice improves the public information environment rather than manipulating it is their seventh. The argument that visibility without outcomes is a vanity metric is their fifth.

I wasn’t wrong. I was imprecise, and imprecision in measurement is how good intentions produce bad incentives.

The Correction

Going forward I’ll try to be more careful about which of AMEC’s three evidence domains I’m discussing in any given piece, more honest about the methodological limitations of single-prompt AI output citations, and more consistent about separating visibility from outcomes.

That’s not a retreat from the argument. It’s the ground-truthing applied to my own position. Which, as I’ve been saying all along, is the only honest way to navigate this.