GEO methodology: how we build and measure AI visibility
Methodology, 2nd edition · last reviewed: September 2026
Building AI visibility runs from classical SEO through GEO and AEO all the way to your brand becoming part of the machine's memory. That is the real, long-term machine equity. This page describes how we work: we measure, we plan, we fix, we implement — then we measure again, to catch citation drift early. Every number we publish carries its run count, and at the end sits the benchmark you can check us against.
Request the AssessmentAssessment · a fixed fee, credited to the work
The terms these methods serve:Knowledge Engine Optimization (KEO)Model Mindshare
The machine-equity loop — GEO methodology: how we build and measure AI visibility
- 1 · MEASURE — “Where do I stand today?” Per engine, with a range (Citation Tracker, Selector)Citation Tracker → · Selector →
- 2 · FIX — “Why do they skip me?” Page by page, then fixed in code (Gap Analysis, Gap Closure)Gap Analysis → · Gap Closure →
- 3 · ANCHOR — “Where does the machine learn about me?” An unambiguous entity, present where models learn (Entity Anchor, Corpus Campaign)Entity Anchor → · Corpus Campaign →
- 4 · PROVE ↺ — “Did it move, and what did it bring?” Closed-book baseline and the delta at every new model generation (Model Memory Audit, Release Audit)Model Memory Audit → · Release Audit →
KG Doctor and the Claim Coherence Engine — the studio's own instruments — serve the entire loop.
Machine equity is the business asset that arises from AI systems being able to recognize, understand, recall, recommend — and, on a user's behalf, transact with — a brand. Composed of three layers: rented reach (retrieval visibility), owned memory (trained model recall), and machine buyability (agent-side transactability). Measured honestly only as a portfolio of ranged, per-layer metrics — never as a single blended score.
Machine equity: the definitionWhat we measure, and why
AI assistants don't answer from a fixed index. They answer from what they retrieve this second and from what they already know. So we measure both layers: whether you appear in live answers, and whether the model knows correctly who you are. These are two different terrains with two different interventions; measuring only one leaves you blind on the other. And we separate being named from being merely used as a source: by Semrush's June 2026 estimate, 61.7% of AI citations were ghost citations — the answer used the source page but never named the brand. So we measure mentions, not mere source presence.
The seven dimensions
Live AI visibility — whether you show up when the model searches — breaks into seven transparent dimensions, not one black-box number. Each is something you can see and check.
- Answer presence
- Do you show up in the live answer at all? When someone asks AI in your category and it searches, are you in the response — or absent while a competitor is named.
- Answer position
- How prominently do you appear? Named up front as a top pick, or buried late. Position shapes whether the buyer even reaches you.
- Named, not just cited
- Does it say your name, or just use your page? The answer can pull facts from your site yet never name you — a "ghost citation." We separate being named from being merely a source.
- Framing
- How does the answer talk about you? Favourably, neutrally, or poorly — and in what context. A bad mention isn't a recommendation.
- Answer stability
- Does it say the same thing each time? AI answers shift between runs. We ask repeatedly and report the range, not a lucky single response.
- Citation footprint
- Which sources does it pull from — and are you among them? The answer is built from a handful of trusted pages; we track where that territory is and whether you hold any of it.
- Intent coverage
- Across the buyer's real questions, where do you surface? Discovery, comparison, "best for X," objections — visibility in one prompt isn't visibility across the journey.
The seven dimensions are measured separately and reported separately, each with its own confidence range. We don't blend them into a single score: that would require calibrated weighting, and we only assign weights based on measured data.
Many measurements work with one composite score, then add that the result is directional. We would rather show how many runs each number comes from.
The live measurement runs on six AI surfaces. On five engines we repeat the same queries ten times; for Google AI Overviews we use a separate measurement method.
We measure the wobble
Ask an AI the same question twice and the answer moves — wording, sometimes even the names. A single response is noise. So we don't report a single answer: we take the real prompts buyers use in your category, ask them across several engines, repeat, and report the result as a range with a confidence interval — the signal and its wobble together. The same honesty runs across the site: the Model Mindshare meter shows a level on a scale, never a promised number.
Each engine is a different job
AI engines don't use sources the same way, so the same intervention isn't worth the same everywhere. The data below comes from our own measurement: 12 August 2026, 12 questions, 3 engines, 5 repeats, 180 runs in total.
ChatGPT
What we measured: It used a live source in 12% of runs. On "who should I avoid" questions it searched in 0 of 30 runs.
What it means: In these answers the model's existing knowledge dominates. Your entity and the corpus about you can matter more than a freshly optimized page.
Gemini
What we measured: It used a live source in 72% of runs.
What it means: Search and the retrievable passage together decide which source can enter the answer. Meet the conditions of entry first; only then worry about position.
Perplexity
What we measured: It worked from a live source in all 60 runs.
What it means: Here it matters most that the source page is fresh, reachable and easy to retrieve. Perplexity assembles its recommendations from its own source and index set.
For Claude, Google AI Overviews and AI Mode we don't yet have our own publishable measurement in this breakdown. We will publish numbers about them once a proper measurement stands behind them.
The full measurement note: engines and source use, August 2026 →
Where the machine learns about you
A significant share of AI answer sources is not on your own domain.
In a December 2025 Ahrefs analysis, the factor most strongly co-moving with AI visibility was YouTube presence, at a correlation of roughly 0.737. Domain authority correlated far more weakly.
So we don't examine only your own website. We also map which external sources the AI answers in your category are built from, and whether your company is present among them.
We call this the Corpus Map. The Corpus Campaign module builds on it.
If your AI visibility work consists solely of optimizing your own website, a large share of the source space is left out.
Ownership: whose job is it?
AI visibility programs often stall not on expertise but on a simple question: who owns the task?
If your company doesn't show up for an important question, someone has to find the cause and carry the fix through. The breakdown below shows who can own each terrain on your side, and which of our modules works the same terrain if you hand the implementation to us.
| Terrain | Owner on your side | Module on ours |
|---|---|---|
| Measurement and question selection | marketing or leadership | Citation Tracker, Selector |
| Content and answer-ready structure | content team | Gap Analysis, Gap Closure |
| Access, rendering, schema | developer | KG Doctor, Release Audit |
| Entity and factual claims | brand owner | Entity Anchor, Claim Coherence Engine |
| Off-domain presence | PR and community | Corpus Campaign |
What the measurement decides — and what it doesn't
We don't promise a guaranteed percentage, ranking or mention. We deliver a sampled signal, with a range and a run count. So don't read the result off a single number: watch how the range moves.
Mentions and click-throughs are two separate measurements. We log clicks arriving from AI surfaces separately, then examine them alongside the mention data. We never convert one into the other.
Nor do we automatically attribute revenue to AI visibility. If a provider promises a numeric revenue impact, it's worth asking what control they used to isolate the effect of AI visibility from everything else.
How to check us — and anyone else
Our methodology comes with an 18-point benchmark. All 18 points are yes-no questions, so any AI visibility provider's public methodology can be checked against it. Ours included.
We publish both playbooks step by step, and the notes and studies from our own measurements are publicly available on the research page.
- The 18-point methodology benchmark →
- The KEO playbook →Building the verified, canonical, readable, citable record AI models learn from.
- How we measure Model Mindshare →Prompt portfolio, repeats per engine, closed-book (web off), named-vs-known, confidence interval.
- Measurement notes and studies →
What we're working on now
- Per-engine pages for the AI surfaces we haven't measured ourselves yet.
- Breaking the seven dimensions out into a factor sheet, with run counts shown everywhere. This will be the Visibility Atlas.
- A public sample of the Corpus Map.