Guide

How to read your AI visibility report

Twenty views, grouped by the question you are actually asking. What each one is, what it tells you, and what to do when it looks bad.
MS
Michael Stratta
Founder & CEO, Arcalea
Current as of August 2026 · Describes the v2 two-layer index
Quick answer

An Arcalea AEO Index has twenty views. They answer five questions in order: where your brand stands, why it stands there, where AI is actually looking, how that relates to your existing search presence, and what you can change. Start with the Delta Quadrant to find your position, open your own row in the entity explorer to see the five signals behind it, then read the recommendations. Everything else is depth. The scoring model itself is published separately as the AEO Index methodology.

Start here

If you have ten minutes, read three sections.

The index is a reference. This is the path through it for someone who has just received their report and wants to know what to do.

1. The Delta Quadrant, to find your position

It places you on two axes at once and tells you which of four situations you are in. Two minutes.

2. Your own row in the entity explorer, to find the cause

Open the row and read the five-by-two matrix behind your score. The largest gap between your memory and retrieval cell for the same signal is your most specific problem. Five minutes.

3. Recommendations, to find the first move

Generated from your position across every section. Take the top two or three. Three minutes.

Come back for the rest when you want to understand why, or when someone asks you to defend a number.

Before you start

What counts as a good score?

It is the first question everyone asks, and the honest answer is that the absolute number is the least informative thing on the page.

Scores are comparable within an index and across its two layers. They are not comparable between indexes, because different categories ask structurally different questions, so a 0.6 in one category and a 0.6 in another are not the same achievement. There is no universal threshold, and anyone offering you one is guessing.

Two readings do mean something. Your position relative to your own category, which is what the quadrant and the leaderboard show. And the gap between your two layer scores, which is what tells you whether you are coasting on reputation, entering the category, or absent from both. The gap is the diagnostic. The number is just the label on it.

The Delta Quadrant, read on real data Every entity in the M7 Business Schools index plotted by model memory on the horizontal axis against live retrieval on the vertical. Two median lines split the field into four positions. Harvard Business School sits top right as a dual leader. INSEAD sits above the retrieval median and below the memory median, a category entrant. GMAT Club sits bottom right, known but not confirmed by the live web. Poets and Quants sits bottom left. The Delta Quadrant, read on real data ALL 29 ENTITIES, M7 BUSINESS SCHOOLS INDEX, AUGUST 2026 median memory median retrieval CATEGORY ENTRANT Found when AI searches, not yet remembered DUAL LEADER Known and found ABSENT Neither known nor found COASTING ON REPUTATION Known, not confirmed by the live web Harvard INSEAD Poets & Quants GMAT Club Model memory, what AI knows without searching → Live retrieval, what AI finds → Your quadrant matters less than your distance from the lines. The bigger the gap between your two scores, the more specific the problem, and the more fixable it usually is.
One category, plotted. Position matters, and distance from the median lines matters more.

The question

Where does my brand stand?

What is the Delta Quadrant, and where should I be on it?

What it isThe signature view. Every entity in your category is plotted twice: model memory on the horizontal axis, what AI knows about you without searching, and live retrieval on the vertical, what it finds when it does search. Two median lines split the field into four positions.

What it tells youYour quadrant is a diagnosis, not a grade. Top right means known and found. Top left means AI finds you when it looks but has not internalised you, which is where category entrants sit. Bottom right means AI knows you but the live web is not confirming it. Bottom left means neither.

What to doRead your distance from the lines, not just which box you are in. A brand sitting far from the median on one axis and close on the other has a specific, usually fixable problem. A brand near the centre on both has a diffuse one, which is harder.

See this section on a live index

How do I read the blended leaderboard?

What it isThree columns per entity: memory, retrieval, and the blended headline that combines them at 0.32 memory and 0.68 retrieval.

What it tells youThe blended number is for ranking. The two layer scores are for understanding. Two brands with the same blended score can be in completely different situations, and the leaderboard is the fastest place to spot that.

What to doIgnore your rank for a moment and read the gap between your own two columns. That gap is the thing you can act on. Note also the recommendation-quality chip, which reports how AI frames you when it names you. It is a separate measure from First-Position Rate and the two are easy to confuse.

See this section on a live index

Why is head-to-head measured both ways?

What it isAsk a model to compare two brands and a large part of the answer is an artifact of which one you named first. Every matchup in the index is run in both orders and the results are averaged.

What it tells youThe order-balanced number is the one to trust. The toggle lets you see the size of the correction, which is often larger than people expect.

What to doIf your naive win rate looks strong but your order-balanced rate does not, you were benefiting from question phrasing rather than from position. Treat the balanced number as the real one in any competitive claim you make.

See this section on a live index

What is in the entity explorer?

What it isThe drill-down. Per-entity two-layer and organic detail, and any row opens the five-by-two matrix behind that brand's scores.

What it tells youThis is where a diagnosis stops being directional and becomes a specific cell. Strong memory on Entity Mention Frequency with weak retrieval on First-Position Rate is a different brief from weak Topical Range Score in both layers.

What to doOpen your own row first, before reading anything else on the page. It is the fastest route from a score to a content decision.

See this section on a live index

The question

Why does it stand there?

How is the score actually calculated?

What it isTwo layers, each scored on the same five signals: Entity Mention Frequency (EMF), Topical Range Score, Position Power Score, First-Position Rate and Cross-Platform Stability.

What it tells youThe panel on the index summarises the model. It is deliberately brief here because the full method, with weights, a worked example and the stated scope of what the numbers support, is published separately.

What to doRead the AEO Index methodology once, properly, before you argue with a number. Everything else in this guide assumes it.

See this section on a live index

What does attribute association tell me?

What it isWhich factors the models associate with each brand in the category. A darker cell means a stronger association.

What it tells youThis is where you find out you are known for something other than what you sell on. A brand can be highly visible and still be associated with the wrong attributes, which converts badly.

What to doCompare your row against the two or three competitors you actually lose to. If they own an attribute you consider yours, that is a positioning problem showing up in the measurement, not a measurement problem.

See this section on a live index

Is my category dominated by a few brands in AI answers?

What it isConcentration and saturation. Whether a small number of names absorb most of the visibility, or the field is fragmented across many.

What it tells youThis changes the strategy rather than the tactics. A concentrated category means displacement, which is slow and expensive. A fragmented one means occupation, which is faster and cheaper.

What to doIf your category is concentrated and you are not one of the concentrated names, do not start by attacking the leaders. Start with the sections below on white space and fan-out.

See this section on a live index

What is white space in AI search?

What it isHigh-volume query clusters where nobody in the cohort has built a strong position.

What it tells youUnclaimed ground. These are questions your buyers are asking that no competitor is currently answering well enough to be cited for.

What to doThis is usually the fastest-moving part of an index for a client, because there is no incumbent to displace. Treat the list as a content brief rather than a report.

See this section on a live index

The question

Where is AI actually looking?

What is search fan-out?

What it isOne question quietly becomes many. When a model answers a category question, it runs a set of narrower sub-searches first. This section shows which ones.

What it tells youIt amounts to a content brief the model wrote itself. If a sub-search theme dominates the fan-out and you have nothing addressing it, you are absent from the part of the process that decides the answer.

What to doMap your existing content against the fan-out themes. The gaps are not hypothetical keyword opportunities; they are questions the model is demonstrably asking on your behalf.

See this section on a live index

Where do AI citations actually go?

What it isRead it left to right: the kind of question on the left, the source the model cited on the right.

What it tells youThis is usually the most uncomfortable diagram on the page. In most categories, aggregators and a broad long tail absorb the majority, and the brands being discussed are a thin slice.

What to doFollow the thickest bands. Those sources are your real competition for the answer, and they are frequently not the competitors in your pitch deck.

See this section on a live index

Which of my pages does AI actually cite?

What it isCitation sources and page types. Of the pages on your own domain that do get cited, which kinds they are.

What it tells youTwo readings. First, how much of the category's citation volume reaches brand-owned domains at all. In the M7 index, across 10,255 cited URLs, 5.7% point at school sites, 35% at aggregators and 59% at everything else. Second, which of your page types earn what you do get.

What to doBuild more of whatever is already being cited, and stop assuming your service pages are doing the work. In most categories they are not.

See this section on a live index

How often do AI answers cite sources at all?

What it isCitation rate across the category, and the domains the models lean on most.

What it tells youA high citation rate means the answers in your category are grounded in retrievable sources, which means being retrievable matters. A low one means the models are answering from memory, which changes where to invest.

What to doUse this to decide which layer to work on first. Heavily cited categories reward retrieval work. Lightly cited ones reward the slower work of entity corroboration.

See this section on a live index

How is Google AI Overviews different?

What it isGoogle's AI layer, measured separately because it behaves differently from the chat engines. In the M7 index an AI Overview appears on 79% of the 772 category keywords, making it the highest-volume AI surface measured.

What it tells youEven Google's own AI leans on third parties over the brands themselves. Brands tend to be cited on their own names and much less on non-branded discovery queries, which are the ones that acquire.

What to doDo not treat AI Overviews as an extension of your organic ranking. It is a separate surface with separate source preferences, and a strong blue-link position does not guarantee a citation.

See this section on a live index

The question

How does this relate to my search presence?

Does my SEO ranking predict my AI visibility?

What it isThe divergence view. AI visibility plotted against organic standing for every entity.

What it tells youMostly, no. The two diverge often enough that treating one as a proxy for the other is a mistake. This is the single most useful section for anyone who already invests seriously in SEO.

What to doIf you rank well and are invisible in AI answers, the problem is citability rather than authority: structure, extractability, and whether the sources models read are reflecting you. If the reverse, your organic footprint is the thing lagging.

See this section on a live index

How does organic share of voice relate to AI visibility?

What it isYour share of the estimated addressable organic clicks in the category.

What it tells youThe traditional-search baseline. It is here to be read against the AI layers, not instead of them.

What to doUse it as a reality check on scale. A brand with negligible organic share and strong AI retrieval is punching above its weight and should understand why before assuming it will last.

See this section on a live index

What does SERP and keyword coverage add?

What it isWhere you appear across the keyword universe the index is built on.

What it tells youThe demand map underneath both layers. It shows which parts of the category's question space you are present in at all.

What to doRead it alongside white space. Coverage tells you where you show up; white space tells you where nobody does.

See this section on a live index

Does branded search demand explain AI visibility?

What it isBranded-search footprint per entity, measured against the AI layers.

What it tells youThis section earns its place by carrying a counterintuitive result: branded search volume and AI visibility are close to uncorrelated. Being well known to people does not reliably make you well known to models.

What to doStop using brand awareness as evidence that AI visibility will follow. It does not, and this is the section to show anyone who assumes otherwise.

See this section on a live index

The question

What can I actually change?

Does updating old pages improve AI visibility?

What it isHow recently each entity's most-cited pages were last updated.

What it tells youPages refreshed within 90 days earn far more AI citations than stale ones. This is one of the clearest effects in the index.

What to doProbably the highest ratio of impact to effort available. Start with pages that are already earning citations and have been left to go stale. Refreshing those beats writing new ones that nothing points at yet.

See this section on a live index

What makes a brand machine-readable to AI?

What it isCitability infrastructure: structured data, an encyclopedia presence, an AI-crawler guide, and named authors, combined into one score.

What it tells youWhether a machine can recognise you as an entity at all, as distinct from whether your content is good. A brand can publish well and still be hard to resolve.

What to doThis is the layer that moves memory rather than retrieval, so it is slow and worth starting early. Missing structured data and a missing encyclopedia presence are the two that show up most often.

See this section on a live index

How do I turn this into a plan?

What it isThe recommendations pane, generated from your own position across the sections above.

What it tells youIt is a prioritised starting point, not a delivery plan. The index diagnoses; it does not execute.

What to doTake the top two or three, and sequence them against the freshness and infrastructure sections, which are the levers with the shortest and longest payback respectively. If you would rather start from a plan than a score, the 21-step Marketing Planning Diagnostic is free and open.

See this section on a live index