The complete AI-visibility index for M7 schools.

What AI already says about this category

Ninety seconds. AI answers this category from two places at once, what the models already remember about it and what they retrieve live, and this index measures both. Every figure below is drawn from a panel further down this page, and none of it is an estimate.

The Delta Quadrant

Every school placed by what AI remembers (model memory) against what AI finds when it searches (live retrieval). Bubble size is mention volume. Tap any school to focus it; switch Layer, Platform or Signal to move the category.

What this category supports

Measured before the page was shaped. It answers one question: do AI answers about this category name real schools often enough for position to be the story?

How big a move has to be before it is real

Each figure is a minimum detectable change: the smallest move that is real at 95% confidence, measured by running the whole instrument twice over the 18 tracked schools. Occasions: Model memory both on 2026-08-22; Live retrieval both on 2026-08-22. These occasions are hours apart, so every figure is a floor: a day-separated repeat is expected to raise them, not lower them.

What may be published, dimension by dimension

A cell is rankable when its bands separate. It is not measurable when the dimension moves more between two runs of the same instrument than it varies across the 18 tracked schools. Where the cells disagree, the disagreement is the diagnosis.

Ranked where the noise allows it

A break between bands opens only where the gap exceeds the noise measured at that point on the leaderboard, which is why the bands are uneven. Entities sharing a band are not ordered, because their order would not survive a re-run.

Entity Mention Frequency (EMF)

Topical Range Score

Position Power Score

First-Position Rate

Cross-Platform Stability

Blended summary

A single weighted number over all five dimensions (w_p 0.32 / w_g 0.68). It is a summary and not the headline: blending one reliable dimension with four less reliable ones moves the result toward the less reliable ones, so the per-dimension standings above carry the ordering. Shown because it is the number this index has always published. Its own noise floor is 0.0432, which separates the 18 tracked schools into just 4 bands: 14 of the 17 gaps between neighbours are inside it. The rank column is a reading order, not a finding. Rows sharing a band are marked and are not distinguishable from each other. Tap a row to focus it across the whole index. The chip shows recommendation quality: how AI frames the school when it names it, a top pick (green) versus hedged with caveats (red). Hover for the recommended / neutral / cautioned split.

How this index is scored

Two layers, scored on the same five signals. The layers are what AI knows about a school without looking versus what it finds when it searches. The signals are how each layer is measured.

Head-to-head, honestly measured

When AI compares two schools, the naive answer is mostly an artifact of which is named first. We ask every matchup both ways; the number to trust is order-balanced. Toggle to watch the correction.

Attribute association

Which factors the models associate with each school. Darker cell = stronger association. Each row is a share of that school's own mentions, so a school needs at least 20 mentions before its shares mean anything.

Search fan-out

One question quietly becomes many. These are the hidden sub-searches the models run to answer a category question, the content brief they wrote themselves: rankings and career outcomes dominate.

AI citations

Citation rate and the domains AI leans on.

Citation flow

Read left to right: the kind of question on the left, the source AI cited on the right. Aggregators and a broad long tail dominate; school sites are a thin slice.

Citation sources & page types

Of the school pages AI does cite, the mix concentrates in program (37%) and outcomes (20%). The chart below sets that against where citations go overall, across 7,509 cited URLs.

Google AI Overviews

An AI Overview appears on 81% of the 772 category keywords, the highest-volume AI surface. Even Google's own AI leans on third parties over the schools: schools are named mostly on their own brand terms, and on non-branded discovery queries the naming concentrates on a few. The count below is how many keywords return an AI Overview naming that school, out of 772. It is not a count of citation instances: one Overview naming a school twice still counts once. Tap to focus.

Organic share of voice

Share of estimated addressable organic clicks. The 16 highest shares of 18 measured entities are shown; the totals below the table cover all 18.

SERP and keyword coverage

Keyword-universe SERP appearances per school.

"not ranked" means the school ranks for none of the tracked keywords, so no composite is computed. That is absence from this keyword universe, not a score of zero.

Category concentration and saturation

How many cohort SERP appearances each question category actually has, and how concentrated they are where there are enough to tell. A concentration index over a handful of appearances reports the reciprocal of that handful, not the shape of the category, so it is withheld under 30.

Read the appearance counts first. 776 of 824 cohort SERP appearances land in the residual bucket rather than any named question category, so the keyword universe barely reaches the categories the AI prompt set asks about. That gap is the finding here; a concentration index computed over the handful that do land in a named category would only report how few there are.

White space

High-volume query clusters where no cohort school has built a strong position.

AEO vs SEO

AI standing is the measured presence band on live retrieval, so schools the instrument cannot separate share a band instead of being put in an order. Organic share is of 457,848 addressable monthly clicks across 772 volume-bearing keywords. No AI-versus-organic verdict is published for this category: the whole tracked cohort captures 9.1% of that market, so there is no organic contest among these schools for an AI standing to be ahead of or behind. The share each one holds is shown; the verdict column is withheld.

Entity explorer

Per-school two-layer and organic detail. Tap any row to open the five signals behind its memory and retrieval scores.

Freshness of cited pages

Not measured for this index. Tracked schools do own pages AI cites here, but none is cited often enough to enter the most-cited set this check reads dates from, so the figure is withheld rather than reported as an absence.

Not measured, and the reason is the sampling. Tracked schools do own pages AI cites in this category: 238 citations across 17 of them, led by UC Berkeley Haas (35), Harvard Business School (34), Chicago Booth (29). Of the 5 genuinely cited pages checked (the fetch covered 40 URLs, but 35 of those are domain roots synthesized from a cited hostname rather than pages anything cited), none belongs to a tracked school. This check reads the last-modified date of the category's MOST-cited pages, and no single school page is cited often enough to enter that set, so none has been dated. Freshness is measurable for this cohort; it has not been measured here, and the figure is withheld rather than reported as an absence.

Citability infrastructure

Whether each school has the machine-readable infrastructure AI leans on to recognize and cite an entity: structured data, an encyclopedia presence, an AI-crawler guide, and named authors. Two of those checks are stable enough to report and are shown as counts; no composite score is published, because three of the five are not reliable enough to stand behind. The reasons are below.

Two of the five checks stand up; the score does not. A citability score needs at least 3 checks that can be stood behind and only 2 of 5 qualify here (schema_org, llms_txt), so what those 2 measured is published below and no composite is. Counted over the 14 of 18 tracked schools whose homepage the check could reach; the other 4 are excluded rather than counted as absences.

The other 3 are withheld:

• wikipedia: answer changed for 8 of 18 entities between the two most recent runs
• wikidata: answer changed for 6 of 18 entities between the two most recent runs
• author pages: identical for all 18 entities, so it carries no information

A failed or rate-limited lookup is currently recorded as an absence, so scoring these would report that named organizations have no encyclopedia entry. The rebuild resolves each entity to a stable identifier first and reports three states, present, absent and unverified, rather than scoring unknowns as zeros.

Category findings

6 findings this index establishes about the category, each one carrying the measurement from the panel that produced it. These describe the category, not a single school: the page-level, prioritised plan for one organization is a GEO Audit, which is a separate engagement.

These are findings about the category, measured on the whole tracked cohort. This index does not tell any one school what to change: the page-level plan for a single organization (the specific URLs, entities, schema and crawler fixes, sequenced with owners) is a GEO Audit, a separate engagement.