The Generative Engine Optimization scorecard

Survey findings

Domains measured

75

Scored (retrievable GEO score)

75

Blocked (no retrievable GEO score)

0

Median GEO (over scored sites)

43

Median SEO (over scored sites)

75

Median ASQ™ (computed GEO+SEO composite, over scored sites)

62

Key Findings

Most leading sites have won classic search and have not yet been built for AI answer generation.

This is a read of where 100 well-known sites stand on Generative Engine Optimization — how retrievable and citable their pages are to AI answer engines — measured alongside their classic SEO. We report what the data supports and flag what it does not.

The SEO–GEO gap is the shape of the cohort

Classic-search strength does not carry over to AI answer generation — and for most of these sites it has not. The gap is structural, not a handful of outliers:

• 52 of 75 sites (69%) score at least 20 points higher on SEO than GEO.
• 32 of 75 (43%) show a gap of 40+ points.
• 33 of 75 (44%) sit at SEO ≥ 70 while GEO < 50.
• Average per-site SEO−GEO gap: +31.7 points.

GEO clusters low; big brands crater

GEO is thin at the top: only 3 of 75 sites clear the Pass threshold of 85 (top10lists.us 92, supabase.com 86, cloudflare.com 85), while 43% sit below 40.

The surprise is the size of the brands at the bottom. Amazon (11), Zillow (13), IMDb (16), Facebook (20), ESPN (11.5) and the FDA (3) all hold SEO 56–74 but GEO ≤ 20 — household-name domains whose substantive content is rendered client-side, gated, or buried under interactive UI rather than served as clean, citable text.

A universal structural gap — the text is there, the trust layer is not

The sub-metrics point to one consistent failure mode across the cohort:

• Citation (CITE) = 0 across all 75 scored sites — no machine-detectable citations, mentions, or isBasedOn in the served markup.
• Supported Grounding Rate (SGR): median 0 (n=55) — claims rarely backed by a verifiable source.
• Schema: median 0; 79% of sites carry ≤1 JSON-LD block. Structured data is near-absent.
• Retrievable Ratio (RR): median 0.89 — the one healthy signal. The served text is usable; what's missing is the citation, grounding, and schema structure layered on top of it.

A quarter of the cohort blocked the fetch

25 of 100 sites returned no retrievable score. That is itself a finding: these are disproportionately sites that block, gate, or paywall automated access — listing aggregators (apartments.com, costar.com, rent.com, rentcafe.com), hard paywalls (WSJ, Bloomberg), login-walled platforms (Reddit, Glassdoor, Indeed), and AI vendors themselves (OpenAI, Perplexity).

We keep "unscored" distinct from "scored low" — but for AI answerability, a page an engine cannot fetch and a page it fetches and cannot cite from land in similar practical territory.

The cohort shows a real, measurable separation between being findable in classic search and being usable by an AI answer engine. The content is largely retrievable; the machine-readable trust layer on top of it is mostly not there yet. That gap is the work AnswerShare addresses — making a site's existing, authoritative content retrievable, structured, and grounded for AI answer generation, without disturbing its classic-search standing.

Data notes: figures are computed over the 75 scored sites (not all 100). The NPS family (µNPS / λNPS / ΔNPS) is not yet measured cohort-wide and no conclusion here rests on it. ASQ is reported as the computed GEO+SEO value (round(0.438·GEO + 0.562·SEO)). The uniform CITE = 0 is read as a real coverage gap but is flagged pending a metric-firing sanity check before any client-facing use.

Per-domain GEO scorecard

Each scored domain is measured across the canonical Generative Engine Optimization metric set, with ASQ™ (the computed GEO+SEO composite) shown first, followed by GEO and SEO. The GEO column is the median across five frontier AI engines; where a per-engine split is available it is shown inline. A value of N/A means the metric was not computed for that domain; Blocked means no retrievable GEO score.