The State of AEO 2026: What 13,385 Sites Tell Us About AI Citation Readiness

We have now scored 13,385 unique domains on AEO Rank and published 14,701 audits as of 2026-05-29. The picture that emerges is not subtle. The median AEO Rank score is 47. The mean is 44.5. Only 32 sites in the entire corpus clear a 70, and just 2 reach 80. This is the most complete portrait of AI citation readiness on the public web in 2026, and it has clear implications for any team that wants ChatGPT, Claude, or Perplexity to cite them.

The headline distribution: a market clustered in the 30s and 40s

Across 13,385 domains, AEO Rank scores follow a tight, low-centered distribution. The 25th percentile is 38, the median is 47, the 75th percentile is 53, and only the top 10% (the p90) clears 58. The minimum we observed is 2; the maximum is 92, scored by aeocontent.ai under our ChatGPT engine evaluation on 2026-02-22.

That distribution tells us something important: the AI-citation-readiness curve is bunched, not bimodal. Most sites are not catastrophically broken, and most sites are not winning. They are sitting in a 38-to-53 band where they have some structured data, some Q&A content, and some authority signals, but no one element is strong enough to win a citation against a well-optimized peer.

Score bracket	Sites in corpus	Share of corpus
Under 30	2,232	16.7%
30 to 49	6,849	51.2%
50 to 69	5,588	41.7%
70 to 79	30	0.22%
80 to 89	1	0.007%
90 and above	1	0.007%

The math on the top tier is brutal. Out of 13,385 sites, only 32 are above the 70-point line that we treat as “AI-citation-ready.” That is 0.24% of the public web we have audited. If you score a 60 today, you are already in the top decile of the corpus.

What separates the top 32 from the median 47?

We pulled the 32 sites at 70+ and compared them feature-by-feature against the corpus median. The top-tier cohort is not winning on raw word count or domain age. They are winning on a small set of structural signals that AI engines weight heavily. Named authors are present on the article level. An llms.txt file exists at the domain root. Schema coverage extends beyond Organization markup into Article, FAQPage, and HowTo entities. Comparison tables and direct-answer paragraphs appear early on key pages. And Trust & Authority signals - visible dates, author bios, citation lists - are consistently present rather than absent.

Two examples from our own work make this concrete. aeocontent.ai sits at 92 on the ChatGPT engine and helpsquad.com sits at 82 on the instant engine. Both publish dense Q&A pages, both expose an llms.txt, and both stamp visible date signals on every article. They are not unusual sites; they are sites where the AEO Rank criteria have been treated as a build list, not a checklist.

The five-pillar view: where the corpus is bleeding points

AEO Rank is built on five pillars, each scored 0 to 10. When we average pillar performance across the entire corpus, a clear hierarchy emerges. Answer Readiness leads at 5.7, followed by Technical Foundation at 5.3. The middle is Content Structure at 4.8. The bottom two are AI Discovery at 4.1 and Trust & Authority at 3.7.

Pillar	Corpus weighted average	Gap to ceiling
Answer Readiness	5.7	-4.3
Technical Foundation	5.3	-4.7
Content Structure	4.8	-5.2
AI Discovery	4.1	-5.9
Trust & Authority	3.7	-6.3

Trust & Authority being the weakest pillar is the single most actionable finding in this analysis. It means most sites have technically valid pages with reasonable answer copy, but they have not signaled to AI engines who wrote the content, when it was last updated, or what data backs it up. Those are exactly the signals that determine whether an engine cites a page or quietly ignores it.

The criterion-level disaster: Speakable Schema and Content Licensing

Two criteria fail almost universally. Speakable Schema, which tells voice and AI assistants which sentences to read aloud, is failing on 99.4% of sites in the corpus. The median score is zero out of ten and the mean is 0.1. Content Licensing & AI Permissions, which expresses whether AI crawlers may use your content, fails on 94.3% of sites with a mean of 0.7. These are not edge cases. These are nearly every site on the public web missing two foundational AEO signals.

When we plot the eight weakest criteria across the corpus, the picture is consistent: the cheapest wins on AEO Rank are also the most ignored. Adding a 200-byte llms.txt, exposing a Speakable Schema annotation, and publishing a clear AI licensing statement are sub-hour engineering tasks that move 90% of sites measurably up the distribution.

How We Tested

The figures in this article come from the AEO Content audit corpus as of 2026-05-29. The corpus covers 13,385 unique domains that have been scored on AEO Rank, with 14,701 total audits published over the lifetime of the platform. Scores are produced by the AEO Rank engine (packages/aeorank-engine) using the v5.0 scoring schema: 48 criteria across five pillars, each criterion scored 0 to 10, weighted into pillar averages and a final overall score capped by coherence and duplication gates.

Audits run in three modes: the instant engine (HTML-only, no LLM), the ChatGPT engine, and the Claude engine. For corpus-wide statistics we use the most recent published audit per domain regardless of engine, which matches how the public report cards are displayed. Bracket counts are computed from aeo_audit_versions.overall_score and joined to the canonical domain row. Pillar averages are weighted by criterion weight, not unweighted averages of raw scores, which is why a pillar can average 3.7 even when individual criteria sit higher.

The top-32 cohort was assembled by taking every domain with a most-recent score of 70 or higher and de-duplicating by registered domain. We then cross-referenced llms.txt presence, named-author presence, and schema coverage manually for each site in May 2026.

Where to take this next

If you are below 50 on AEO Rank today, you are in the densest band of the corpus and the next 10 points are the cheapest you will ever buy. Run a fresh audit on your domain to see exactly which of the 53 criteria are pulling your score down and which fixes will move you above the p90 of 58. Start with the AEO Rank methodology for context, then run a site audit to get your own report card with the same engine that produced these corpus numbers.