State of GEO: AI Search Readiness Benchmark

Across sites audited by GeoReady, the basics of AI search readiness (AI crawler access, an llms.txt, and structured data) are still missing more often than not. The live numbers below show the current average and median GEO score and adoption rates, updated as new audits run and counting the most recent audit per domain.

What this means

Among sites that bothered to measure, most are still leaving easy AI visibility points on the table. The gaps are rarely about content quality; they are about machine legibility. AI answer engines cannot cite what they cannot reach, parse, or trust, and the cheapest wins sit in those three layers.

Two signals do much of the work. An llms.txtfile orients AI tools toward your important pages, and validJSON-LD schema lets a model disambiguate your brand and lift clean answers. Across audited sites both are still underused, and both are among the cheapest, fastest fixes: a weekend of work, not a quarter.

AI discovery files are often among the weakest signals. Files like/.well-known/ai.txt and anllms.txt are trivial to publish yet rarely present, so a site that ships them moves ahead of much of the audited sample with very little effort. For marketing, SEO, and dev teams, AI search readiness is mostly hygiene you can act on this week.

How to improve your score

To improve a GEO score: allow AI crawlers intentionally, publish anllms.txt, addOrganization andWebSite JSON-LD, lead key pages with a direct answer, expose product and pricing facts in machine-readable form, and re-audit over time. None of it requires guessing at a model’s internals. It is the machine-readable groundwork that lets AI engines find, parse, and trust a page. Work top to bottom:

1. Allow AI crawlers intentionally. Audit yourrobots.txt so you are not silently blocking GPTBot,ClaudeBot, orPerplexityBot by accident.
2. Publish an llms.txt at your domain root that points to your most important pages. Build a starter with the freellms.txt generator.
3. Add structured data. Use validOrganization,WebSite, and where it genuinely fitsFAQPage orArticle JSON-LD, with consistent brand naming.
4. Improve answer-first content blocks. Open key pages with a direct, self-contained answer a model can lift cleanly, before the marketing context.
5. Expose product and pricing info in machine-readable form.Make the facts an AI needs to describe you available in plain HTML and schema, not buried in images or scripts.
6. Run recurring audits. AI behavior shifts as models update, so a one-time snapshot tells you where you are while monitoring tells you whether your changes worked.

Methodology

This benchmark is built from sites audited by GeoReady, scored with the open-source GEO Optimizer engine. The numbers are aggregated and anonymized: the public dataset is intentionally summarized, and the full internal analytics are not published. It is a directional signal across an audited sample, not a claim about the entire web.

• Audited sites only. This is a benchmark of sites audited by GeoReady, not a sample of the whole web.
• One row per domain. Only the most recent audit per domain counts, so re-audits don’t inflate the cohort.
• Scheduled monitoring excluded. Only user-initiated audits (web, CLI, API, tools) count.
• Aggregated & anonymized. Domains are stored as salted hashes; this page and its public endpoint expose aggregates only.
• Same engine for everyone. The open-source engine’s 100-point score. The current default rubric (v2, since engine 4.19.0) has nine categories: Google AI readiness, robots, schema, content, brand & entity, meta, llms.txt, signals, AI discovery. Audits run on older engine versions used the eight-category v1 rubric.
• Public data is summarized on purpose. Per-category breakdowns and exact distributions stay internal; the public figures are headline aggregates.

The signals behind every category are documented in ourGEO guideand research foundation. The dataset grows with every audit: run one and you are (anonymously) in it.

FAQ