The llms.txt Index: how 187 sites score
We archived the llms.txt files of 187 major websites and scored every one with the same engine that grades a free scan (rubric v2.1, judged against llms.txt v2 (2026-08-10)). The mean score is 59.2 and the median 59.9; most files are present but underpowered. These are aggregate figures: pass rates, grades and averages, with sites named only as good examples.
At a glance
| Measure | Value |
|---|---|
| Sites | 187 |
| Files captured through | 2026-10-10 |
| Rubric | v2.1 (llms.txt v2 (2026-08-10)) |
| Mean score | 59.2 |
| Median score | 59.9 |
| Grade | Sites | Share |
|---|---|---|
| A | 0 | 0% |
| B | 1 | 1% |
| C | 16 | 9% |
| D | 76 | 41% |
| F | 94 | 50% |
The checks most sites miss
Measured audits that apply to at least 20 sites, lowest share of full marks first. Each is a concrete fix; the audit page says how.
- Entity identity: 0% of the 187 sites it applies to earn full marks.
- Link relations: 0% of the 142 sites it applies to earn full marks.
- Intent routing: 1% of the 167 sites it applies to earn full marks.
- API error handling: 1% of the 83 sites it applies to earn full marks.
- Data-freshness tiers: 3% of the 81 sites it applies to earn full marks.
- FAQ blocks: 4% of the 187 sites it applies to earn full marks.
What most sites already get right
- Encoding and parsing: 100% of the 187 sites it applies to earn full marks.
- Single H1: 91% of the 187 sites it applies to earn full marks.
- H2 sections: 89% of the 187 sites it applies to earn full marks.
- Canonical agreement: 88% of the 102 sites it applies to earn full marks.
- Curation, not a sitemap: 88% of the 99 sites it applies to earn full marks.
Every audit
“Applies to” counts the sites where the audit was scored; the rest were Not applicable (the audit doesn’t fit the site) or Not verified (the scan couldn’t measure it). Full marks and mean score are over the sites it applies to.
Retrievability & Access
| Audit | Applies to | Full marks | Mean score | Good examples |
|---|---|---|---|---|
| Covering llms.txt | 187 of 187 | not measured precisely yet | — | — |
| Clean fetch | 187 of 187 | not measured precisely yet | — | — |
| Content type | 187 of 187 | not measured precisely yet | — | — |
| Encoding and parsing | 187 of 187 | 100% | 1.00 | netlify.com, aflac.com, taylormorrison.com |
| AI crawler access | 183 of 187 | 20% | 0.59 | netlify.com, taylormorrison.com, expedia.com |
Spec Conformance
| Audit | Applies to | Full marks | Mean score | Good examples |
|---|---|---|---|---|
| Single H1 | 187 of 187 | 91% | 0.94 | netlify.com, aflac.com, expedia.com |
| Summary blockquote | 187 of 187 | 77% | 0.77 | netlify.com, aflac.com, taylormorrison.com |
| H2 sections | 187 of 187 | 89% | 0.94 | netlify.com, aflac.com, taylormorrison.com |
| Link line format | 174 of 187 | 54% | 0.75 | netlify.com, expedia.com, tavily.com |
Curation Quality
| Audit | Applies to | Full marks | Mean score | Good examples |
|---|---|---|---|---|
| Signal-to-noise | 174 of 187 | 65% | 0.88 | netlify.com, expedia.com, cloudflare.com |
| Link descriptions | 174 of 187 | 29% | 0.76 | cloudflare.com, attio.com, jasper.ai |
| Ordering | 184 of 187 | 86% | 0.93 | netlify.com, aflac.com, taylormorrison.com |
| Token budget | 187 of 187 | 80% | 0.92 | netlify.com, aflac.com, taylormorrison.com |
| URL hygiene | 174 of 187 | 67% | 0.97 | netlify.com, expedia.com, tavily.com |
| Curation, not a sitemap | 99 of 187 | 88% | 0.90 | netlify.com, aflac.com, www.khov.com |
| Freshness signals | 114 of 187 | 7% | 0.30 | qdrant.tech, intel.com, wordpress.com |
| LLM-friendly links | 174 of 187 | 12% | 0.31 | netlify.com, docs.expo.dev, twilio.com |
Agent Readiness
| Audit | Applies to | Full marks | Mean score | Good examples |
|---|---|---|---|---|
| Role and scope | 187 of 187 | 22% | 0.41 | netlify.com, aflac.com, taylormorrison.com |
| Data-freshness tiers | 81 of 187 | 3% | 0.12 | www.khov.com, khov.com |
| Intent routing | 167 of 187 | 1% | 0.10 | www.khov.com, khov.com |
| Journey flows | 40 of 187 | 5% | 0.19 | www.khov.com, khov.com |
| Machine surface | 83 of 187 | 55% | 0.87 | tavily.com, you.com, cloudflare.com |
| Action endpoints | 103 of 187 | 10% | 0.55 | aflac.com, expedia.com, www.khov.com |
| FAQ blocks | 187 of 187 | 4% | 0.23 | www.khov.com, khov.com, docs.apollo.io |
| Entity identity | 187 of 187 | 0% | 0.12 | — |
| API error handling | 83 of 187 | 1% | 0.25 | developers.deepl.com |
Compliance & Trust
| Audit | Applies to | Full marks | Mean score | Good examples |
|---|---|---|---|---|
| Regulatory guardrails | 0 of 187 | not measured yet | — | — |
| Prohibited claims | 0 of 187 | not measured yet | — | — |
| Required disclosures | 0 of 187 | not measured yet | — | — |
| Human handoff | 0 of 187 | not measured yet | — | — |
| Disclaimer text | 0 of 187 | not measured yet | — | — |
| Citation and licensing | 0 of 187 | not measured yet | — | — |
Ecosystem Integration
| Audit | Applies to | Full marks | Mean score | Good examples |
|---|---|---|---|---|
| Named AI crawlers | 183 of 187 | 20% | 0.59 | netlify.com, taylormorrison.com, expedia.com |
| Sitemap freshness | 129 of 187 | 48% | 0.84 | aflac.com, taylormorrison.com, www.khov.com |
| Markdown versions | 122 of 187 | 34% | 0.39 | netlify.com, you.com, cloudflare.com |
| JSON-LD agreement | 122 of 187 | 75% | 0.81 | netlify.com, aflac.com, taylormorrison.com |
| Canonical agreement | 102 of 187 | 88% | 0.94 | netlify.com, www.khov.com, you.com |
| Agent files | 187 of 187 | 6% | 0.07 | zapier.com, apify.com, developers.deepgram.com |
| Link relations | 142 of 187 | 0% | 0.13 | — |
How the Index is built
- We checked 723 domains in July 2026 — the top 100 US websites, 400 more of the Fortune 500, the top 100 global brands and the top 100 AI companies — and re-checked every one of them in October. July found 158 files; October found 182. We archive the raw bytes of each, so every figure here can be re-derived.
- Each archived file is scored by the current engine (rubric v2.1); when the rubric changes, the Index is re-scored, so these figures match what a scan would say today about the archived file.
- An audit that doesn’t apply to a site is left out of that site’s figures, and so is one the capture couldn’t measure. Audits scored by a fixed value today are marked “not measured precisely yet” rather than reported as a pass rate.
- Regulatory guardrails, Prohibited claims, Required disclosures, Human handoff, Disclaimer text, Citation and licensing need page captures the archive doesn’t have yet, so the Index has no figures for them.
- Sites are named only as good examples: sites that earn full marks on that audit. We don’t publish per-site scores here.
The same figures are available as JSON at /research/stats.json. Cite them as: LLMSRank llms.txt Index, rubric v2.1, files captured through 2026-10-10.
Frequently asked questions
How many sites are in the llms.txt Index?
187: the llms.txt files we have found and archived across our sweeps of the same 723 major domains, first in July 2026 and again in October. Files are captured through 2026-10-10.
What do most llms.txt files get wrong?
By share of full marks, the most-missed measured checks are Entity identity (0%), Link relations (0%), Intent routing (1%). Each audit page explains the fix.
How do these scores compare with a scan of my own site?
They come from the same engine and rubric, applied to archived files. A live scan also fetches robots.txt, your sitemap and pages at scan time, so it can measure a few things the archive can’t.
Why do some audits have no figures?
Some audits need captures the archive doesn’t have yet (page markup and link relations), and a few are scored by a fixed value until a precise measurement ships. We mark both rather than report a misleading pass rate.
Can I get the data?
The aggregate figures on this page are published as JSON at /research/stats.json. Per-site scores are not published; scan any site for free to see its full report.