Last updated September 2026. Last checked: Sep. 3, 2026. Figures below carry their measurement date; revisit this page if a third adoption study lands.
The news: v2 shipped on Aug. 10, 2026
The llms.txt maintainers published version 2 of the spec on Aug. 10, 2026, at llmstxt.org, incorporating two years of implementer feedback. Refonte Learning and aeo.press both covered the release the same week.
The update lands at an odd moment for the format. Google has said, on the record, that its Search systems ignore the file, and a large-sample correlation test found no link between having one and getting cited. Meanwhile, two adoption studies published weeks apart in 2026 put the file’s real-world reach anywhere from under 10% to over half of the sites checked. A vendor or a tool that quotes a single “llms.txt adoption rate” without naming its sample is picking whichever number tells the better story.
Two studies, two numbers
Both studies below measured llms.txt adoption in 2026. Neither is wrong. They counted different populations.
| Detail | Rankability | Digital Applied |
|---|---|---|
| Headline adoption rate | 8.7% | 51.8% |
| Sample frame | Tranco top 1,000, the most-visited sites on the open web | Fixed 219-host panel, curated and weighted toward developer tools, SaaS, and AI companies |
| Hosts tested | 1,000 | 219 |
| Hosts reachable | 549 (54.9%) | 218 (99.5%) |
| Rate on reachable hosts only | 15.8% | 51.8% |
| Measurement date | June 2026 | Aug. 3, 2026 |
| What the rate describes | How common the file is across the general, traffic-ranked internet | An adoption ceiling among sites whose visitors already run coding agents daily |
The gap looks like a contradiction until you read the fine print. Rankability’s 8.7% headline divides confirmed adopters by all 1,000 sites in its list, including 451 that never responded to a crawl at all. Count only the 549 reachable roots, and the rate rises to 15.8%. Digital Applied’s panel, by contrast, was 99.5% reachable by design, because it hand-picked sites in sectors, developer tools, SaaS, and AI companies, chief among them, where llms.txt already has traction.
Compare the raw headlines, 51.8% against 8.7%, and the gap is close to sixfold. Compare reachable hosts only, 51.8% against 15.8%, and it shrinks to about 3.3 times. Both comparisons are legitimate. Neither is “the” adoption rate. The number that belongs in your report is the one whose sample matches the population you actually care about, stated with its date and its denominator attached.
What v2 actually changed
Version 2 is a maintenance update, not a rewrite. It formalizes standard link relations, rel="alternate" pointing to a page’s markdown mirror and rel="describedby" pointing to the llms.txt file that covers it, delivered as an HTML <link> element or an HTTP Link: header, so a server can announce both files without editing page content. It gives markdown mirrors two valid naming patterns instead of one, appending .md to a URL or swapping the extension for it. It clarifies that a file covers every page under its path and that the most specific file wins when more than one applies. And it downgrades the spec’s “Optional” heading from a rule tools were expected to enforce mechanically to a naming convention, since that is how implementers were already treating it. None of the four changes touch whether an AI engine reads or ranks the file at all.
The evidence hasn’t moved
A v2 spec bump does not change what the citation research already found. A correlation study across roughly 300,000 domains, published in November 2025, tested whether publishing llms.txt tracked with AI-citation frequency and found no measurable link; a gradient-boosted prediction model actually got more accurate once the llms.txt feature was dropped, the signature of noise, not signal. Google has said the same thing from the other direction: its Search documentation states that AI Overviews and Search rankings ignore the file, and Google’s John Mueller has compared it to the retired keywords meta tag on the record.
What this means for your GEO checklist
Don’t let “implement llms.txt” sit at the top of your generative engine optimization list. The spec got a real update; the case that it moves citation odds still is not there. Spend the engineering hour on crawler access, structured content, and page-level markdown clarity instead, and revisit llms.txt only if a future study breaks the current null result.
None of the three tools this directory tracks in this space treat llms.txt as a core lever, though each covers it. Semrush publishes a how-to guide and lets its Site Audit module confirm a file is being crawled correctly. Profound publishes its own explainer on the format’s role in agent-driven web interactions and serves an llms.txt file on its own site. Temso does not list llms.txt implementation as a required step in its own guidance, treating crawler accessibility and structured content as the higher-priority work.
If you track AI visibility and want a platform that timestamps what changed and when, instead of a single static score, compare options on the full AI SEO tools index, scored against the published methodology.
Sources
- llmstxt.org. “The /llms.txt file, v2.” Accessed Sep. 3, 2026.
- Refonte Learning. “Implementing llms.txt.” Accessed Sep. 3, 2026.
- aeo.press. “The State of llms.txt in 2026.” Accessed Sep. 3, 2026.
- Rankability. “LLMS.txt Adoption: 8.7% of the Top 1000.” June 2026.
- Digital Applied / llmtxt.info. “llms.txt in Practice: Adoption Data, Evidence, and Setup.” Aug. 7, 2026, citing a fixed 219-host panel measured Aug. 3, 2026.
- SE Ranking. “LLMs.txt: Why Brands Rely On It and Why It Doesn’t Work.” Correlation study across roughly 300,000 domains, published November 2025.