Between 0.005% and 28% of websites publish an llms.txt file in 2026, depending on the sample, but evidence that it works is thin. Ahrefs found 97% of 38,000 files received zero requests in May 2026, Google says Search ignores the file, and SE Ranking found no link to AI citations across 300,000 domains.
llms.txt is a plain Markdown file at the root of a website that lists the pages a language model should read first. Jeremy Howard proposed it on 3 September 2024. Two years on, it is one of the most heavily marketed tactics in AI search optimisation and one of the least supported by evidence.
This article compiles 18 sources: the proposal itself, eight adoption measurements, five studies or statements on whether AI systems read or reward the file, and the platform announcements from Google. I reconcile why adoption estimates differ by a factor of more than 5,000, report what Google, OpenAI and Anthropic have and have not said, set out my own reading, and finish with a template. If you are working on citations more broadly, start with how to get cited by Perplexity and Google AI Overviews and how to get your brand mentioned by ChatGPT.
#The numbers at a glance
Adoption estimates range from under 0.005% of all websites (Sistrix, Nov 2025) to 28% of 137,210 tracked domains (Ahrefs, May 2026). Large top-site scans in September 2026 cluster near 8% to 9%. Readership is far lower: 97% of Ahrefs' files got no requests at all.
| Measure | Figure | Source and date |
|---|---|---|
| All websites using the file | under 0.005% | Sistrix, 10 Nov 2025 |
| Sites in the HTTP Archive dataset | about 2% | Web Almanac 2025, reported 29 Jan 2026 |
| About 300,000 domains | 10.13% | SE Ranking, 20 Nov 2025 |
Common Crawl sampled /llms.txt URLs returning a text body |
11.72% | Common Crawl, July 2026 crawl |
| Top 10,000 domains (Tranco) | 8.3% | Rankability, 18 Sep 2026 |
| Top 1,000 domains (Tranco) | 9.3% | Rankability, 18 Sep 2026 |
| Top 100,000 domains (Cloudflare Radar list) | 8.6% of all, 11.4% of responding | Seomator, 13 Sep 2026 |
| 137,210 domains with traffic in Ahrefs Web Analytics | 28% | Ahrefs, 15 Jun 2026 |
| Files receiving any request in May 2026 | 3% (97% received none) | Ahrefs, 15 Jun 2026 |
The rest of this article explains why those figures are all defensible and none is "the" adoption rate.
#What llms.txt is, and what the spec requires
The proposal, published by Jeremy Howard on 3 September 2024 and last modified on 10 August 2026, defines the file as follows (llmstxt.org; original post on Answer.AI):
- The file is named
llms.txtand sits at the site root, or at a subpath such as/docs/llms.txt. A file covers the URLs under its path, and agents should use the most specific file that applies. - Section order: an optional byte-order mark, then an H1 with the project or site name (the only required element), then a blockquote summary, then zero or more detail sections without headings, then zero or more H2 sections containing lists of links.
- Each list entry is a Markdown list item with a required link and optional notes after a colon:
[name](url): notes. - An H2 section named Optional holds secondary links that an agent can skip when it needs a shorter context.
It is a convention, not a standard. No standards body ratified it, and it is not part of robots.txt or sitemap protocols. Its job is different from robots.txt: robots.txt says what crawlers may fetch, while llms.txt is meant to say what is worth reading. A companion file, llms-full.txt, holds the full text of the linked pages in one document.
#Who has adopted it
Eight measurements, eight populations
Sistrix, 10 November 2025. "Less than 0.005% of all websites worldwide are using the file." The post does not report how many domains were checked. It concludes that adoption is very low, that major providers do not use the file, that OpenAI has made no official statement, and that typical business sites have no urgent need (Sistrix, modified 20 Nov 2025).
SE Ranking, 20 November 2025. About 300,000 domains crawled, 10.13% with an llms.txt file (Search Engine Journal reporting on SE Ranking).
HTTP Archive Web Almanac 2025, reported 29 January 2026. About 2% of sites in the dataset. The Search Engine Journal write-up says much of the growth came from SEO tools adding llms.txt as a default or optional feature, and does not state the sample size (Search Engine Journal).
Ahrefs, 15 June 2026. 137,210 domains with traffic in Ahrefs Web Analytics in May 2026. 28% (about 38,000, or 38,360 in the chart) published an llms.txt that returned HTTP 200 and was confirmed as Markdown. Ahrefs says its customers skew technical and SEO-aware, so 28% is an upper bound (Ahrefs).
Common Crawl, July 2026 crawl. Of sampled /llms.txt URLs, 11.72% returned a text body, 22.32% returned HTTP 200 and 68.4% returned 404. Common Crawl notes this is well above the roughly 2% Web Almanac reported for 2025 but that populations differ. It analysed 584,107 files after removing 14,191 empty responses (Common Crawl).
Seomator, 13 September 2026. 100,000 domains from Cloudflare Radar's top list; 75,353 responded. 8.6% of all 100,000 (8,598 real files) and 11.4% of responders. A further 10,775 domains returned an HTML page with a 200 status at /llms.txt, so a status-code-only check overcounts by about 2.35 times (Seomator).
Rankability, 18 September 2026. Top 10,000 Tranco domains: 8.3% (830 confirmed adopters, with 4,211 domains classed unknown and kept in the denominator). Top 1,000: 9.3% (93). llms-full.txt appears on 1.2% of the top 10,000. The page itself says changes in sample and method prevent a like-for-like growth calculation (Rankability).
Digital Applied, 7 August 2026. A curated developer-weighted panel of 218 reachable hosts: 51.8% (113). Developer tools 42 of 61 (68.9%), SaaS 32 of 48 (66.7%), media 0 of 14. It warns against averaging figures from different sampling frames (Digital Applied, a secondary compilation I use only for this panel and for the log study described below).
Who is in the numbers
Adoption is lopsided by sector and by top-level domain.
- By category in Rankability's top 10,000 (directional): travel 16.0% (8 of 50), financial services 15.4% (12 of 78), music streaming 12.8%, reference 9.8%, e-commerce 7.4% (8 of 108), technology 6.4% (49 of 764), health and medical 6.3%, sports 5.7%.
- By TLD in Seomator's scan:
.ai39.3%,.io24.9%,.co20.4%,.au13.9%,.com13.4%,.uk9.4%,.de7.0%,.jp6.5%,.net5.5%. - Named adopters in secondary coverage: Anthropic, Cloudflare, Stripe and Cursor publish their own files, and Cloudflare, Azure, GitHub, Fastly, WordPress.org, Adobe and Samsung appear in a traffic-ranked sample (Digital Applied). Anthropic publishing its own file is not a statement about what its crawlers do.
Much of the file population is generated
- Wix accounts for 41.34% of the Common Crawl corpus, and 68.27% of all files are templated. The top ten skeletons cover 38.71%. 44.87% mention the Model Context Protocol, mostly through a line Wix inserts.
- In Seomator's scan at least 29.8% of files carry a platform signature: Shopify 18.2% and WordPress SEO plugins 9.6%.
- Common Crawl found 49.90% of files follow the full structure (H1, summary, link sections), 32.94% give link notes, 22.56% contain no links, 2.54% are domain-for-sale pages and 99.23% are Markdown.
- Seomator found 16.4% of files (1,407) contain no links; the median file has 15 links and 7 sections; 186 domains serve their robots.txt at
/llms.txt.
So a large share of "adoption" is a platform flipping a default, not a team deciding to write one.
#What Google, OpenAI and Anthropic have said
Google says its Search ignores llms.txt but that creating one for other services is fine. OpenAI and Anthropic have published no on-record statement that their crawlers fetch or parse it. Perplexity has published none either, and Sistrix noted in November 2025 that OpenAI had made no official statement.
| Date | Statement | Who |
|---|---|---|
| 17 Jun 2025 | "FWIW no AI system currently uses llms.txt." He added that server logs show consumer LLMs and chatbots fetch pages for training and grounding, but "none of them fetch the llms.txt file." | John Mueller, Bluesky, via Search Engine Roundtable |
| 15 May 2026 | Generative AI search does not need new "machine readable files, AI text files, markup, or Markdown." Google may crawl such a file, but "this doesn't mean that the file is treated in a special way." | Google AI optimization guide, as quoted by Frase; guide at Google Search Central |
| 5 May 2026 | Lighthouse added an llms.txt audit in its agentic browsing category | Reported by Digital Applied |
| 2026 (Lighthouse 13.3) | Lighthouse 13.3 audit: flags a server error retrieving the file, marks a 404 as not applicable because the file is optional "for now"; Chrome's docs say agents may spend more time crawling without it | TechWyse |
| 15 Jun 2026 | A self-reported manifest cannot differentiate sites, because "by design, can't trust what is here as a way of differentiating between different websites." Automated systems may help once someone is already on the site. | Mueller, as reported by Digital Applied |
| 17 Jun 2026 | "It's completely fine if you decide to create and maintain LLMS.txt files (or other similar files)." This does not affect Search rankings because "Google Search ignores them." The special-markup statement is limited to "Google Search (including its generative AI capabilities)". | Google guidance update, reported by Search Engine Journal |
The June 2026 change matters. The earlier wording could be read as saying no AI surface needs the file. Google narrowed it to say that Search does not use it, and removed the discouragement. Digital Applied also reports that Gary Illyes described the format as at best neutral, comparing it to the keywords meta tag, with no date given.
OpenAI, Anthropic, Perplexity
I could not find a primary, on-record statement from any of the three confirming or denying that their crawlers fetch or parse /llms.txt. Digital Applied reached the same finding, and Sistrix wrote in November 2025 that OpenAI had made no official statement. What exists is behaviour in logs:
- In Ahrefs' May 2026 data, GPTBot accounted for 4.51% of requests to llms.txt files and was the top AI bot, ClaudeBot for 0.8% and OAI-SearchBot for 0.74%. Slackbot fetched llms.txt more often than PerplexityBot (Ahrefs).
- In an April 2026 server-log study of about 900 domains between 4 September 2025 and 13 April 2026, 1,227 llms.txt-family requests were recorded and none came from GPTBot, ClaudeBot, PerplexityBot or Google-Extended. The largest requester was a commercial data aggregator with 794 requests (64.7%) (reported by Digital Applied).
Those two logs seem to conflict. They do not necessarily: they cover different sites and periods, and the second says it rules out identifiable lab crawls, not on-demand agent retrieval.
#What the evidence says about whether it works
Evidence from readership
Ahrefs' study is the most detailed. Its core numbers:
- 28% of 137,210 domains publish a file. 97% of those files received zero requests in May 2026. The remaining 3% (about 1,100 domains) received all the measured traffic.
- About 22,000 requests in total, of which 930 (4%) were human and 96% bots.
- Requests by category: SEO audit tools 21.7% (4,776), other or unidentified 14.9%, general web crawlers 13.1%, tech profiling tools 11.6%, AI agents and agentic infrastructure 10.5% (2,302), GEO and AEO tools 5.8%, AI training crawlers 5.3%, llms.txt discoverability bots 3.6%, service and social bots 2.9%, research bots 2.7%, AI assistants 2.5%, AI retrieval bots 1.1% (233).
- The AI categories together made 19.5% of requests. Googlebot made about 900 fetches, which Ahrefs attributes to normal crawling. Chrome Lighthouse made 22 requests, roughly one in 1,000.
- Requests to missing files were 98% human, and the AI bot share of requests to 404 versions was zero. In Ahrefs' words, "AI tools never go looking for llms.txt files that aren't there."
- Claude-Code fetched more llms.txt files than every AI retrieval bot, assistant and training crawler except GPTBot and one indexer, which points to coding agents as the realistic readers.
- Ahrefs' verdict: "If your goal is showing up in ChatGPT, Perplexity, or AI Overviews, an llms.txt file is largely decoration," and "the cons outweigh the pros right now."
Ahrefs lists its own caveats: its customers skew technical, the study did not check files against the spec, and "fetched" does not mean "read," so all consumption figures are ceilings.
Evidence from citation outcomes
SE Ranking's crawl of about 300,000 domains compared llms.txt presence (10.13%) with how often each domain was cited by major LLMs. It used correlation tests and an XGBoost model. It found no significant correlation, and removing llms.txt from the model improved the model's accuracy. The authors concluded it "doesn't seem to directly impact AI citation frequency. At least not yet." (Search Engine Journal, 20 Nov 2025).
I found no published controlled experiment in which a site added the file and showed a citation lift against a matched control. Absence of a study is not proof of no effect, but it is the current state of the evidence.
Evidence for the pro case
The case for the file rests on four points, none of them a measured citation gain:
- Coding agents read it. Claude-Code's position in Ahrefs' logs and the 68.9% adoption among developer tools in Digital Applied's panel line up. Documentation sites are where fetch-on-demand makes sense.
- It is cheap. It is one text file that you can write in an hour.
- Chrome is looking. A Lighthouse audit exists, with a not-applicable result when the file is absent "for now" (TechWyse).
- Google now says it is fine. The June 2026 update removed the discouragement.
Risks on the other side
- Prompt injection. Common Crawl found 3,793 files with mild steering language, 102 promotional, and 10 at the top severity tier, 4 of which it judged genuine. It also found 287 files containing ChatGPT citation markers and 1.43% showing traces of a chat window, which suggests some files are pasted model output. Ahrefs advises treating the file as code for the same reason.
- Policy language with no standing. 6.59% of Common Crawl files carry policy language (rate limits 3.05%, copyright 0.93%, citation demands 0.46%). Of 1,570 files naming a specific crawler, 32 denied CCBot, and none of those sites blocked CCBot in robots.txt. The file is not an enforcement mechanism.
- Maintenance drift. A manifest that lists dead pages misleads agents that do read it.
#What the data disagrees on
1. Why is adoption 0.005%, 2%, 10% or 28%?
Four effects stack up.
- Population. Sistrix measures all websites; most sites on the internet are small, unmaintained or non-commercial. Ahrefs measures domains that run Ahrefs Web Analytics, an SEO-minded group that Ahrefs says makes 28% an upper bound. Rankability and Seomator measure top-ranked domains, which are larger and more likely to sit on platforms with defaults.
- Detection method. Seomator found that counting any HTTP 200 overstates real files by about 2.35 times because soft 404s return HTML. Ahrefs and Common Crawl filtered for Markdown or text bodies; the studies that did not filter would run higher.
- Platforms. Wix is 41.34% of Common Crawl's files, and Shopify and WordPress plugins are a further 27.8% of Seomator's. A platform decision can add millions of files overnight.
- Time. Sistrix is November 2025. Web Almanac is data from 2025. The rest are May to September 2026. Rankability's top 1,000 moved from 0.3% (June 2025) to 8.7% (June 2026) to 9.3% (September 2026), but the page says the method changed, so no growth rate should be derived.
My conclusion is that "x% of websites have llms.txt" is not a meaningful sentence without a population. For top-ranked sites in 2026 the usable range is about 8% to 12%, which is where four independent measures converge (Rankability 8.3%, Seomator 8.6% of all and 11.4% of responders, Common Crawl 11.72%, SE Ranking 10.13%). For developer-tool sites, one panel measured 68.9%.
2. Does Google read it or not?
Mueller said no AI system uses it (June 2025) and the Web Almanac coverage says Google has said it does not plan to use it. Google then added a Lighthouse audit for the file (May 2026). Both are true. Googlebot fetching about 900 files in Ahrefs' logs is ordinary crawling. The Lighthouse audit is a Chrome tool for agentic browsing, a different product from Search. The June 2026 guidance resolves the apparent conflict: Search ignores the file, other Google-adjacent tools may look.
3. Do AI bots fetch it?
Ahrefs shows GPTBot, ClaudeBot and OAI-SearchBot all fetched files in May 2026. The April 2026 log study shows none of GPTBot, ClaudeBot, PerplexityBot or Google-Extended did over a seven-month window on about 900 domains. Differences in sample (137,210 against about 900), window (one month against seven) and date are enough to explain it. My own arithmetic, from Ahrefs' numbers: 4.51% of about 22,000 requests is roughly 990 GPTBot requests across about 38,000 files, and 22,000 requests across about 38,360 files is about 0.57 requests per file in a month, and since 97% of files had none, the average among the 3% that were requested is far higher.
4. Is the file harmless or risky?
Ahrefs says the cons outweigh the pros; Google says it is completely fine; Common Crawl documents injection attempts. These answer different questions. Google is saying it will not penalise you. Ahrefs is weighing effort against measured return. Common Crawl is describing attacker behaviour in the wild, where almost all injection samples were mild steering, not exploits.
#Shruti's analysis
This section is my interpretation. The arithmetic uses only figures cited above.
1. Treat llms.txt as documentation for agents, not as a ranking input. Every reader of the file that the evidence identifies is an agent acting for a user: Claude-Code, agentic infrastructure, SEO and GEO audit tools. None of the evidence shows a search index using it to decide what to cite. If your buyers reach you through a coding assistant, as with an API or developer product, it is a reasonable bet. If they reach you through ChatGPT recommending a vendor, it is not where citations are decided.
2. The "97% never read" figure is a call to action to measure, not a verdict on every site. Of Ahrefs' 38,000 published files, about 1,100 domains drew all the traffic. Those domains are probably not random. I would guess documentation and developer sites are over-represented there, since that matches Claude-Code's rank and Digital Applied's 68.9% developer-tool adoption, but Ahrefs does not break the 3% down by site type, so I mark this as inference. The test for your own site is cheap: search your server logs for /llms.txt after a few weeks.
3. The real return today is hygiene, not traffic. A good llms.txt forces you to decide which 10 to 30 pages best describe your company. That exercise is the same one that makes a good About page, a clear sitemap and a tidy documentation index. If publishing the file makes you fix a stale pricing page or an orphaned guide, that is worth more than any agent fetch.
4. The citation work is elsewhere. Ahrefs' 75,000-brand study found branded web mentions correlate with AI visibility at 0.664 to 0.709, against 0.266 for Domain Rating on ChatGPT. Reddit and YouTube together hold 42.4% of Perplexity's top 50 mention share. The pages that decide whether you are named are other people's pages. See Who Does AI Cite for B2B Buyer Questions, where in 50 Google AI Mode answers YouTube appeared in 48%, LinkedIn 44% and Reddit 30%, and Why Reddit matters for B2B.
5. Do not pay for it. At roughly a 0.57 requests-per-file monthly average, no study showing a citation effect, and platforms that generate files by default, there is no case for buying a standalone llms.txt service. Use your CMS's built-in option if it has one, or write it by hand in an hour.
6. What would change my mind. A controlled before-and-after test with matched sites, or a primary statement from OpenAI, Anthropic or Perplexity that a retrieval system reads the file. Either would show up as a change in the log data. Until then my position matches the weight of the sources: low cost, low evidence, optional.
For a check of what AI crawlers can already see on your site, run the free AI search readiness checker, and for a baseline of whether the engines name you, the AI visibility check.
#How to write an llms.txt file
If you decide to publish one, this takes about an hour.
Step 1. Choose 10 to 30 pages. Pick the pages you would send a new hire or a journalist: what you do, who it is for, pricing page (if public), documentation or guides, key research, contact. Skip tag pages and archives.
Step 2. Write the H1 and summary. The H1 is the only required element. Add a blockquote of one or two sentences saying what the company does and for whom, in plain terms.
Step 3. Group links under H2 headings. Use names a reader would recognise: Docs, Guides, Research, Company. Give each link a one-line note after a colon.
Step 4. Use an Optional section for the rest. Per the spec, agents may skip this section when they need a shorter context.
Step 5. Save it at the root and serve it correctly. Path /llms.txt, HTTP 200, text/plain or text/markdown. Make sure it is not an HTML page or a soft 404, the mistake Seomator found on 10,775 domains. Do not serve your robots.txt there; 186 domains did.
Step 6. Consider llms-full.txt. Only if you have documentation that fits in one document. Seomator found half of the 2,516 llms-full.txt files were within 5% of their llms.txt size, which suggests many are copies and add nothing.
Step 7. Review for injection and accuracy. Write every line yourself. Do not paste model output (Common Crawl found 287 files with ChatGPT citation markers). Do not include instructions to the model, promotional claims or policy demands; they have no standing and could be read as manipulation.
Step 8. Link it and log it. Reference it from your docs or footer, add it to your monitoring, and check server logs monthly for requests and user agents. If a month passes with no requests, you have your own answer.
A template
Replace the placeholders with your own details.
# Example Co
> Example Co makes scheduling software for dental clinics with 2 to 20 chairs. This file lists the pages that best describe the product, its documentation and its published research.
Example Co was founded in YEAR. The product is available in the US and Canada.
## Product
- [Product overview](https://example.com/product): What the software does and who it is for
- [Pricing](https://example.com/pricing): Plans and what each includes
- [Integrations](https://example.com/integrations): Systems it connects to
## Documentation
- [Getting started](https://example.com/docs/start): Setup in under an hour
- [API reference](https://example.com/docs/api): Endpoints, authentication, limits
## Research
- [2026 clinic scheduling survey](https://example.com/research/survey-2026): Method, sample size and results
## Optional
- [Company history](https://example.com/about): Founders and milestones
- [Press](https://example.com/press): Coverage and logos
#Decision guide
| Your situation | Publish? | Reason |
|---|---|---|
| Developer tool, API or documentation site | Yes | Coding agents are the best-evidenced readers (68.9% adoption in one dev-tool panel; Claude-Code in Ahrefs' logs) |
| B2B SaaS with a public help centre | Yes, cheap | Same logic, lower stakes; maintain it with the docs |
| Consultant, agency or personal site | Optional | No evidence of citation gain; your time is better spent on third-party mentions |
| Ecommerce on Shopify | Check first | Shopify may already generate a file (18.2% of Seomator's files); review it for accuracy |
| Publisher or media site | Low priority | Media was 0 of 14 in the developer-weighted panel and Google says Search ignores the file |
#Limits of this evidence
- Every adoption study uses a different population, so figures cannot be compared as a trend.
- Ahrefs, SE Ranking, Rankability and Seomator each sell or promote SEO tooling, and Ahrefs' sample is its own customers. I report their caveats where they state them.
- Log studies measure fetches, not use. "Fetched" is a ceiling on "read," as Ahrefs says.
- Some sources (Web Almanac, Mueller's 2026 remarks, the April 2026 log study, the Google guide's wording) reach me through secondary reports, labelled where used. I could not open the Google guide directly in this session.
- The evidence cannot rule out private use by AI vendors that leaves no trace in public logs.
#Methodology
Source selection. I looked for four kinds of evidence: the proposal and specification (llmstxt.org, Answer.AI), adoption measurements with a stated sample (Sistrix, SE Ranking, Web Almanac, Ahrefs, Common Crawl, Rankability, Seomator, Digital Applied), statements from platform owners (Google Search Central guidance, John Mueller, Lighthouse documentation), and outcome studies (SE Ranking, Ahrefs, the April 2026 log study). I included a source if it stated a sample or documented an official position, had a visible date, and could be traced to a page. I excluded marketing posts that quoted adoption percentages without a sample, and any claim I could not trace.
Date range. 3 September 2024 (the proposal) to 18 September 2026 (the latest scan), with Google statements from June 2025 to June 2026. The analytic weight sits on the 2026 sources. Pages were checked on 30 October 2026.
How I handled differences. Adoption percentages are never combined. Where I calculate something, I show the inputs, and I label it as my arithmetic. Where an article gave inconsistent figures I used the number in the primary table or the one the author highlights, and say so.
Statements I could not source. I found no on-record statement from OpenAI, Anthropic or Perplexity on whether their crawlers read llms.txt, so I report none. I make no claim about my own server logs, because I have not published them.
Conflicts of interest. I write about AI visibility and sell related services, which is a reason to prefer a low-claim conclusion here. Nothing in this article is a price, a client result or an estimate I have not sourced.
#Sources
- Jeremy Howard, "The /llms.txt file", proposed 3 Sep 2024, modified 10 Aug 2026. https://llmstxt.org/
- Jeremy Howard, "/llms.txt: a proposal to provide information to help LLMs use websites", Answer.AI, 3 Sep 2024. https://answer.ai/posts/2024-09-03-llmstxt
- Sistrix, "llms.txt: How Useful is the File for LLM Optimisation?", 10 Nov 2025, modified 20 Nov 2025. https://www.sistrix.com/?p=175210
- Search Engine Journal on SE Ranking, "LLMs.txt shows no clear effect on AI citations, based on 300k domains", 20 Nov 2025. https://www.searchenginejournal.com/llms-txt-shows-no-clear-effect-on-ai-citations-based-on-300k-domains/561542/
- Search Engine Journal, "What the latest Web Almanac report reveals about bots, CMS influence and llms.txt", 29 Jan 2026. https://www.searchenginejournal.com/what-the-latest-web-almanac-report-reveals-about-bots-cms-influence-llms-txt/565958
- John Mueller via Search Engine Roundtable, "Google: No AI System Currently Uses LLMs.txt", 18 Jun 2025. https://seroundtable.com/google-ai-llms-txt-39607.html
- Ahrefs, "We Analyzed 137K Sites: 97% of llms.txt Files Never Get Read", 15 Jun 2026, modified 27 Aug 2026. https://ahrefs.com/blog/llmstxt-study/
- Common Crawl, "A content analysis of llms.txt files from the July 2026 crawl archive". https://commoncrawl.org/blog/a-content-analysis-of-llms-txt-files-from-the-july-2026-crawl-archive
- Rankability, "llms.txt adoption", 18 Sep 2026. https://www.rankability.com/blog/llms-txt-adoption/
- Seomator, "llms.txt: We Scanned the Web's Top 100,000 Domains", 13 Sep 2026. https://seomator.com/blog/llms-txt-adoption
- Digital Applied, "llms.txt in practice: adoption evidence 2026", 7 Aug 2026. https://www.digitalapplied.com/blog/llms-txt-in-practice-adoption-evidence-2026
- Search Engine Journal, "Google says it's fine to use llms.txt for AI SEO", 17 Jun 2026. https://www.searchenginejournal.com/googles-says-its-fine-to-use-llms-txt-for-ai-seo/579608/
- Google Search Central, "AI optimization guide", 15 May 2026. https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
- Frase, "AEO and GEO Are Still SEO: What Google's Official AI Guide Actually Says", 7 Jun 2026. https://frase.io/blog/google-ai-optimization-guide
- TechWyse, "Google Publishes First Official AI Search Optimization Guide, Then Chrome Starts Auditing llms.txt", 2026. https://www.techwyse.com/news/ai-search/google-ai-search-optimization-guide-llms-txt-lighthouse-audit
- Ahrefs, "Top Brand Visibility Factors in ChatGPT, AI Mode, and AI Overviews (75k Brands Studied)", 12 Dec 2025. https://ahrefs.com/blog/ai-brand-visibility-correlations
- Ahrefs, "The 50 Most-Cited Websites in Perplexity", updated 2 Sep 2026. https://ahrefs.com/blog/most-cited-domains-perplexity/
- Shruti Kaushik, "Who Does AI Cite for B2B Buyer Questions? 50-Query Study", 17 Oct 2026. /writing/who-ai-cites-b2b-buyer-questions-2026