Somewhere on your marketing team’s to-do list is probably a ticket that says “add llms.txt.” Maybe an agency pitched it. Maybe a developer read about it on Hacker News. Maybe a competitor announced theirs on LinkedIn like it was a product launch. Before your team spends an engineering sprint on it, it’s worth asking a question most vendors selling the tactic will not ask for you: does building one actually change whether ChatGPT, Perplexity, or Gemini cite your SaaS product?
The honest answer, based on the data available in mid-2026, is: probably not by itself, and not for the reason most teams assume. This piece walks through what an llms.txt file actually is, what the current evidence shows about whether AI systems read it, and a four-part test you can run in under a day to decide if your SaaS company should bother.
What an llms.txt File Actually Is
The llms.txt proposal comes from Jeremy Howard, co-founder of Answer.AI and fast.ai, who published the original specification in September 2024. The idea is simple: a plain markdown file at your site’s root, written for language models instead of browsers, that gives a short summary of your product plus curated links to the pages you most want an AI system to read. Per the published specification, the format is deliberately minimal: an H1 heading, a one-line blockquote summary, and H2-delimited sections listing the URLs you consider canonical.
It was built to solve a narrow, real problem: language models have limited context windows, and turning a marketing-heavy HTML page into clean text is messy. A curated file sidesteps that. It was never pitched by its creator as an SEO ranking signal or a guaranteed citation lever. That distinction matters, because most of the content telling SaaS marketers to build one treats it as the second thing.
What the 2026 Data Actually Shows
Three independent data points, from three different organizations, point the same direction.
Google has explicitly declined to use it. At a Search Central event in the Asia-Pacific region in July 2025, Google’s Gary Illyes stated plainly that Google does not support llms.txt and has no plans to. As reported by Search Engine Land, his guidance for AI Overviews visibility was to focus on standard SEO fundamentals, since AI Overviews draw on the same index and ranking systems as regular search.
Almost nothing requests the file once it’s live. Ahrefs analyzed 137,210 domains that received traffic in May 2026 and checked each for a valid, live llms.txt file. Roughly 28% of those domains had one published. Of the domains that published a file, 97% received zero requests to it that month. Of the small slice of files that did get traffic, retrieval bots (the category of crawler that would actually pull content into an AI answer) accounted for just 1.1% of requests. SEO audit tools, unidentified bots, and general web crawlers made up most of the traffic that did occur.
Where gains showed up, the file wasn’t the cause. Search Engine Land tracked ten sites across five industries for 90 days before and after they published an llms.txt file. Two sites saw AI referral traffic increase (12.5% and 25%), but both increases traced back to something else happening at the same time, like new downloadable templates or a product page rebuild with extractable tables, not the file itself. Eight of the ten sites saw no measurable change.
None of this means llms.txt is useless. It means it is not currently a meaningful lever for AI search visibility on its own, and treating it as one is how marketing budget gets spent on the wrong fix.
What Ahrefs’ Bot Traffic Data Actually Shows
| Requester category | Share of requests to accessed files | What it means for SaaS teams |
|---|---|---|
| SEO audit tools | 21.7% | Other agencies and rank trackers checking whether you have one, not AI systems reading your content |
| Unidentified bots | 14.9% | Unclassified traffic, can’t be attributed to AI retrieval |
| General web crawlers | 13.1% | Standard search engine crawling, not AI-specific |
| Tech profiling tools | 11.6% | Stack-detection tools like BuiltWith, unrelated to AI citation |
| AI agents / coding infrastructure | 10.5% | Developer tools like IDE agents, the one category where the file has a plausible real use |
| AI training crawlers | 5.3% | Feeds future model training, not today’s citations |
| AI assistants | 2.5% | Tools like ChatGPT plugins fetching a page on request, not organic discovery |
| AI retrieval bots | 1.1% | The category that would actually generate a citation in an AI answer |
Source: Ahrefs, “We Analyzed 137K Sites” study, based on May 2026 server log data from sites with a live, requested llms.txt file.
The 4-Signal SaaS Fit Test
Instead of a generic “how to write an llms.txt file” checklist, run this test first. It’s built directly from the gap in the data above: the file has a plausible use case (developer and agent tooling) and a mostly unproven one (AI search citation). The test forces you to figure out which one applies to your site before you spend engineering time.
Walking Through Each Signal
Signal 1: Check what your traffic already looks like
Open GA4, segment organic sessions by landing page path, and separate anything under /docs/, /api/, or /changelog/ from everything else. If your SaaS product is developer-facing and a meaningful share of organic traffic already goes to technical pages, an llms.txt file scoped to that content has a real, if modest, use case: helping coding agents and IDE tools pull accurate API context. If your traffic is almost entirely marketing and blog pages, the file’s most-supported use case doesn’t apply to you yet.
Signal 2: Look at your logs, not your assumptions
This is the step most teams skip because it takes more than five minutes. Server-side log file analysis tells you what’s actually crawling your site today, as opposed to what a vendor’s dashboard estimates. Filter for known AI agent user agents and see whether they’re requesting your existing pages at all. A site with zero AI crawler visits in its logs has no evidence that publishing a new file will suddenly attract them.
Illustrative example: a 30-day log filter
Illustrative example using placeholder figures, not data from any real client account. Built to show the kind of filter to run, not an actual log export.
Signal 3: If you build it, scope it narrow
The evidence points to developer and agent tooling as the only category with a defensible use case right now. That means the highest-value version of an llms.txt file for most SaaS companies links to API reference, integration guides, and changelog entries, written in plain, structural language, not marketing copy. This is a different exercise than the technical foundation that does have evidence behind it: crawlable pages, valid structured data and entity markup, and a site architecture search engines and AI crawlers can actually parse. If your technical SEO foundation isn’t solid, that’s a higher-priority fix than a new markdown file.
Signal 4: Measure before you scale it up
If you do publish a file, don’t treat it as done. Set a log-based measurement window, 90 days is a reasonable baseline given the Search Engine Land study design, and track AI agent requests to the file plus any change in AI-referred sessions in GA4. If nothing moves, that’s a valid result. It tells you not to invest further in llms-full.txt, automated generation pipelines, or expanding the file’s scope, and to put that engineering time into work with stronger evidence behind it.
FAQ
Does Google use llms.txt for AI Overviews?
No. Google’s Gary Illyes stated directly in July 2025 that Google does not support llms.txt and has no plans to, and that standard SEO practices are what determine AI Overviews visibility, since AI Overviews draws on the same index as regular search.
Should a SaaS company create an llms.txt file?
It depends on your traffic mix. If a meaningful share of your organic traffic already goes to developer-facing pages like docs and API reference, a narrowly scoped file has a plausible, if modest, benefit for coding agents and IDE tools. If your site is primarily marketing content, current data doesn’t support it as an AI search visibility tactic on its own.
Do ChatGPT, Perplexity, and Gemini actually read llms.txt files?
Based on Ahrefs’ analysis of 137,210 domains, retrieval bots, the category that would pull content into an AI answer, accounted for just 1.1% of the requests made to llms.txt files that received any traffic at all. The vast majority of files got zero requests in the study period.
What should go in a SaaS llms.txt file if I build one?
Scope it to what the evidence supports: API reference, integration and setup guides, and changelog entries, written as plain summaries rather than marketing language. Keep it to a curated set of your most important pages rather than a full site export.
If your technical foundation, crawlability, structured data, and site architecture, isn’t where it needs to be, that’s the higher-leverage fix before a new file format. Our technical SEO for SaaS work starts with exactly this kind of log-based audit. If you want a second set of eyes on where your engineering time is best spent, book a strategy call.
