I shocked a fellow agency owner last week.
They were fired up about the results they’d seen after adding llms.txt files to client sites. I laughed and told them the data shows no positive impact from adding it.
I like this agency owner. They do good work. But chasing the fad of the month wastes your time and energy at best, and at worst it hurts you. So before you spend another afternoon writing markdown versions of every page on your site, let’s look at what the research actually says about llms.txt.
If you’d rather watch than read, I walked through the same data here: The llms.txt File is a WASTE OF TIME (Here’s the Data) on the MarketingAlec YouTube channel.
The data: 300,000 domains don’t lie
SE Ranking analyzed nearly 300,000 domains to measure whether llms.txt does anything. The findings were brutal:
- Only 10.13% of sites have adopted it. Nine out of ten haven’t bothered.
- Zero correlation between having an llms.txt file and how often AI cites you.
- Pulling llms.txt out of their predictive model made the model more accurate.
Read that last one twice. The file made their predictions worse. That’s not “no benefit.” That’s noise.
AI doesn’t care about your llms.txt
Look at what the labs themselves say, and what they pointedly don’t:
OpenAI keeps its documentation on robots.txt controls and says nothing about llms.txt touching ranking or citations.
Google has been blunt about it. Per multiple reports, Google reps have “likened llms.txt to an early, unused idea…comparable to the old meta keywords tag in terms of current impact.”
Anthropic, the maker of Claude, publishes its own llms.txt file and still “haven’t confirmed their crawlers rely on it.” The company that helped popularize the format won’t vouch for it.
Perplexity went the other way entirely. Cloudflare research caught them “repeatedly modifying their user agent and changing IPs to hide crawling activity,” dodging robots.txt, never mind a voluntary standard like llms.txt.
The smoking gun: server logs
Here’s where it gets damning.
A Reddit analysis of 1,000 domains with llms.txt files went to the server logs to see which bots actually requested the file.
The result? No GPTBot. No ClaudeBot. No PerplexityBot. Instead, 95% of the requests came from Google’s traditional desktop crawler.
Independent analyses keep landing in the same place: “Many webmasters who implemented llms.txt reported that known AI bots never requested the file at all.” You can’t get value from a file the AI never opens.
Why ScaledOn doesn’t use llms.txt
The missing evidence is reason enough. But maintaining the thing carries real costs on top of that:
Version drift. Sitemaps generate themselves. llms.txt needs you to babysit it by hand. Let it fall out of sync and “LLMs may ingest outdated or misleading data, leading to hallucinated outputs.” Now you’re feeding the model stale facts about your own business.
No standardization. Nothing authoritative defines what makes a page “LLM-worthy” or how often these files should update. The whole implementation stays “ambiguous and brittle.”
Resource drain. You’d maintain llms.txt, probably llms-full.txt too (which can run past 200,000 words), keep all of it in sync with your live content, and spin up markdown versions of everything. “This burden hits smaller teams hardest.”
Canonical URL confusion. When an LLM links to your raw .md or .txt file instead of your real page, “users land on unstyled text instead of your polished website.” There goes your navigation, your experience, and your funnel.
The bottom line: busted
llms.txt launched in September 2024. Writing this in 2026, I can tell you no rigorous study has shown a measurable benefit from it. The SEO community loves a shiny new tactic. This one isn’t it. Skip it.
Put that energy somewhere it pays off. Getting cited by AI runs on the same fundamentals that win generative engine optimization: clear, authoritative, well-structured content that machines can parse and people actually want to read. Search Engine Journal put out a strong rundown of the factors that genuinely move whether ChatGPT cites you. Start there.
And if a vendor is pitching llms.txt as the next big thing, ask them for the data. I’ll wait.
Test the next “must-do” tactic before you buy it
Every quarter ships a fresh AI-SEO shortcut someone swears by. Most of them fold the second you check the data, exactly like llms.txt just did. I pressure-test these claims so you don’t have to, then send you the ones that survive. Subscribe free and you’ll get the next myth-busting breakdown in your inbox before you burn an afternoon on a fad.