Back to the Campfire Blog

The Current Consensus on LLMs.txt: Where Are We Now? (Still No)

Written by: Adrienne Kmetz Tags: future of seo, data, content

Published: Nov 27, 2025  |  Last Updated: Apr 28, 2026

In the midst of the infrastructure, thought leadership, and budget landgrab that is the “AI disruption”, we’ve come across a lightning rod: The LLMs.txt file. 

Somehow more controversial than buying links, the LLMs.txt file has emerged as a trend in experimentation and in memes. As SEOs try to find any way possible to structure more data and let’s face it, exploit any kind of shortcut in these early days, more and more websites are implementing it without any hard data to support it. 

As of April 2026, the consensus is still that llms is not readable. In this article by Tom Cranstoun at CogNovaMX, he explains how the idea is logical, but functionally does not work. Except, there may be some benefit to "getting it into training data". 

TLDR: It's not necessary and not a hack

• Start with no, especially if you are a small team without ways to automate it, and you haven’t yet completed more impactful projects like structured data or high-EAT content improvements. 
• There's likely bigger fish to fry in your organization; for instance, if you’re still writing FAQs, or if your content needs refreshing.  
• Or, if you want to hedge your bets just in case. (Of what, though? Just make sure you noodle on that question).
• Remember that providers have not endorsed, standardized, or even referenced this file. And, there is no data yet that shows correlation between implementing this tactic, and seeing results. 
• At Search Central Live in July 2025, “Gary Illyes clearly stated that Google doesn't support LLMs.txt and isn't planning to.” (h/t Kenichi Suzuki).
• Balance your curiosity with prioritization and act accordingly. 

John Mueller summed it up nicely in The Community with this comment: 

"Do what gives you joy and don't let random internet strangers tell you what to enjoy" - John Mueller

Annnnd since he wrote that, this happened: 

Which definitely has everyone scratching their heads. Is it natural for Google's hands to not know what its legs are doing? Sure! Departments are known to experiment in there, and good on them. Google's spokespeople are also known for saying things that we as SEOs have come to know are true or not true. 

So while last's week's news about Mintlify, Anthropic, and Google’s Agent2Agent protocol may not have changed much, this little flash in the pan may prompt some folks to want to play around too.

Just remember llms-txt.io reminds us that whatever traction so far has been in niche tech organizations: “llms.txt has achieved rapid penetration in developer tools, AI companies, technical documentation, and SaaS platforms, while remaining completely absent from the mainstream web” and consumer facing websites. 

A phrase I like to say when I notice my bias is: Or, not!

Just like many things, I've been wrong before. There's incentive for LLM companies to create resource-intensive first party data mechanisms that don't rely on the Google index. It starts getting original data in simplified format that it can easily store, and integrates further into the marketing operation, taking "market share" of an SEO's daily tasks.

The cynical marketer in me wonders whether the LLM providers' push for "too big to fail"-sized partnerships, will win over the practicality of Google essentially posting the Jordan "just stop" meme. We'll see! 

Instead of recapping all the other great articles out there, I’ve organized the quick hits for you. 

You know what they say about good intentions

The original purpose of Jeremy Howard’s proposal in November of 2024: “A proposal to standardise on using an /llms.txt file to provide information to help LLMs use a website at inference time.” It makes sense logically; with more context, crawlers can better understand the audiences and relationships between your pages. 

The takes are hot

Who am I to get in your way of the chisme. Here’s some: 

Google's John Mueller compared llms.txt to the old meta keywords tag. Carolyn Shelby has a rebuttal on why that analogy is a miss, and that new things take time. If the argument is to wait until the providers confirm they are using it, imagine if we as SEOs said that about anything having to do with Google. Let’s wait til Google confirms it, said no SEO ever. 

SE Ranking research findings: Its analysis of 39,000 domains (they say 300k but only 0.13% of the sites had one. It's early days, so this makes sense) found that llms.txt "doesn't seem to directly impact AI citation frequency. At least not yet."

Search Engine Journal noticed a "misinformation loop" where SEO tools flag missing llms.txt as an issue, users feel anxiety about AI visibility, tools offer it meet demand, and the perception of necessity reinforces itself. 

Slack has managed to turn it into a brand moment

Knowing we're all going to be checking these out, Slack has given us their usual sharp wit and a cute visual to go with it. 

The markdown efficiency argument needs more data to truly validate

Some experts claim it has value for agent workflows and token efficiency, even if it's not boosting SEO visibility. LLMs must parse messy HTML containing menus, ads, JavaScript, and CSS to find the relevant text; which spends tokens on understanding non-content elements. 

• An open-source tool called llm-docs-builder claims it strips 85-95% of noise from documentation for AI systems.
• Benword reports “a 10x reduction in token usage” using markdown.

If your site is so massive that these kinds of gains might mean delivering 5-10 times more information to the LLM than previously in the same context windows, well then, maybe you can do a little scope work to see if this math is worth testing. 

The main issue with this argument is that the the token savings comes from serving all of the content in markdown. Which brings llms-full.txt into the equation. A full file is all of your most important content, in full markdown. The Cloudflare file is 3.7 million tokens by itself. 

Now this is the part that feels a little ridiculous. If we cannot tell the machines to understand how to parse and/or ignore anything that is not Main Content (MC), I have a very hard time imagining that replicating all content on the site in a text file is even a short-term solution for token efficiency.

• Automation is currently out of reach for most small teams. 
• It can conflict with your sitemap. 
• Sounds like a maintenace and alignment nightmare (opportunity?) to some. 

Questioning the practicality 

Community member Mike Buckbee explains for us: 

"Couple different things:

• We have quite sophisticated tools for taking HTML+Images and making it plain text of all sorts, so I'm not sure it helps anything there.
• It's built on the assumption that LLMs just want text, but the newest models are quite mutli-modal and support images, video, etc. which makes this feel outdated.
• One larger strategic change is that you want more `Brand Statement` content as part of your overall content strategy than before AI, and that might not fit cleanly into how you present your marketing, so it might be helpful for that."

Do more important things instead

There is an opportunity cost to chasing down trends that don’t show strong signals that they’re going to pan out. If your site needs fundamental work, don’t chase down an unconfirmed perceived shortcut. Work on: 

• schema.org structured data
• EAT improvements 
• Clean HTML with good internal linking
• Digital PR and strengthening your brand

Okay but like.. I still wanna

I knew it! Go forth responsibly, my friend. Check out these guides and examples: 

• A primer: What is it and how does it work
• /llms.txt directory: A curated list of products and companies using the llms.txt file. 
• llms.txt sites directory: A list of all llms.txt file locations across the web with stats and links to the files.
• Take a look at the PokeVerse file. 

As always, new things require time in order to see how it all shakes out and to gather enough data to understand how it all works. What’s hype today could quietly become standard, or totally irrelevant, tomorrow. 

At the end of the day ask yourself, does this exercise get me closer or farther away from where I need to go? Then base your decision on that.