There has been genuine confusion in the industry about whether llms.txt files matter. Last month, Google said they don’t read them.
“LLMS.txt files and other “special” markup: You don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in generative AI search.”
Case closed, then?
Not quite. It rcently emerged that Google has quietly added a check for llms.txt to its own Lighthouse audit tool. Whaaaatt? You don’t build something into your official auditing infrastructure if you think it’s irrelevant.
At Mogul, we’ve been implementing llms.txt files across client sites since mid-2025. Here’s what we’ve actually learned.
What is an llms.txt file, anyway?
Think of it as a plain-language briefing document for AI systems. It sits at the root of your website (e.g. yoursite.com/llms.txt) and tells AI crawlers, in structured, readable text, who you are, what you do, which pages are worth reading, and which areas should be off-limits.
It’s an emerging standard, not yet a formal requirement. But the fact that Google is checking for it is a nudge in a clear direction.
What we’ve implemented (and what we’ve learned)
A regional law firm
One of our early implementations. We generated a custom llms.txt file, got the client to review it for accuracy, and deployed it to the site root. What struck us was the internal reaction: sharing the file triggered a whole series of conversations at board level about how the firm was being described online. It had forced them to think clearly about their brand positioning in a way that a website audit never had.
The technical lesson: we set up a plugin so the page list generates automatically, but the introductory summary stays editable by the client. That balance works well.
A premium winery
This one gave us what we now call the “Bordeaux blend problem.” Our client produces wines with a very specific style and story. Left to their own devices, AI systems reach for generic descriptors and get it wrong. The llms.txt file lets us control that narrative directly: pointing AI models to the official vintage notes and winemaking descriptions, and away from checkout pages and private areas.
The broader point: for any client with a nuanced brand story, llms.txt is as much a reputation management tool as a search optimisation tool.
A professional services firm
We built a full AI/LLM search optimisation scope: structured data, llms.txt, content guidelines. This became the blueprint for how we now package and sell GEO services. Key learning: the out-of-the-box llms.txt output from SEO plugins is a reasonable starting point, but it needs proper customisation. Generic schema and generic file content looks budget and does budget work.
A software company
A more technical issue here. The llms.txt file was being silently rewritten by the WordPress multisite configuration. It took a few rounds of cache-clearing and plugin regeneration to get it showing correctly. The lesson: always verify the live URL after deployment. Assume nothing.
The plugin vulnerability problem
One thing nobody in the “just flip a switch in Yoast” camp talks about: the website-llms-txt plugin has had multiple security vulnerability notifications flagged across our client sites. This is not a reason to avoid the plugin, but it is a reason to keep it updated and to include it in your standard security monitoring workflow. Set-and-forget is not a strategy.
What the data actually shows
AI referral traffic is, right now, negligible for most businesses. One of New Zealand’s larger tertiary institutions received 273 visits from ChatGPT in a single month. That was 0.6% of its total traffic.
That stat matters because it reframes the conversation. llms.txt is not a traffic play. The real value is brand accuracy: ensuring that when an AI does talk about your business, it gets the story right. That’s worth doing even when the traffic volumes are small, because the influence of AI on purchasing decisions is growing fast regardless of whether the clicks are showing up in GA4 yet.
The higher-leverage work for AI visibility is getting your business cited in places that models actually trust: agency directories, industry publications, review platforms, and authoritative third-party sources. llms.txt helps with accuracy once you’ve earned that presence. It can’t manufacture it.
Time for some radical honesty
We’ve landed on this position internally: llms.txt is a quick win. It’s low-effort and low-risk. It takes minutes to implement, protects brand accuracy, and signals responsible digital governance. On that basis, we now include it as standard in every new WordPress build.
Agencies pitching for GEO work have to be honest about the limitations. The value is brand accuracy, competitive share of voice, and getting ahead of a curve that is moving fast. It is not a quick fix for more traffic.
What Google’s Lighthouse move tells us
It confirms what we suspected: Google’s public statements about llms.txt were designed to manage behaviour, not describe reality. They said “we don’t read it” to prevent a slop tsunami of low-quality, keyword-stuffed AI-generated llms.txt files of negligible value. But at the same time, they built a check for it into their own gold-standard site quality audit tool used by developers and agencies worldwide.
Watch what they build. Not what they say.
If you haven’t got an llms.txt file on your site yet, it’s a straightforward fix. If you want to know how your business is actually being described (or not described) by the major AI models, that’s a different conversation.
We’re running AI visibility audits now. Get in touch if you’d like to know where you stand.
Matthew Miller is Managing Director of Mogul Digital, a digital marketing and web development consultancy based in Noosa, Queensland, serving clients across Australia and New Zealand.
