If you have spent any time looking at GEO over the past year, you will probably have come across advice to add an llms.txt file to your website or create Markdown versions of important pages for AI crawlers.
The theory is easy to rationalise. If LLMs prefer clean, structured text, then surely providing a simplified file or a Markdown copy of your content should make your site easier to understand and improve your chances of appearing in AI search.
At the moment, though, there is very little evidence that either tactic materially improves visibility across the major platforms.
In fact, Google’s current guidance is clear: it does not use llms.txt, special AI text files or Markdown versions as part of its generative search systems.
Everyone’s favourite Google spokesperson, John Mueller, has also been sceptical about the practical value of llms.txt, comparing the idea to the old keywords meta tag and pointing out that major AI services have not indicated that they use it.
What is llms.txt actually supposed to do?
The proposed llms.txt standard gives site owners a way to publish a simple Markdown-formatted file, typically an llms.txt file, that contains a curated description of the website and links to important content.
In theory, this gives AI systems a cleaner and more deliberate map of the site than they might get from navigating the full HTML experience. For documentation-heavy websites or products with large technical knowledge bases, this can sound particularly attractive.
The problem is that a standard is only useful if the systems you care about actually adopt it.
At present, there is no evidence that the major AI search platforms rely on llms.txt as a meaningful discovery or ranking signal. Google has explicitly said that Search does not use these files, while OpenAI, Anthropic and Perplexity continue to publish crawler guidance centred on ordinary web crawling and robots.txt, rather than asking publishers to create a separate LLM-specific index.
Focus on the site, not the file
The bigger issue with llms.txt is that it risks becoming a distraction from the work that actually makes a site easier for search and AI systems to understand.
A declaration file can tell a crawler what a site owner considers important, but it does not fix unclear positioning, weak content, poor site structure, or inaccessible pages. If the underlying website is difficult to interpret, llms.txt won't solve that problem.
For most brands, the better question is whether the content they already publish is clear, crawlable, authoritative, easy to retrieve, and high-quality. If that foundation is strong, llms.txt is unlikely to be the thing that materially changes visibility.
The crawl data is not especially encouraging
In June 2026, Ahrefs analysed roughly 137,000 websites and found that around 28% had an llms.txt file. Of those files, 97% received no requests at all during the study period.
That does not prove llms.txt will never matter. Standards can gain adoption over time, and as we all know, crawler behaviour changes quickly. It does, however, make it difficult to justify spending meaningful technical or content resources on llms.txt today if the objective is simply to improve visibility and reputation across major AI platforms.
For most teams, there are far more important things to fix first.
What about separate Markdown versions of your pages?
The Markdown argument is closely related. Because Markdown is simpler than a typical web page, some practitioners recommend creating a parallel Markdown version of important content so LLMs can process it more easily.
Again, the logic sounds plausible, but it doesn’t exactly scream ‘top priority’ for most.
Google explicitly states that site owners do not need to create new machine-readable files, AI text files, markup, or Markdown for content to appear in generative Search. Its advice remains focused on making the main site technically accessible and ensuring the content is useful, clear, and easy to find.
If an AI crawler struggles to extract the important information from your existing HTML, the answer is usually to improve the page itself rather than maintain a second shadow version of the same content.
A clean, well-structured web page helps search engines, AI systems and human users at the same time. A separate Markdown layer introduces another version to maintain, another place where information can become outdated, and another workflow for teams to manage.
There are some legitimate use cases
None of this is to say that llms.txt or Markdown are completely pointless.
There are situations where they can be useful, particularly for developer documentation, agent workflows or systems where a clean machine-readable index genuinely improves usability.
If your CMS can generate an llms.txt file automatically and keeping it current requires almost no effort, there is little downside to publishing one.
Likewise, if you already maintain Markdown documentation because your technical audience uses it, there is no reason to remove it.
The mistake is treating either tactic as a meaningful GEO shortcut.
Where we’d spend the time instead
If the goal is to improve visibility across LLMs, the priorities are clear.
Make sure important pages are crawlable and technically sound
Clarify the way your products, services and markets are described
Publish original, useful content that adds something to the wider information environment
Strengthen third-party evidence for your brand through PR, reviews, partnerships, and relevant external coverage
Understand which sources are actually being cited in AI answers and where competitors are present while you are not.
Most importantly, measure whether the work is changing the answers.
That gives you evidence about which signals matter in your category rather than relying on a new tactic simply because it sounds AI-specific.
Don’t optimise for hypothetical crawlers
The temptation in GEO is to chase every new file format, schema proposal or technical workaround as soon as it appears.
Sometimes those ideas become important, but often it’s another shiny thing that takes up time, headspace and focus from the things that matter long-term.
Right now, llms.txt and dedicated Markdown versions sit firmly in the “nice to have, if effortless” category rather than the “important visibility tactic” category.
The better use of time is to make your website clear, accessible, and authoritative, and then strengthen the external evidence supporting your brand across the web.
If adoption changes, you can tackle it then, but for now, we would keep llms.txt low on the list.
See what is actually influencing your AI visibility
Surfacd shows which sources, prompts and competitors are shaping AI-generated answers, so you can prioritise the work that has a measurable impact rather than the latest GEO distraction.
