The file almost nobody reads
llms.txt is a small text file meant to show AI services at a glance which content on a website matters. The idea sounds reasonable, and setting it up takes about fifteen minutes. The problem: the major AI providers simply do not evaluate it in practice.
John Mueller of Google said this as plainly as a corporation rarely does, in spring 2025. Asked whether AI services actually use the file, he wrote in a developer forum:
“As far as I know, none of the AI services have said they use llms.txt. And looking at server logs, they don’t even look for it. To me, it’s comparable to the keywords meta tag.”
John Mueller, Google, April 2025
The comparison lands. The keywords meta tag was a promise made twenty years ago that nobody believes any more, because it was too easy to abuse. The data backs Mueller up: an analysis of around 300,000 websites found no connection between having an llms.txt file and appearing more frequently in AI citations. A server-log experiment showed that only a fraction of a percent of AI traffic ever requests that file at all.
It does no harm. It does not make you visible, either.
WordPress or custom-built? The wrong divide
The second favourite question is about the system behind the site. WordPress, Webflow, a custom-built stack: these get turned into a debate about religion. For AI visibility, that debate is almost meaningless. Any of these systems can produce citable content, and any of them can fall badly short.
Where they differ is in friction. WordPress with a decent schema plugin lowers the barrier to marking up a page cleanly. Webflow needs extra code or an extension for more demanding structured data. A custom-built stack can get everything technically right, but it requires care and skill to do so. That is a difference in effort, not in what is achievable.
Put simply: a well-maintained WordPress beats a careless custom build. A carefully built custom stack is not inferior to anyone. Anyone who blames the technology stack for missing visibility has not yet found the real problem.
What AI systems actually base decisions on
If it is not the file and not the platform, what is it? The answer is uncomfortable because it sounds like work rather than a quick checkbox. A figure from Ahrefs makes it concrete: in mid-2025, roughly three out of four AI-cited pages still came from the top ten Google results. About half a year later, that share had fallen to barely four in ten. Ranking well is no longer enough on its own. What matters is what the page says and how it says it.
- Topical depth. AI systems break a question into many sub-questions and check who covers them. Pages that treat a subject thoroughly and broadly are cited far more often than those that only hit the main keyword.
- Freshness. Recently updated content is preferred. A page left untouched for three years loses ground, even if the information is still correct.
- External authority. Being linked to often, and by reliable sources, signals trustworthiness. No file and no CMS can shortcut that reputation.
- Clear sentence structure. Statements that hold up even when read out of context are more likely to be picked up. Mike King of iPullRank puts the goal this way: a statement should be able to stand alone, removed from its surroundings, and still be complete.
- Clean markup. A machine-readable description of who you are and what you offer helps AI systems place you accurately. The details of how to do that are a topic in their own right.
Aleyda Solis, who advises internationally on AI search, deliberately published her optimisation checklist without naming a single system. Every point applies equally to every CMS. Lily Ray of Amsive makes the confusion explicit: much of what gets sold as new GEO advice is, word for word, the same guidance SEO professionals have given for years. AI visibility is getting the credit; solid foundational work earned it.
The problem most people do not even see
There is one figure that says more about the state of the field than any ranking-factor study. Almost nine in ten brands already appear in AI search results today. But only about one in seven businesses actually measures whether and how that happens.
Most are flying blind. They do not know whether an AI mentions them, for which questions, or how they compare to competitors. And what you cannot see, you cannot improve. The situation also shifts quickly: an answer given today may be different tomorrow. That is why AI visibility is not something you set up once. It is something you watch and adjust over time.
See where you stand first, then turn the right dials
Before you invest time in a file or a platform migration, the sober question is whether AI services mention your business today and where the biggest gap is. A website assessment shows exactly that, without promises the data does not support.
Frequently asked questions
Do I need an llms.txt file for ChatGPT and Perplexity?
Based on current evidence, no. The major AI providers do not evaluate the file in practice, and Google has confirmed this explicitly. It causes no harm and can complement a well-structured site, but it does not replace one. Whether the effort makes sense in your specific case is a question for a proper position analysis.
Is WordPress worse for AI visibility than a custom-built site?
No. A well-maintained WordPress with clean markup is just as citable as a carefully built custom system. The platform decides how much care is needed. The outcome stays the same. Quality of implementation is what matters.
When does a custom-built stack make sense over WordPress?
Only at genuine complexity: very large page counts, demanding data models, or multilingual setups. For most business sites, WordPress with a decent schema plugin delivers the same benefit for AI visibility at lower effort.
Is a good Google ranking enough to be cited in AI answers?
Less and less so. In early 2025 most AI citations still came from the top Google results. By now, the majority come from pages outside the top ten. Ranking well helps, but it no longer guarantees AI visibility.
What actually makes an AI cite a page?
Mainly four things: covering a topic with real depth, keeping content current, earning links from reliable sources, and writing clearly enough that individual statements hold up on their own without context. These factors work regardless of which system the site runs on.
How do I even know whether I appear in AI answers?
Only by checking regularly. You query the important topics in the relevant services and observe whether and how your business is mentioned, including how that compares to competitors. Very few businesses do this, yet it is the foundation of any meaningful improvement.