Concepts
Does llms.txt help with AI citations?
Ahrefs (2026) studied 137,000 sites and found 97% of sites that added an llms.txt file saw zero change in AI-bot traffic — no measurable citation effect. No major answer engine (ChatGPT, Google AI Overviews, Perplexity, Gemini) reads llms.txt as a ranking or citation signal.
What is llms.txt?
llms.txt is a plain-text file proposed in 2024 as a convention for helping AI systems understand a website's content. Site owners place a file at /llms.txt summarizing content, permissions, or preferred framing for LLMs, modelled loosely on robots.txt. No major answer engine — not Google AI Overviews, not Perplexity, not ChatGPT search, not Gemini — has announced support for llms.txt as a citation or ranking input.
robots.txt is a standard enforced by every major crawler because every crawler has explicitly committed to reading it. llms.txt has no such commitment. The 4 largest answer engines (Google AI Overviews, Perplexity, ChatGPT search, Gemini) have not confirmed that llms.txt influences any retrieval, ranking, or citation decision.
What does the evidence say?
Ahrefs (2026) analyzed 137,000 sites and found 97% of those that added an llms.txt file saw zero change in AI-bot traffic after adding it. The study found no statistical evidence that llms.txt influenced which pages were crawled, cited, or surfaced by any major answer engine. The effect was indistinguishable from zero.
Seeklyn's own "Don't bother" audit category surfaces this finding directly: llms.txt is labeled as a myth in the landing page analyzer with an evidence grade of moderate, sourced to the Ahrefs 2026 study. Seeklyn shows it not to mock the idea, but because the honesty moat is the product — no competitor will tell you when a tactic doesn't work.
Why does the myth persist?
The myth persists because llms.txt sounds plausible by analogy to robots.txt. Vendors selling "AI SEO" products benefit from a growing checklist. The file costs nothing to add, and without a controlled experiment, absence of harm gets interpreted as benefit.
6 structural signals demonstrably raise citation rates (Princeton GEO, KDD 2024): self-contained 100–300-word passages, fact density of ≥1 figure per 100 words, attribution, answer-first structure, named entities, and freshness signals. Time spent on llms.txt is time not spent on any of these.
What should you do instead?
Spend the effort on the 6 structural levers the Princeton GEO study (KDD 2024) correlated with measured citation lift: self-contained passages, fact density, attribution, answer-first structure, named entities, and freshness signals. Seeklyn scores all 6 in every content analysis. All 6 work regardless of whether your site has an llms.txt file.
- Write self-contained 100–300-word passages that answer a question without requiring prior context.
- Include at least 1 concrete figure per 100 words with a named source.
- Lead every section with an answer, not background.
- Name entities: companies, products, standards, authors.
- Date your content — freshness signals help.
- Skip llms.txt — it is not read by major answer engines.
Frequently asked questions
Should I delete my llms.txt if I already have one?
Keeping it costs nothing (a small text file), and removing it has no benefit. The point is not to delete it but to stop treating it as a citation lever. If it took more than 5 minutes to create, that effort would have been better spent on a self-contained answer capsule.
What about robots.txt for AI bots?
robots.txt is different — major AI crawlers (GPTBot, PerplexityBot, ClaudeBot, Google-Extended) do read it and respect Disallow rules. If you want to block a specific AI crawler from indexing your content, robots.txt is the correct mechanism. llms.txt is a separate convention with no enforcement.
Could llms.txt become useful in the future?
Possibly, if a major answer engine announces support. As of 2026, none has. Seeklyn will update its guidance if the evidence changes — evidence-first, no adjectives.
Does Seeklyn flag llms.txt as a failure in its audit?
No. Seeklyn shows llms.txt in its "Don't bother (myths)" section with a status of info — it doesn't hurt your score, it just doesn't help. The section is informational only and never penalizes your overall score, because the honest finding is that the file is neutral, not harmful.