llms.txt in 2026: Standard, Hype, or Dead on Arrival?

Every six months the SEO industry discovers a new file you are supposed to put at the root of your website. llms.txt – proposed by Jeremy Howard in September 2024 – is the latest. The hype cycle is in full swing. The data is much less flattering.

What llms.txt is supposed to do

The proposal: a Markdown file at /llms.txt that lists your most important pages with short descriptions, helping LLM agents efficiently discover and consume your content. A companion /llms-full.txt bundles entire content as a single Markdown file.

The technical argument is solid. SerpApi reported that an agent reading their HTML page burns ~79,000 tokens; the Markdown equivalent uses ~1,800 tokens – a 10x reduction. For agentic workflows that programmatically read your docs, the efficiency gain is real.

What the adoption data actually shows

Three independent studies tell the same story:

• SerpApi cites ~10% adoption in 2026, heavily inflated by WordPress and Drupal plugins that auto-install the file.
• Rankability’s scan of the top 100 websites found zero adoption among market leaders.
• QuickSEO’s analysis across 2025-2026 found zero measurable impact on AI citations, traffic, or rankings.

What major AI engines actually say

• Google. Multiple Google representatives have publicly confirmed they do not use llms.txt and have no plans to.
• OpenAI / ChatGPT. No announced support. ChatGPT’s crawler reads standard HTML.
• Anthropic / Claude. No announced support.
• Perplexity. No announced support.
• Microsoft / Copilot / Bing. No announced support.

If none of the engines that drive AI citation use the file, it cannot influence citation. That’s not skepticism – that’s arithmetic.

When llms.txt actually makes sense

Two narrow, legitimate use cases:

1. API/SDK documentation. If your users are developers writing agents that programmatically read your docs, an llms-full.txt at /llms-full.txt measurably improves token efficiency and accuracy. SerpApi added it for exactly this reason.
2. Internal AI infrastructure. If you’re building MCP servers or RAG pipelines over your own content, having a curated content manifest is genuinely useful.

If you’re a marketing team hoping for a citation lift, the opportunity cost of building llms.txt vs. improving your actual content structure is too high. Skip it.