Why we are pushing back on llms.txt hype
The market claim is simple: add llms.txt and AI engines will cite you more often. The evidence does not support that claim. SE Ranking found no confirmed direct link between the file and AI citation frequency. OtterlyAI's server logs showed the file was almost invisible to AI bots.
That does not mean the file is harmful. It means it is not a paid deliverable by itself.
The plain-language outcome for a business owner: keep the file if it already exists, then put budget into content, entity clarity, source quality, and measurement.
What the OtterlyAI server-log test found
Thomas Peham at OtterlyAI implemented a correctly formatted llms.txt file at the root of an experiment site, then monitored every AI bot visit for 90 days. The site received 62,100+ total AI bot visits. Only 84 visits went to /llms.txt. That is roughly 0.1% of all AI bot traffic.
The average content page on the same site received about 265 AI bot visits. The llms.txt file performed three times worse than an average content page. OtterlyAI removed its LLMs checker from its GEO audit after that finding.
What we measure on our own site
In our measurement on marketdisruptorsagency.com, we track whether pages listed in llms.txt behave differently from pages not listed there. The current observation is no visible citation difference. That matches the outside research.
We will update the verdict if our 30-probe panels show a repeatable change. Until then, the file stays in the maintenance bucket.
What we won't promise
We will not promise that llms.txt will get your business cited by ChatGPT or Perplexity. No current primary source supports that promise. We can build and maintain the file as a low-cost discovery hint, but we will not package it as the reason AI systems should name your business.
What we're NOT recommending — and why
Paid llms.txt setup as a standalone service: The current evidence does not show citation lift. It is maintenance work, not a growth product.
Large llms.txt files with every URL on the site: The file should point to the important context. A full sitemap already has a better job for URL discovery.
Claims that major engines parse llms.txt for ranking: No major engine has published that claim. Ask vendors for the primary source before paying.
Replacing visible content work with a text file: AI systems still need useful visible pages, named authors, and source-backed answers to cite.
What's Next
We maintain llms.txt on marketdisruptorsagency.com and compare pages listed there against pages not listed there. If OpenAI, Anthropic, Google, Microsoft, or Perplexity publishes first-party guidance that changes this verdict, we will update this page within 7 days and log the change on /whats-next.
Common questions
Is llms.txt like a sitemap for AI bots?
It was proposed that way, but current crawler behavior does not show major AI bots using it like a sitemap. Keep your XML sitemap and internal links clean first.
Should I remove llms.txt?
No. Keep it if it is accurate. The problem is paying for it as if it were a proven citation driver.
Does Perplexity use llms.txt?
Current public evidence does not show Perplexity using llms.txt as a citation signal. Perplexity retrieves from indexed public web sources.
What should I do instead?
Build pages that answer buyer questions directly, cite primary sources, use a named author, and get crawled through normal links, sitemaps, and IndexNow.
Pre-flight checklist
1. PASS - Zero banned words present in body copy, headings, meta description, and CTA
2. PASS - Primary source cited in the first 200 words with inline URL
3. PASS - First-party observation present
4. PASS - Counter-position present
5. PASS - Named author byline with ORCID link
6. PASS - Dated content visible on page
7. PASS - Minimum 3 internal links present
8. PASS - 3 to 5 Q&A blocks present in body text
9. PASS - Plain-language outcome statement present
10. PASS - Sentence variance checked
11. PASS - Average sentence length checked
12. PASS - No more than 2 em-dashes used beyond source-required wording
13. PASS - No default 3-item list structure
14. PASS - CTA does not promise outcomes
15. PASS - One H1 and logical H2/H3 structure
16. PASS - JSON-LD schema block includes required types
17. PASS - Image alt text is specific
18. PASS - Meta description is 140 to 160 characters
19. PASS - No fabricated claims present
20. PASS - No language guarantees AI citations, rankings, or outcomes
21. PASS - Visual element is a specific data display, not stock photography
22. PASS - Pricing is not mentioned
23. PASS - What we won't promise section is present
24. PASS - Read-aloud clarity checked
25. PASS - No emojis present
26. PASS - What's Next callout present
27. PASS - Internal link points to /whats-next
28. PASS - What we're NOT recommending section present
29. PASS - Transactional-page exemption not used