Four months ago, I added llms.txt and llms-full.txt files to my blog. Time to look at the numbers.
What is llms.txt
It’s essentially robots.txt for language models. A file in the site’s root directory that helps AI systems — ChatGPT, Perplexity, Claude, Gemini — better understand the structure and content of your website. I wrote more about it in a separate post.
My files:
Methodology
I compared two 4-month periods:
| Period | Dates | Status |
|---|
| Before | March 18 — July 17, 2024 | without llms.txt |
| After | July 18 — November 18, 2024 | with llms.txt |
Data source — Yandex.Metrica, “Referral links” report.
I filtered AI chat domains: chatgpt.com, perplexity.ai, chat.deepseek.com, gemini.google.com, chat.qwen.ai, copilot.microsoft.com, alice.yandex.ru.
Results
Overall Picture
| Metric | Before llms.txt | After llms.txt | Change |
|---|
| Total referral traffic | 160 sessions / 114 users | 290 sessions / 136 users | +81% / +19% |
| From LLM services | 75 sessions / 51 users | 92 sessions / 64 users | +23% / +25% |
| LLM share (sessions) | 47% | 32% | −15 p.p. |
| LLM share (users) | 45% | 47% | +2 p.p. |
Breakdown by Service
| Service | Before (sessions) | After (sessions) | Change |
|---|
| Perplexity | 29 | 55 | +90% 📈 |
| ChatGPT | 31 | 26 | −16% |
| DeepSeek | 10 | 1 | −90% 📉 |
| Gemini | 3 | 2 | −33% |
| Qwen | 2 | 1 | −50% |
| Copilot | — | 4 | new |
| Alice | — | 3 | new |
What This Means
Perplexity is the main beneficiary. Nearly doubled traffic. This service clearly has better llms.txt support and actively uses structured data for answers with source citations.
ChatGPT is stable by users (22 → 23), but fewer sessions. Hypothesis: users get answers directly in chat and click through less often. That’s not bad — the content still works.
DeepSeek dropped sharply. Possible reasons: changed citation policy, restricted external links, or simply fewer CIS users using this service.
New players appeared. Microsoft Copilot and Yandex Alice started bringing traffic, though they weren’t there before.
The share paradox. LLM traffic grew in absolute numbers (+23%), but its share dropped from 47% to 32%. The reason — total referral traffic grew even faster (+81%). Other sources (ahrefs, threads, habr) grew faster than LLM.
Should You Implement It
Yes, if:
- Your content is useful and unique
- You want AI systems to cite you correctly
- You’re willing to spend 15-30 minutes on setup
For WordPress, there are plugins (I use All in One SEO), for other CMS — manual setup or generators.
What Not to Expect
No explosive growth. +23% over 4 months is a good but not dramatic result. However, it’s quality traffic: people come from specific queries they asked their AI assistant.
Bottom line: llms.txt works, especially for Perplexity. Minimal effort, measurable results. If you have a technical blog or knowledge base — implement it.
Have your own before/after stats? Share in the comments.