Shipped llms.txt — what should I track first? (sanity-check my plan)

From my test (see seed-010 if you want the methodology): Perplexity reads llms.txt, ChatGPT is inconsistent, Google AI Mode seems to not read it at all. So your hypothesis is right in my data too. I'd keep it lean — 'use cases' section yes, but max 3-4 bullet points. My 300+ word llms.txt performed worse than the trimmed 170-word version in terms of how accurately Perplexity described my product.

The SE Ranking study from earlier this year showed basically zero citation correlation with llms.txt adoption. I'm not saying don't do it, but temper expectations. If your real bottleneck is entity recognition and external mentions, llms.txt is just a nice-to-have right now.

I've been running the llms-full.txt experiment for 6 weeks. Honest answer: not worth the maintenance overhead yet. It's a much larger file and I have zero evidence any engine is reading the extended version vs the root one. Maybe in a year.

Petr here — we have llms.txt on roughly 40% of monitored domains now. Honest take: the measurable lift in citation rate is small and concentrated on Perplexity. The bigger value we observe is editorial — the act of writing the file forces a clean one-page positioning statement that people then port into their hero copy and About page, and THAT moves things. So don't expect the file itself to be the win.

If you want a cleaner A/B than 'before vs after deploy', stand up a second subdomain (or a noindex staging clone) without llms.txt and run the same prompt battery against both. That isolates the file from concurrent content/PR/schema work, which is usually the real confound. We ran this for 6 weeks across two product lines and saw a measurable delta only on Perplexity for the variant with the file.