658 Top Websites Use Home-Made AI Meta Tags Like ai-content and llms-txt, and 63.1% of Them Also Publish llms.txt

Summary

Picture a consultant who tells a dental practice to add a few "AI meta tags" so ChatGPT will understand the site. The developer adds ai-description, ai-authority and llm-preferred to the page head. The practice pays for the work and hopes for the best.

Meta tags only work when the software reading the page knows what they mean. Search engines document the tags they support. The AI crawler pages from OpenAI, Google and Anthropic describe robots.txt as the control and do not list any ai- or llm- meta tag names (checked October 21, 2026).

We looked for these home-made tags across the top 1 million homepages and found 658 sites using them. The names vary widely, which is itself a sign that there is no shared standard behind them.

What we measured

We used the HTTP Archive October 2026 mobile crawl (httparchive.crawl.pages) for top-1M homepages and read every meta tag name recorded in the almanac meta-nodes metric. We grouped names into three families: names starting with ai (ai-, ai:, ai_), names starting with llm, and names equal to an AI crawler's user-agent token (for example gptbot or claudebot).

We left out the standard robots and googlebot tags. We then compared these sites with llms.txt adoption from the same crawl.

Three families of made-up names

A site can fall into more than one family.

The ai- family alone holds dozens of different names. Many look like attempts to describe the business to AI in a structured way: topic, audience, summary, trust.

The most common names

Names used on at least 20 homepages.

Some names try to give instructions (ai-training, ai-access), while others try to give facts (ai-authority, ai-description). Without a standard, a crawler has no reason to read either kind.

These owners tend to try other AI files too

Counts by popularity rank.

These sites are over four times as likely as average to serve llms.txt. Rates are low in every bucket, between 0.07% and 0.27% of homepages.

What this means

Before paying for "AI meta tags," ask which AI system documents support for them. If no one can point to that documentation, the tags are guesswork. The time is better spent on things AI tools do read: visible page text, clear headings, accurate structured data and a robots.txt that lets AI search crawlers in.

If you already added such tags, they are unlikely to hurt, but do not count them as optimization work. Real optimization starts with seeing what assistants say, and a free AI visibility audit from our team shows you exactly that.

Limits

Our name patterns may miss home-made tags with other naming styles and may include a few unrelated names that start with ai.

We checked documentation from three AI companies; other systems may read some of these tags.

Meta tags are read from the homepage only.

Single crawl, October 2026.

Frequently asked questions

Related studies

• AI training opt-out signalsOnly 0.9% of News Homepages Mark Content as Paywalled
• AI training opt-out signalsWeb Bot Auth Keys: 0.01% of Top Sites Publish Them
• AI training opt-out signalsOnly 1 in 4,544 Top Sites Sends an AI Opt-Out Header
• AI training opt-out signalsOnly 0.13% of EU Sites Declare a TDM Reservation for AI