A Guth Labs publication ↗
Back to News

Cloudflare adds analytics and tracing for Markdown delivered to AI agents

Cloudflare announced new analytics, logs and tracing for Markdown for Agents on November 9, giving website operators more ways to investigate how their pages are converted for AI systems. Its announcement describes the service as generally available and adds a dashboard playground for comparing a page’s HTML with its Markdown output. Announcement

The dashboard reports conversion results, problem pages and content limits. A GraphQL dataset provides up to seven days of conversion information at account or zone level, while sampled traces expose conversion outcomes, byte counts and CPU time. These tools can help distinguish pages that converted successfully from those that were truncated or failed. Changelog

The measurements come with important limits. Cloudflare calculates estimated tokens from bytes rather than using a model’s tokenizer. Savings calculations count only full conversions, and even a full conversion is not a promise that every element of the HTML survives. User-agent strings in analytics are supplied by clients and do not establish a bot’s identity. A completed conversion also does not prove that the requesting client received it. Technical documentation

The service supports HTML input up to 6 MiB after decompression. It is available without an additional charge on Pro, Business and Enterprise plans and for SSL for SaaS customers. The documentation distinguishes this automatic conversion path from separate tools that render dynamic pages in a browser. Availability and limitations

For an AI application team, the practical value is a more inspectable input pipeline. A smaller page representation may reduce the material an agent needs to process, but a token-reduction chart alone cannot establish answer quality. A useful evaluation would compare representative outputs with the original pages, checking whether tables, qualifications and links remain understandable. Operators can now investigate conversion behavior more directly; they still need to test what their agents actually read and do with the result.

Sources

Sources and citations

Revision history

Published .