A lot of websites are starting to switch to serving custom content for agents - instead of requiring agents to parse the output. OpenCode fetch does requesttext/markdown, with fallback for html.
Ideally, the extract requests would request text/markdown as an Accept'ed response type.