I built https://t.co/5NktuMjX9i
Put https://t.co/ZxgaXfljH1 before any URL → get clean Markdown back.
Cloudflare's Markdown for Agents is great, but only works for enabled sites. https://t.co/5NktuMjX9i works for ANY website on the internet.
80% fewer tokens. Also converts PDFs, images, audio.
Free. No signup.
https://t.co/5NktuMjX9i
Time to consider not just human visitors, but to treat agents as first-class citizens. Cloudflare’s network now supports real-time content conversion to Markdown at the source using content negotiation headers.
https://t.co/B7wYH4PtA8
We told the campaign. About an hour later: "Please reactivate our chatbot."
We switched it back on, and WIRED changed the headline to "...Is Back Online."
Happy ending. Very stressful few days.
Full story: https://t.co/mxBeE7LDYy
I told this story at Fuckup Nights Ankara.
In 2024, RFK Jr.'s presidential campaign ran its chatbot on our product, @LiveChatAI .
Then a WIRED reporter emailed asking how, since OpenAI bans political use.
We had no idea that rule existed. 🧵
But there was one detail. We didn't use OpenAI directly. We used it through Microsoft Azure.
Microsoft told WIRED its customers weren't bound by OpenAI's terms, and the bot wasn't breaking its rules.
Banned at OpenAI. Fine on Microsoft's OpenAI.
@broIMxd agree the engines drift. that's exactly why the score never asks one. it checks the page against a versioned ruleset: crawler access, extractable passages, schema, plumbing. same page + same rules = same number next month
Deterministic & open-source GEO audits for AI search. Claude Code skills that score how well AI engines can read, quote and cite a site.
https://t.co/vzDWZM14n1
@swordBluesy please do, that's exactly the test i'd want someone to run. rules live in src/geo_audit/data (weights, thresholds) and src/geo_audit/scoring (the signals). one gotcha: it blocks localhost by default, so serving the fixture locally needs --allow-private
@Kantorcodes the raw HTML. every page is stored gzipped under its sha256, with robots.txt and llms.txt next to it, plus status, headers and the redirect chain. so --rescore reruns extraction and every signal with today's code. a changed rule gets re-applied to the same HTML, not replayed