Hardin Filter checks AI answers from ChatGPT, Claude, Gemini, and Copilot for what actually matters: fabricated citations, references that don't exist, numbers that don't add up, contradictions, leaked data, and outdated info presented as current. Every check produces a signed, verifiable receipt (via our partners at TBN Protocol) — proof of what was checked, not just a green tick. Chrome extension, VS Code extension, or paste-in web tool. Free, no account needed.

Hey everyone 👋 I built Hardin Filter because I kept seeing AI answers that sounded completely confident but were quietly wrong — a made-up source, a package that doesn't exist, numbers that didn't add up. Once you start looking for it, it's everywhere, and there was no easy way to catch it in the moment. The approach that stuck was: don't try to make the AI smarter, just check what it actually said. Hardin Filter runs each answer through a set of checks for fabricated references, numeric consistency, contradictions within the same answer, leaked data, and staleness — then gives you a signed, verifiable receipt instead of a plain pass/fail you have to trust. It started as a Chrome extension, but pretty quickly it was obvious people needed it in more places — so a VS Code extension and a paste-in web tool came next (the web tool also covers anyone on mobile or using an app instead of a browser). A Safari version is built and ready, just waiting on a Mac-owning volunteer to help test it — if that's you, there's a link on the site. Would love feedback, especially on where the checks feel too strict, too lenient, or miss something you'd want caught.

The signed receipt approach is genuinely clever, turning a trust signal into something you can actually verify instead of just another "trust me" badge.

Hardin FIlter

Catch AI hallucinations before you trust the answer

Catch AI hallucinations before you trust the answer