What changed in detectors this quarter

Detectors change without announcements, and a result from three months ago may not hold today. This is a recurring post (we'll refresh it each quarter) tracking what actually moved and why it matters for your writing.

How to read this series

Each edition notes, where they're public: detector model or version updates, changes to how a tool sets its 'likely AI' threshold, notable false-positive reports (especially those affecting students and non-native writers), and any published accuracy claims with their dates. We link sources and we don't editorialise a vendor's numbers into a pass-rate claim of our own.

The reason this is a series and not a one-off is the whole point: there is no stable 'best detector' or fixed 'pass rate'. The responsible thing is to track the movement and help you interpret it, not to freeze a flattering moment.

The constant across every quarter

Whatever changed, the guidance for writers doesn't: vary your rhythm, be specific, keep your drafts, follow your institution's or employer's rules, and treat any score as an estimate. Tools that promise to keep you ahead of every detector update forever are promising something no one can deliver.

If you want the current method behind our own testing, see 'How we test our output against detectors'. When we have measured pass rates we'd stake our name on, they'll appear here, dated and sourced.