AI-writing detectors LIE a lot.

I wrote an article, then did a grammar and consistency check it with Grammarly Premium. Then I checked it for AI content on QuillBot and Grammarly for AI. It showed 0–8% AI-written, which is fine.

But when I checked it on GPTZero, it said, ‘we are confident this is 99% AI-generated.’ Some AI detectors are grossly inaccurate.

I just imagine a lecturer marking down or failing a student or a company firing its writer on this ground, or does everyone take extra caution of using three AI detectors instead of trusting one?

I have experienced situations where I tested out the same piece of article for AI with the same AI detector but got different results in terms of percentages. Even Grammarly AI Checker once highlighted a different part of an article as AI-written, different from the part it initially selected. I was amazed!

I think AI-detectors are great, but their inaccuracies may be costly. I prefer Grammarly’s phrasing, something like “similar to AI” and not GPTZero definite term “we are confident”.

Photo by John Cameron on Unsplash

Have you experienced or learnt of a similar inaccuracy with AI-checkers?