This is part of an email note I got from someone who runs an annual writing contest in the U.S.

“I’ve gotten numerous inquiries about AI detection as part of the prize process, so I thought I’d reach out to share what I did in order to dispel any concerns. 

Before starting this year’s prize, I researched numerous AI detectors, including reviews, user ratings, testimonials and the like. Then, I tested each of them several times with work I knew to be 1) 100% human-authored; and 2) 100% AI-generated. Some were far better detectors than others. I settled on using WinstonAI. 

Once I narrowed the 956 prize entries down to 125, I wanted to read a second (and third and fourth) time, I ran the entries through WinstonAI. Nearly all the entries I tested rated 98-100% human in the AI detector, so those that didn’t stood out. 

I understand that these AI detection tools are not perfect, and I took the results with a grain of salt.

I directly contacted those whose work showed the possibility of AI use, and if the writer assured me they didn’t use AI, I chose to trust them rather than the detector. 

As much as I’d prefer not to have to even think about this (and do the extra work it entails), I feel it’s only fair to all the honest human submitters to conduct due diligence to root out AI-generated submissions.”