To writ: I’m STILL editing my next feature because of this stupid Pangram scanner, even though it’s been mangled beyond recognition at this point, because yes, polish really IS what it most flags as suspicious… funny enough, the raw draft came back as 100% human, and it was only after I went through my piece to edit and clean it up that the number jumped to 30% AI, and it’s fluctuated between 3% and 25% in the following ~8 editing passes (for a +20k-word piece! ⌛️). Funny enough, adding images and captions seemed to REALLY throw it off. The integrated Substack scanner and the scanner on Pangram’s site also often give you a different number. They’ll even spit out different numbers on different days, even when I haven’t changed anything!

It’s currently sitting at 4% AI because it keeps flagging my “Games Featured” list and “share my work” boilerplate segment as AI, even though I’ve structured both the same way for over four years, since before AI was unleashed into the world (see screenshots for the before and after, as well as the specific part it flagged, then compare that to the example linked here, from early 2022 (which, if you’re smart, you’ll recognize as being before the start of the GenAI era): gameandword.substack.co… )

And yes, I know, I shouldn’t care, I should just do me and to hell with what this machine thinks… and on principle, I agree! But we also know Substack Team is exploring ways to filter out flagged content (they said as much in their announcement), and with Pangram now working on a similar filter for Gmail, these scores could directly impact my discoverability, and since they’ve given precisely zero guidance on what their possible thresholds might look like (is it 100%? 50%? 25%? 1%? Will you get filtered if you opt out of the scan entirely? Who the hell knows! They sure as hell aren’t telling us!), I really have no choice but to dance for the machine if I want my readers to actually read my work.

This thing is a joke. 0.01% false positive rate, my ass.