Yesterday evening, I finished writing a chapter in my current WIP. When I opened my social media this morning, there was an avalanche of posts about the current AI witch hunt. I know I shouldn’t have, but I couldn’t resist running a little test.
The AI checker I have been reading a good deal about is TextGuard. I copied the text of my chapter’s first draft, totally new, totally unedited, totally me, and ran it through the TextGuard AI checker. Here is the result:
So, the report states that the text was 60% AI generated. However, I knew that to be a false positive. Following that, I had to download the App because I wasn’t allowed to paste anymore text into the online version (which is supposed to be free).
I downloaded the app, removed the three instances of en dashes (signifying a longer pause than a comma) and the one instance of an em dash (signifying an interruption) and ran it again. Here is the result:
I made no other changes. Only removing the dashes, and yet, the report went from 60% AI generated to 9% AI generated. How can the use of a few dashes signify a document has been 51% generated by AI? I am afraid that I can only draw one conclusion: the claim to be a “comprehensive AI detector” is a marketing ploy. Their website also claims the tool is 100% free, but in fact requires a subscription.
As authors, we are already under serious assault because the market is flooded by AI generated slop. To now have that exacerbated by witch hunts caused by tools that claim accuracy when they are anything but accurate, using mainly em dash and en dash to detect AI, which was an idea bandied about on social media because of a sudden surge in their usage by non-professionals. Professional writers have always used them and it is mostly non-professionals who see them as a sure sign of AI.
Of course, the side of the story that I have rarely seen mentioned is that the Large Language Models used to generate AI text were stolen from original writers (myself included: A Prelude to War and Anvil were included in the Anthropic theft). So, when I run my text through an AI checker, is it finding writing traits that are there because I wrote some novels that were included in their LLMs? It seems quite plausible.
No doubt, the developers will say TextGuard reports are only an indicator of possible AI generation, but that isn’t how they are marketing the app. TrustPilot has (at time of writing) 1395 reviews, most of them positive. I sincerely hope that they take a look at TextGuard, because they must be bots or other such scams.
PerchedCrowPress 