04/27/2026
Why Ronald Reagan's Words, "Trust but verify," Apply to AI Today.
I can't yet trust AI... for anything.
My first novel (Operation Hoplon: The Joshua Booker Origin Story) has a long history. I started it around 2016, set it aside when life intervened, picked it back up in 2022, and finally finished and published it in 2024. By then, ChatGPT existed, but I had never touched it. Artificial intelligence, for all practical purposes, wasn't part of my world.
Now, deep into my second novel (about 90% finished), I've become more familiar with what AI can and can't do. I've even started using a product called Grammarly, though not for ideas or phrasing. I use it strictly for consistency in punctuation. High school taught me one set of comma rules, college introduced me to AP style, and now the Oxford comma reigns supreme. Three competing standards are two too many for me to keep straight, so I let Grammarly help with that.
That's where things got interesting and very troubling.
I noticed that Grammarly has added a new tool to check for plagiarism and AI-generated content. Out of curiosity, I ran my second novel through it. The verdict: 40% AI-written. That is terribly wrong, as it should have been 0%. So I ran my first novel through the same check, the one I finished before AI writing tools were even a realistic option. It came back 31% AI-written.
That result sent me down a rabbit hole. I tested both books across several AI-detection tools. The verdict on my first novel ranged from 13% to 77% AI-generated, depending on which tool I used. One declared it 100% plagiarized, apparently having "learned" the published text and assumed I had copied it rather than written it.
I wasn't alone in noticing this problem. While researching, I found a writer who ran the opening chapter of Mary Shelley's Frankenstein through multiple AI detectors. One tool declared it 100% AI-generated. For that to be true, artificial intelligence would have needed to exist before 1817, when Shelley completed the manuscript.
So here's the question I can't shake: if AI tools are this unreliable at detecting AI writing, why would I trust them for anything else? When I do use AI for research, I make a point of spot-checking results in areas I know well enough to verify. So far, that habit has proven wise.
Ronald Reagan's old line, "Trust, but verify," turns out to apply just as well to algorithms as it did to arms treaties.
AI is improving exponentially every year - even every month. I am sure that one day soon it will produce great, accurate results. But as of this moment, April 27, 2026, I simply don't have much confidence in getting information I can consistently trust.
Have you encountered your own examples of AI getting things embarrassingly wrong? I'd genuinely like to hear them.