A few weeks ago an AI tool handed me back a paragraph that sounded completely confident, totally polished, and was just wrong. Not obviously wrong, the kind of wrong that slides right past you if you’re in a hurry. That moment is basically this whole episode. AI is a genuinely useful tool, but the same thing that makes it fast and confident is exactly what makes it dangerous to lean on without checking.
Logistics AI has always done well. Recommendation engines, calendar assistance, first-pass drafts, these are the same category as auto pay or a calendar reminder: friction removed, nothing lost. AI is genuinely great at the blank page problem, at restructuring a messy voice memo, at comparing two spreadsheets, at giving you a first draft of show notes.
Where AI is different from a calendar reminder. A calendar reminder never tries to be right about facts. It just tells you the time. AI will state a false thing in the exact same confident tone as a true one. I’ve caught it misidentifying birds on my nature podcast and inventing plausible-sounding history on my Bible podcast. Unlike a human expert, who’ll tell you when they’re not sure, AI states everything with equal confidence unless you specifically prompt it not to.
The Dome of the Rock thumbnail story. I asked for an image of Jesus looking over Jerusalem, and AI put a mosque that wasn’t built until roughly 600 years later into the scene. It’s a good drafting partner, not a source of information, and that distinction matters.
The context problem. I share a story about someone who had AI rewrite an internal email for a different audience without checking it, and the tool didn’t know which details didn’t apply to the new group. The email went out confused, and leadership had to send a correction. The failure wasn’t really AI being wrong, it was nobody reviewing the output before it went out under someone’s name.
Three questions before you publish anything AI touched. Can I verify this myself, especially any hard fact, date, or name? Does it still sound like me, or has it turned into smoothed-out generic internet voice? And would I be comfortable if someone traced this sentence back to me? I tell the story of an early Substack draft where AI invented a detail in a personal story about my dad during a snowstorm when I was eleven. No one else would have caught it. I did, because it was my story.
How I actually use it. I start with my own draft, let AI suggest structure and flow, then rewrite it myself, then let it help with grammar. Different creators layer it differently, but the point is the same: your voice in your own project has to stay yours.
One resource: look up “automation bias,” the research term for our tendency to over-trust confident-sounding systems.
Jill’s Links
https://jillfromthenorthwoods.com/ - Find my podcasts, videos, downloads and ways to contact me
https://www.buymeacoffee.com/smallstepspod
https://x.com/NorthWoodsJill
Email the podcast at [email protected]
By choosing to watch this video or listen to this podcast, you acknowledge that you are doing so of your own free will. The content shared here reflects personal experiences and opinions and is intended for informational and educational purposes only. I am not a software developer, data scientist, or AI professional. Any tips, tools, or suggestions offered should not be considered a substitute for professional technical advice. AI tools and platforms change frequently — always verify current features, pricing, and terms directly with the providers. You are solely responsible for any decisions or actions you take based on this content.
