Why AI invents confident statistics
By Chrysti Reichert, independent AI trainer in Central Florida • Published
"91% of people will trust a statistic if it has a decimal in it."
My AI made that number up. Told me so mid-sentence. "According to a 2019 Harvard study that I am inventing right now."
And honestly? It had a point.
Stick a decimal on a claim and it grows a little lab coat. 91% sounds researched. "Most people" sounds like a guess. Same sentence, fancier outfit.
That is not the robot being sneaky. It watched us trust the lab coat for years and picked up the bit.
So next time a number struts by in its little decimal, ask it the one thing it cannot answer. Says who? That single question is the cheapest data-literacy habit your team can build, and it catches most of the damage before it lands in a slide deck.
Questions teams ask before booking
Yes. It will produce a precise-sounding number with no real source behind it. This is a hallucination, and it is most dangerous when it looks researched. A made-up "91.3%" reads as more credible than an honest "most people," which is exactly the problem.
Stick a decimal on a claim and it grows a little lab coat. We have trusted precise-looking numbers for years, so the model learned the bit. The format signals rigor even when the number is invented. Same sentence, fancier outfit.
Build one habit: ask the number the thing it cannot fake, which is says who. Require a real source for any figure that informs a decision, and treat an unsourced statistic as a draft, not a fact. This data-literacy habit is part of every workshop. Vendor-neutral, Tampa, Orlando, and Lakeland or virtual, flat-fee.
Keep reading
Want your team to stop trusting the lab coat?
Tell me where numbers drive your decisions. I will show you the data-literacy habits that catch the invented ones. Independent, flat-fee, no upsell.