My AI wrote that titanium won't dent like steel. A second AI with web search disagreed.

I was writing a buying guide for automatic field watches on Watch To Wear. The AI writer gave me a clean draft, and somewhere in the middle was this sentence: titanium "won't dent like steel."

It sounded right. It was stated with total confidence. (Of course it was.)

I put the claim into Google, and Google's own AI answer told me the opposite: steel is generally harder and dents less than titanium. Now two AIs disagreed with each other, and I had to decide who to believe.

That's the problem I built a fix for. After SyndiCalito writes an article, a second pass searches the web for the claims writers most often get wrong (specs, numbers, comparisons, history). It shows me every change right next to the original sentence, with the reason, a link to its source, and an Undo button.

On the dent claim, the answer turned out to depend on the grade. Grade 5 titanium is harder than typical watch steel, while grade 2, the more common and cheaper one, is softer and shows marks more easily. My draft had picked one side of that, and so had Google's summary.

Here's what surprised me. I ran it three times on the same article and got three different answers on that one claim. A single pass can't settle a fact that depends on a condition. So now, when the second pass keeps a claim that depends on one, it lists it under "Worth knowing" with the exact condition and a source, and the decision stays with me.

It also caught that my field watch sizes ran large (a few classics sit at 34 to 38mm, like the Bulova Hack and the Marathon General Purpose), and it gave me the accuracy numbers behind a daily range I'd written as "10 to 20 seconds." It also lists the claims it couldn't confirm.

(Tested on Watch To Wear, October 12, 2026. A run takes about a minute and a half and a handful of web searches.)

It's part of the Pro and Agency plans.

What's the last thing an AI told you with total confidence that turned out to be wrong?

Carl