I once caught myself about to repeat a statistic I genuinely believed was true, until I tried to find where it actually came from and couldn’t. It had just sounded right, and I’d heard it enough times that it felt established. Sounding right and being true are not the same test, and I almost failed to notice the difference on something I was about to put my name on.
Why every statistic your AI hands you needs a source attached.
A number with no source is indistinguishable, on the page, from a number that’s completely made up. That’s the uncomfortable core of it. “73% of customers prefer X” reads exactly as authoritative whether it came from a rigorous study of ten thousand people or whether a language model assembled a plausible-sounding percentage out of thin air three seconds ago. The confidence is identical. The formatting is identical. The only thing that separates a real stat from a fabricated one is whether you can trace it back to somewhere real, and if you can’t, you have to treat it as fabricated, no matter how much you’d like it to be true.
This matters more with AI specifically because of a quirk in how these tools behave. A human who’s genuinely unsure of a number usually signals it, they hedge, they say “I think it was around,” they trail off. That hesitation is useful information, it tells you to double-check. AI often delivers a fabricated number with the exact same fluent confidence as a real one, no hedge, no tell, because it doesn’t experience uncertainty the way a person does. You lose the natural warning signal you’ve relied on your whole life. So you have to replace that missing instinct with a hard rule, because your gut, which is calibrated for human tells, will not save you here.
And the stakes aren’t abstract. A fabricated statistic in your marketing is a small time bomb. Most of the time nobody checks and nothing happens. But the one time a knowledgeable customer, a competitor, or a journalist does check, and finds a number you can’t back up, you don’t just lose that one claim. You lose the benefit of the doubt on everything else you’ve ever said, because now you’re a business that publishes numbers it can’t support, and people extrapolate. One unsourced stat can quietly cost you the credibility of every true thing around it.
The discipline is simple, even if it’s a little tedious: if you can’t link it, don’t publish it. If a stat is worth putting in front of customers, it’s worth the two minutes to confirm the source actually exists and actually says what you think it says. This is the single cheapest insurance policy in all of marketing, and almost nobody buys it.
Takeaway: If you can’t point to where a number came from, don’t use the number.
Try this: Before you publish anything with a statistic in it, click through to the source and confirm it actually says what you think it says. Not that the source exists, that it says the specific thing you’re claiming. You’ll be surprised how often a real source got summarized into something it never actually said. This takes two minutes and saves you from a genuinely bad day later.
Talk soon, Tiffany