The Trouble With Asking People What They Want Is That They'll Tell You

Kate O’Keeffe is CEO and co-founder of Heatseeker, the AI-native platform for live-market customer validation.

gettyMy father made his first confession at 7 and hit a snag. He couldn’t think of a single sin.

I know my father. He was, and remains, thoroughly spoiled and entirely capable of misbehavior. Supply wasn’t the issue. Disclosure was.

So he improvised. He told the priest his mother had served meat on a Friday, and he’d eaten it.

My grandmother was incandescent when she found out. Her son had lied to a priest and framed her for it. She was still bringing it up 60 years later. Dad, meanwhile, gave his interviewer a satisfying answer and went out to play.

I think about my father every time someone shows me a survey.

People are hopeless narrators of their own behavior. We want to be helpful, we want to sound reasonable, and occasionally we’d rather not admit what we’ve been up to. Even when we’re perfectly sincere, we’re terrible at predicting what we’ll actually do.

A 2024 study in the Journal of Industrial Economics worked with an online supermarket to see how shoppers responded to discounts. You’d think telling someone about a discount would get them to buy the discounted product.

Not necessarily. Shoppers who got the sale email went straight for the products they always buy, discounted or not. Sales of full-price items in the promoted categories rose 44%.

Then came the survey. Ninety-one percent said they’d choose organic if it cost the same. Eighty percent said they’d switch brands for 20% off.

When it came to actual shopping decisions, habit won.

Apparently our deeply held commitment to a bargain occasionally loses to the inconvenience of thinking about which bananas to buy.

A 2026 study in the Journal of Accounting and Economics found the same pattern across more than 24,000 American households. Telling people about companies’ good environmental and social practices lifted their intention to buy. Their actual spending barely moved.

Which brings us to the sustainability-conscious shopper browsing Shein at midnight.

She cares about the environment. She really does. She also wants something fabulous for Saturday, has three children asleep upstairs and has just discovered free shipping.

Both things can be true. Her survey answer goes to the planet. Her credit card goes to Shein.

And we’re about to make this problem a lot bigger.

You can now spend 45 minutes with a very patient chatbot discussing your relationship with a moisturizer. It will probe your motivations, ask thoughtful follow-ups and help you unpack your feelings about hyaluronic acid.

You’ll leave with a beautifully articulated skincare philosophy. You’ll still buy the one with the nice lid.

Will it predict what you buy? A review of more than 400 studies found that what people said they’d do explained only about 28% of what they actually did. A friendlier interviewer doesn’t change that.

Synthetic research goes one step further. Skip the humans entirely and ask AI-generated personas what people might think or do.

Large language models are excellent at producing plausible answers. So was my father at 7.

Most synthetic personas learn from what people say, so they inherit the fibs. At Heatseeker, the company I founded, ours learn from what people actually did in live experiments. One gives you faster confessions. The other lets you watch the dinner.

Here’s my prediction. By 2030, customer research as we know it will be gone, replaced by continuous experiments with real people.

Want to know whether people care more about convenience or sustainability? Put both in front of comparable audiences and see which one they click. Want to know whether they’d buy it? Let them try.

We saw exactly this with a food delivery company. The obvious play was 30% off. Everyone loves a discount. Customers have said so in every survey since surveys were invented.

We tested it against a full-price offer that simply reminded people they could add a new delivery address.

Nobody would have ranked “new delivery address” above “30% off” in a focus group. Nobody would have put it in the questionnaire. Which is exactly the point.

That’s what gets me out of bed. Not running research faster, but finding the things you’d never have thought to ask. Now imagine AI generating dozens of hypotheses like that, testing them in market, and designing the next experiment before anyone has commissioned a research project.

I still love talking to customers. Some of the most interesting things I’ve learned about people came from listening to them.

But asking someone to predict a decision we could simply let them make is backward. Using those predictions to decide which ideas deserve a test is worse.

The companies that understand their customers best in 2030 will be the ones most willing to be surprised by them. They’ll still need human curiosity and judgment. But they’ll spend far less time asking people to predict their behavior, and far more time giving them the chance to reveal it.

So before you commission your next survey, ask one question: Could we just let customers make this choice instead?

My grandmother spent 60 years protesting her innocence because a 7-year-old was asked a question he didn’t want to answer.

If the priest had simply dropped by for dinner on a Friday, he’d have seen the fish. The only sin on the table would have been how Nan cooked it.​

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?