I have spent a good part of my working life trying to figure out what consumers want. And for the longest time there was only one way to do it: go ask them. Surveys, focus groups, store visits, long interviews. It was slow, it was expensive, and a lot of what you heard was plain noise. (The best answers usually came when you least expected them!)
But here’s what I keep coming back to lately: we don’t need to ask them anymore. Or at least, that’s the pitch.
Synthetic consumers. You describe a person to an AI – her age, her income, her habits, what she buys – and it answers your survey as her. Then as a thousand others like her. No recruiting, no incentives, no six-week wait. Results before your coffee gets cold. With companies spending over $130 billion a year on market research, no wonder everyone is leaning in. And to be fair, the answers are good. Scarily good sometimes.
Think about it this way though: where do those answers come from? From everything real people have already said, bought, reviewed and posted. A synthetic consumer is an echo of them. A very sharp echo, but an echo. It can tell you what people like her have thought so far. It can’t tell you what she will think next, because she hasn’t thought it yet.
Take GLP-1s. A few years back, ask any model how people would eat and snack in 2026 and you would have got sensible talk about protein, convenience, price. Nobody’s model had a consumer who simply felt less hungry because of a weekly injection. Today that consumer is changing what goes into the shopping cart in millions of homes. The echo was faithful. The voice had just not spoken yet.
There is a second, quieter problem. Models are built to find the pattern, and the pattern is the middle. But anyone who has worked on a new product knows the middle is rarely where the next idea is hiding. It’s with the odd ones… the person who uses your product “wrong”, the one who complains about something nobody else noticed. A model treats them as outliers. A good researcher treats them as a lead.
It is not the same – reading a thousand reviews of a restaurant versus eating there once.
So the question for me is not synthetic or real. It’s what the synthetic is standing on. If you have years of genuine listening behind you, these tools multiply it. If you don’t, they are just confidently repeating the internet back to you.
And this is the part that worries me. Synthetic is cheap, and cheap things get used to cut budgets. One less round of groups, a smaller tracker, fewer days in the market. Every cut makes sense on its own. Add them up and you get what I’m calling “listening debt” – the growing gap between what your models believe about consumers and what consumers are actually doing. You won’t see it on any dashboard. You will see it when a launch that tested beautifully sits on the shelf.
It’s the same irony I wrote about with SEO, isn’t it? When machines made content cheap, trust became the scarce thing. Now that machines make consumer answers cheap, real listening becomes the scarce thing.
By all means, use the tools. They are great for narrowing forty ideas down to five, and for a quick gut check before the real work begins. Just don’t let them be the only voice in the room.
The machines can tell you what your consumer has said. Only she can tell you what she’ll say next.