Notes
The same model, asked two ways
If you paste your product into a chatbot and ask “would people buy this?”, it will usually be kind. That is not the same as research. We checked.
Chatbots are built to be helpful. Helpful often means agreeable. A research panel is built to tell you when an idea is weak. We used the same AI both ways, on the same five products. Only the way of asking changed.
What we did
We picked five made-up products at very different prices: a $4.99 app, a $29 tracker, $85 skincare, $199 software, and an $899 tool for big companies. For each one, we collected ten answers.
Way 1
Just ask the model
One question: on a scale of 1 to 5, how likely is someone to buy this? No customers. No back-and-forth. The model answers as itself.
Way 2
Run it through Synthser
We build a small panel of made-up buyers who match the people you’d actually sell to. Each one is asked alone, in their own words — they never see what the others said. Then we read the whole set.
If “just ask the AI” were enough, both tests would look alike. They don’t. One is a mirror. The other is a room full of people who don’t work for you.
What happened
When we just asked the model, it stayed polite. For two of the five products, all ten answers were the exact same number. It also liked the $899 company tool more than the $29 tracker — which is a strange thing for a buyer to do. No one pushed back. Nothing useful came out of it.
Through Synthser, the same AI argued with itself. Some buyers leaned in. Some walked away. That is what a real group of people sounds like. That mix is the point.
Five products, two ways of asking
| Product | Just ask the model | Through Synthser |
|---|---|---|
| $4.99 budget app | All 10 answers were the same number | People disagreed |
| $85 skincare | All 10 answers were the same number | People disagreed |
| $29 fitness tracker | Polite, clustered together | Mixed — some interested, some not |
| $199 software | Enthusiastic | More cautious |
| $899 enterprise tool | Ranked above the $29 tracker | Treated as a harder sell |
The numbers themselves matter less than this: one column agrees with you. The other does not.
A useful mistake
The first time through, we asked a generic “everyday shopper” panel about all five products — including the $899 company tool. They liked the cheap app most, and the expensive tool least. That is the right answer for those people. Ordinary shoppers are not supposed to want enterprise software.
A chatbot does not know who it is speaking for, so it flatters whoever asked. Synthser only helps if the panel is the people you actually sell to. Asking the wrong room will get you a clean “no.” That is still more honest than a kind “yes.”
What this does not mean
This does not mean you can skip talking to real customers. Synthser is a first draft: a way to find out, quickly, which ideas are worth spending real time and money on.
This was a small test — ten voices per product, one pass. We have not yet put these five ideas in front of thirty real buyers. We are not claiming a magic “they will buy it” score. We are claiming something simpler: the same AI, asked like a chatbot, agrees with you. Asked like a panel, it can say no.
If we cannot show our work, do not trust the number. Every Synthser brief comes with how the study was run. Read that before you believe the headline.
26 September 2026. Five products. Ten answers each. Same AI in both tests — the only change was how we asked. We are not using this page to prove a precise market score. We are using it to show the difference between a yes-machine and a panel.