Synthetic market research works with AI models that respond to ideas, products and messages the way customers would. It promises answers in hours rather than weeks. More interesting than the speed is the question of how far these answers can be trusted.
In marketing, I work with a team of 22 people across several European markets. Fast insight before a decision is valuable, as long as it stays clear how robust it is.
- AI personas prepare decisions; they do not make them.
- Only your own data makes their answers specific.
- Fine differences between target groups are easily lost.
- The benchmark remains the judgement of real people.
Contents
What synthetic market research can do
An article in Harvard Business Review describes how large language models can act as synthetic customers, responding to product concepts and reaching conclusions similar to surveys or focus groups, only faster (see here: Harvard Business Review, Using Gen AI for Early-Stage Market Research). The authors recommend fine-tuning the models with your own data and using them to complement real research, not to replace it.
This matches my experience: for filtering the most promising ideas out of many, synthetic market research is a smart tool. Which idea carries a season is ultimately decided by real people.
At the beginning, not the end
Synthetic market research belongs early in the process. It helps condense ten concepts into three, sharpen wording and test questions for a later survey.
One thing holds: a general model does not know your customers. Only with your own data, from earlier surveys or the CRM for instance, do the answers become specific. Without that foundation, you get the average of the internet.
Where the nuances get lost
AI personas smooth over differences. The fine distinctions between people who have worn a brand for years and those just discovering it are captured only vaguely. Yet it is precisely these nuances that often decide claim, imagery and channel.
Real people as the benchmark
Every result from synthetic market research deserves to be checked against real answers. Those who put both side by side learn over time where the model is right and where it errs. How your own data can be used for forecasting is something I describe in my article on predictive analytics.
A team might first test four campaign ideas with AI personas. Two fall clearly behind, two convince. Only those two go into a survey with the target group. The team gains time without leaving the decision to a simulation.
Synthetic market research is a fast filter, not a verdict. With your own data, a clear place at the start and a check against real people, it makes research faster without replacing it.