Research insights
From survey to shelf: the life of a new product idea
Every new product on a store shelf started as an idea. Learn how survey takers shape new items from the earliest concept to the final launch.
Synthetic respondents: can AI replace real opinions?
A new idea is being discussed across the research industry: using AI to simulate survey answers instead of asking people. These simulated answers are called synthetic respondents. Here is a balanced look at what they can and cannot do, and why your opinion still matters.
Key points
- Synthetic respondents are AI models prompted to answer questions as if they were a certain type of person.
- They can be useful for early brainstorming and testing a questionnaire before it goes live.
- They cannot have real experiences, and they tend to repeat patterns from their training data, including its gaps and biases.
- Decisions that affect real customers still need evidence from real people.
What synthetic respondents are
A synthetic respondent is a simulated participant. A researcher gives an AI model a description, for example "a 34-year-old parent in Chicago who shops online weekly", and asks it to answer a survey or take part in an interview in that role. The model generates plausible answers based on patterns in the vast amount of text it was trained on.
Some tools go further, combining AI models with real survey data to produce simulated answers for groups that are hard to reach. The idea has attracted attention because it promises quick, cheap answers at any time of day.
Where they might help
Used carefully, synthetic responses may have a place in the research process:
- Testing a questionnaire. Running a draft survey through a simulation can reveal confusing questions or broken logic before real people see it.
- Brainstorming. Generating a wide list of possible reasons, ideas or objections that researchers then test with people.
- Early exploration. Getting a rough sense of a topic before designing a proper study.
In each case, the simulation is a starting point, not the answer.
Why they cannot replace real people
No real experience
An AI model has never tasted a new snack, waited on hold with a bank or struggled to open a medicine bottle. It can describe what such experiences are usually like, but it cannot report a new reaction to something it has never encountered. New products are, by definition, new.
The average is not the person
AI models tend to produce typical, reasonable-sounding answers. Real people are surprising. They have contradictions, strong feelings and unusual habits, and those details are often where the most useful insight sits.
Gaps and bias
Models learn from the text they were trained on. Groups that are underrepresented online, such as some older people, people in rural areas, people who speak less common languages or people with disabilities, may be poorly represented in simulated answers. Relying on synthetic data can quietly exclude the very people research should include.
Hard to check
When a real person answers, a researcher can follow up, verify the answer and understand context. When a model answers, it can sound confident even when it is wrong, and there is no person behind the answer to ask.
Real voices, real decisions. If a company is deciding whether a product is safe to use, easy to understand or worth its price, it needs to hear from the people who will actually use it.
What it means for panel members
You may hear that AI will replace surveys. We do not believe it will, at least not for decisions that matter. More likely, AI tools will take over some routine tasks, such as checking questionnaires, while research with real people becomes more focused on what simulations cannot do: genuine reactions, lived experience and the unexpected.
That makes careful, honest participation more valuable, not less. We wrote about the broader changes in AI in market research.
A note on your own answers
Simulated answers have one more effect: they make it more important than ever to tell genuine responses from generated ones. Research teams are using stronger checks to detect AI-written answers in surveys. Writing in your own words, even briefly, is the best way to show you are a real person with a real view.
Our position
Reward Hours exists to connect real people with research. We are interested in tools that make research better, and we will be open about how they are used. But we will keep making the case that decisions about people should be informed by people.
Common questions
Are companies already using synthetic respondents?
Some are experimenting with them, usually alongside traditional research. Practices vary widely, and the industry is still debating standards.
Will this mean fewer surveys for members?
Not for studies that need genuine experience. Some simple tasks may change, but research that informs real decisions still needs real participants.
Is my data used to train synthetic respondents?
Your data should only be used as described in the privacy notice you agreed to. Responsible research companies do not use participant data to train AI models without a lawful basis and clear transparency.
