Artificial intelligence can produce convincing answers about consumer behaviour, but those responses should not be treated as proof of what real people think, feel or experience, according to qualitative research expert Raquel Torres.
Torres, founder and chief executive of Focusquali, argues that the growing sophistication of synthetic consumers is creating a methodological challenge for researchers, particularly when AI-generated statements are presented as findings from human participants.
In an article published on November 4, Torres said the distinction between predicting a human response and observing one is becoming increasingly important as businesses adopt AI to explore consumer attitudes and preferences.
The Brandspur Brand News reports that the debate has implications for companies relying on consumer research to shape products, marketing strategies and brand positioning, as synthetic data become more capable of producing responses that resemble those of real people.
Torres said synthetic consumers can offer useful insights into how people might respond to a product, brand or situation. However, a model generating a plausible answer does not establish that a consumer actually experienced the feelings or circumstances described.
She explained that qualitative research involves more than collecting statements. Researchers also examine how people arrive at their views, including the influence of personal experiences, social relationships, memories and the research environment.
A consumer’s response may change during an interview as questions prompt new memories or other participants introduce different perspectives. Such shifts can reveal tensions between what people say about themselves and how they behave.
For example, a participant may insist that price does not influence purchasing decisions, only to later explain that a product was abandoned after a price increase.
Torres said the apparent contradiction could offer valuable insight into how consumers understand their own choices. It may reveal differences between their stated priorities and actual behaviour, or show how they interpret the value of a product.
If an AI system generates the same two statements, the wording may be identical, but the evidence is different. The model has produced a plausible contradiction, while the human participant has expressed one within an actual research encounter.
The researcher cautioned against treating synthetic data as inherently unreliable. She said such tools can support exploratory work, help researchers develop hypotheses, test questions and examine possible interpretations before conducting interviews with real participants.
Synthetic consumers can also help researchers explore scenarios and identify issues that deserve further investigation, potentially reducing costs at certain stages of a research project.
The concern, she said, arises when simulated responses are presented as evidence of real consumer experiences.
An AI-generated persona may describe a product preference in convincing detail, but the fluency of the response does not establish that a person made the decision or experienced the emotion being described.
Torres argued that researchers should be clear about the type of evidence they are using and the conclusions it can support.
Her position is that synthetic data and human research can complement one another, but they serve different purposes. AI can help identify what consumers might plausibly say, while research involving real people allows investigators to examine how opinions and meanings emerge.
As synthetic consumers become more convincing, Torres said, researchers must pay closer attention to the difference between a plausible answer and an observed human experience.
