If you want to know whether Gen Z women in different income brackets would prefer one style of dress over another when shopping for a beach vacation, you could gather up women who fit the description and ask them.
Or you could have artificial intelligence create them for you.
Synthetic consumer research, where AI is used to simulate real consumers, is emerging as a cutting-edge way for companies to better understand how their customers think, minus the actual customers.
The way it operates is not so different from the startups that promise to connect you with a virtual therapist, or a simulated version of your favourite celebrity. Brands define the consumer persona they want to know about to a large language model, and it acts like a real shopper that can answer questions. You can create individual personas and talk to them like you would any chatbot, or generate thousands and survey them to provide a more comprehensive picture.
More sophisticated approaches involve training the AI on a brand’s own data to better reflect its customers’ actual behaviour. PwC’s system, which hit the market a few months ago, not only creates AI agents to answer questions but to ask them, too.
“Envision it like a chat, like you’re talking to a consumer,” said Kelly Pedersen, a partner at PwC who leads the consultancy’s global retail practice. “I can start to ask it questions about its preferences: Do you like short sleeve? Long sleeve? Those types of things.”
While fashion businesses have long used customer profiles with quantitative details like age, income and spending levels to help them make decisions about everything from product design to marketing, Pedersen said one strength of synthetic personas is they allow companies to incorporate a range of qualitative information as well, like consumer preferences and habits, bringing those customer profiles to life.
Getting that level of detail has typically required traditional consumer research, where human shoppers are surveyed about their opinions. AI’s advantage is that it’s faster and cheaper. Assembling panels of humans can take weeks or months and often requires hiring an external research company. But consultancy Bain & Company found that tests with synthetic consumers can take half the time and cost one-third as much.
Industries from political polling to macroeconomic forecasting have jumped to make use of AI-generated research, with fashion and retail now joining them. Large luxury brands are among the clients PwC conducts synthetic research for (Pedersen said he couldn’t disclose any names), while Amazon researchers have proposed their own framework for using the concept in retail. Target is perhaps the only large retailer to have spoken about publicly about using synthetic audiences to help it predict how different groups could respond to campaigns and products before they’re launched.
“This allows our marketing and design teams to test, learn and refine products, promotions and messaging with incredible speed and efficiency,” Michael Fiddelke, Target’s chief operating officer, told analysts and investors during a November 2025 earnings call.
Synthetic consumers aren’t perfect. Because LLMs are prone to offering inaccurate and made-up information, known as hallucinations, synthetic personas can do the same. If personas aren’t carefully defined, they can reflect racial and demographic biases buried in their training data. Data also needs to be refreshed regularly to make sure it matches real consumers’ fast-changing attitudes, said Jane Frost, chief executive of industry association Market Research Society.
“My approach is AI is good if it’s good AI, and synthetic is good if it’s good synthetic,” Frost said. “However, people do get this tendency to be blinded by jargon and tech.”
The hope among businesses, and a growing number of startups like BluePill, Aaru and Deepsona pitching synthetic research, is that the advantages will outweigh the shortcomings.
How Fashion Can Use Synthetic Consumer Research
For fashion brands, AI-powered personas promise a way to get insights into the shopping preferences of different consumer groups.
BluePill, which raised $6 million in seed funding in November, creates narrowly focused “micropersonas” to capture subtle variations within a population, according to founder and chief executive Ankit Dhawan. If a fashion brand were interested in learning whether an ad might resonate with Gen Z, BluePill could create a thousand different micropersonas that represent the entire Gen Z population of the US, and use AI to survey them and analyse the results.
“You can show them an image and ask them if they like it or not, and they’ll react like real humans,” Dhawan said.
Companies can create synthetic consumers based on actual customers, too. In the framework outlined by Amazon’s researchers, which they call persona aligned agentic retail shoppers, or PAARS, a real customer who bought a pair of men’s waterproof hiking shoes and a guide to solo travelling is translated into a persona between 30 and 45, single, likes hiking and camping, has a certain set of brand preferences, is willing to invest in outdoor gear and so on.
The goal, the researchers wrote, is to be able to infer how a shopper thinks and makes decisions, such as gauging their price sensitivity, or how much they rely on reviews or a brand’s reputation. These personas also let them better appeal to underrepresented groups, they noted, which is particularly useful when introducing new products or features.
“By generating synthetic personas that model diverse backgrounds and purchasing behaviors, PAARS enables businesses to proactively design more inclusive shopping experiences,” they wrote.
Another advantage of AI-generated consumers is their accessibility. If you want to know what shoppers in another country think, you don’t need to be on the ground to find out.
“We’ve helped some Japanese brands understand the US consumer at an age level, a preference level — as well as ones that even like to travel to Japan — to understand how they can better tailor assortments, where they want to put their stores,” Pedersen said.
Can AI Really Be a Substitute for Humans?
Whether AI-generated consumer personas can be a substitute for surveying real humans is open to debate.
One study by researchers at Stanford University and Google that created AI agents based on interviews with 1,052 real individuals found the AI was able to replicate the human participants’ responses to survey questions with 85 percent accuracy.
On the other hand, Aaru, a startup that simulates large populations to make predictions, got most of its forecasts wrong in the 2024 US presidential election — but then so did many human pollsters surveying real people. Co-founder Cameron Fink pointed out in an interview with the outlet Semafor after the election that its predictions were within the margin of error, which it considered a success. In December the company raised a Series A round, with some of its equity reportedly achieving a $1 billion valuation.
Frost of Market Research Society has argued publicly that synthetic research can be a useful tool, but real insights still require real people. The value of in-person research is that it’s inherently unpredictable. Data-backed hypotheses might be upended entirely.
“The remoter you get [from real people], the more you miss, and it’s particularly true of emotions,” Frost told The Business of Fashion.
BluePill still surveys human consumers to test how closely its AI’s responses align with those of actual people. Even Amazon’s researchers concluded that synthetic consumers may enhance the efficiency of market research but “cannot fully replace human participation.” There are social and psychological factors AI personas may not capture, not to mention that AI can’t interact with physical objects, meaning there’s no good way to simulate the experience of a person trying on a product. Their guidance was to use synthetic personas to complement human-centred research.
Technology doesn’t have to be perfect to be useful, however. PwC’s Pedersen said the firm’s synthetic personas have been “very accurate” in early testing, but maybe more important, he believes the technology will continue to learn and improve as it’s trained on more data, enabling fashion brands to make better predictions about what their customers will like and respond to.
As with many uses of LLMs, the value of synthetic research may come down to whether it helps companies do what they’re already doing, just smarter and faster.
“Everybody we’ve shown is like, ‘Wow, these insights are very intuitive, they make sense, and it would’ve either been impossible or taken me so much time to get to this level of insight,’” Pedersen said.
