Will synthetic respondents replace real consumer research? What researchers and vendors told us in 2026

Will synthetic respondents replace real consumer research?
What researchers and vendors told us in 2026
Written by: Mark Nguyen, ND Research
Originally Published: October 23, 2026
Labels: Field Notes, AI-assisted, Continuously Maintained
Topics: Synthetic data, Synthetic respondents, Digital twins, Market research, AI
TLDR: Not yet. In our 2026 conversations, research firms were curious but cautious, and the synthetic-data vendors themselves said their products need real, observed data underneath.
Key Ideas:
- 4 of the 19 industry conversations we recorded in 2026 were skeptical of synthetic respondents. The skepticism came from research firms and from the synthetic-data vendors themselves.
- A research-agency founder whose firm invested in synthetic data and digital twins told us clients were curious but few were buying.
- The synthetic-data vendors we spoke with drew their own lines: boost small segments inside a big enough real sample, don't use it to model what drives choices, and don't invent things nobody observed.
- Where we heard support, it came with a condition: synthetic, grounded in observed data.
- Our read: synthetic respondents are a multiplier, and a multiplier needs something real to multiply.
Preamble
Synthetic respondents are the topic that comes up at every conference now, usually with a demo and a big claim. We went into this year's conversations expecting to hear that they were eating the survey business. That is not what we heard, and interestingly, the most careful people were the ones selling the stuff.
Here's what they told us, with the obvious caveat that we sell observed data and have a horse in this race.
Are synthetic respondents replacing real consumer research?
Not in the conversations we had this year. Interest is real and buying is cautious, and some of the most careful voices were the vendors themselves.
Between April and October 2026 we recorded 19 conversations with agency, brand, research, measurement and data professionals (we break down the full set in our notes from 19 industry conversations [1]). Synthetic data or AI hype came up in 6 of them. 4 of the 19 were skeptical of synthetic respondents specifically, and 2 more were wary of AI hype in general. Two of the four were with synthetic-data companies.

4 of 19 - conversations skeptical of synthetic respondents. Two of them were with synthetic-data vendors.
What do research firms think of synthetic data?
Curious, and not yet convinced.
The founder of a US research agency told us his firm had put real effort into synthetic data and digital twins over the past couple of years. Clients were curious, he said, but few wanted to pay for it. He also said it's hard, especially with AI, to tell hype from reality. In his day-to-day, AI is a productivity tool, not the thing that changes the work.
A measurement leader at a global research firm was sharper. He had no interest in synthetic respondents built from a language model with nothing real underneath (his example was inventing a synthetic doctor to interview), and called that kind of work bad research. But he wasn't against synthetic data as such. His firm does model up from a real metered panel. His position was basically: synthetic, grounded in reality.
What do synthetic-data vendors say their products can't do?
More than you'd expect, and honestly, we found that reassuring.
The product lead at a synthetic-data startup described their approach as statistical boosting for small segments, not a chatbot answering survey questions. The guardrails he described:
- It needs a real sample underneath. At least a couple of hundred real respondents, with the segment being boosted making up only a modest share of them.
- It helps small segments most. Once a segment already has roughly 100 to 150 real people, the gain from boosting gets small.
- It isn't for driver modelling. Synthetic data doesn't behave under the same assumptions as real data, so they don't use it for driver analysis or similar choice modelling.
The founder of a synthetic-survey company made a similar point from a different angle. His product treats a client's real data as strong priors and aims to behave like a hundred more respondents to the same survey, not like a probability sample of a country.
He said plenty of providers simply wrap a general-purpose chatbot, and his company's view is that this isn't enough. He also said people regularly arrive with five real respondents and hope to leave with a thousand.
When the idea of generating what people see in their social feeds came up, his answer was no: without real signal underneath, he didn't think it would be truthful to present the result as someone's feed.
When does synthetic data make sense in market research?
Putting together what research firms and vendors told us, roughly:
- Good fit: steadying a read on a small segment inside a decent real sample. Think: you surveyed 400 people, 60 of them are in the group you care about, and you need a less jumpy number for that group.
- Good fit: extending a real panel whose behaviour is actually observed, like a metered or measured panel.
- Poor fit: replacing fieldwork with an audience you've never actually measured.
- Poor fit: modelling what drives people's choices.
- Poor fit: inventing behaviour nobody observed, including what people were shown in their feeds.
Why does observed data matter more as synthetic data grows?
Because every synthetic respondent is only as good as the real data it was built from. The more synthetic answers get used, the more the real data underneath becomes the bottleneck.
We've also seen interest from the other direction: at least one synthetic-survey company wants to test whether observed feed-exposure data [2] makes synthetic respondents more accurate. That's an early conversation, and we have an obvious interest in the answer, so weigh our read accordingly.
That's our read (talking out loud, work in progress here), and we hold it loosely. The tools are improving quickly, and we could be wrong about how fast this moves. But in 2026, the people closest to the work kept setting the same condition: ground it in real data. Our shorthand is that synthetic data is a multiplier, and you still need something real to multiply.
How we gathered this
- Source: 19 recorded conversations with 17 distinct people and organizations (agency, brand, research, measurement, advisory and vendor), April 1 to October 20, 2026 [1].
- Synthetic data specifically: 4 conversations were skeptical of synthetic respondents, 2 of them with synthetic-data companies. 2 more were wary of AI hype more broadly.
- Limits: a small, mixed sample, not a survey. Several participants are advisors to us. Treat this as field notes, not a market estimate.
- Disclosure: we sell observed feed data, and we've discussed a research collaboration with one of the synthetic-data companies quoted here.
- Privacy: people and companies are described by role only, and statements are paraphrased from private conversations.
FAQ
What is a synthetic respondent?
A survey answer generated by a statistical or AI model to stand in for a real person, usually to fill out small segments or to answer new questions without new fieldwork.
Can synthetic data replace consumer surveys?
Not on its own, based on what we heard in 2026. Vendors themselves describe it as a way to boost real samples, not a substitute for them.
What is a digital twin in market research?
A model of a person or segment built from real data, used to answer questions that person or segment wasn't directly asked. How good it is depends on the real data behind it.
How many real respondents do you need before boosting with synthetic data?
One vendor's rule of thumb: at least a couple of hundred real respondents overall, with boosting reserved for segments too small to read on their own. Once a segment has roughly 100 to 150 real people, the extra value of boosting gets small.
How to cite this
Nguyen, M. (2026, October 23). Will synthetic respondents replace real consumer research? What researchers and vendors told us in 2026. ND Research. https://www.nguyendigital.co/blog-posts/will-synthetic-respondents-replace-consumer-research
List of References
(Continuously Maintained)
[1] ND Research. Nguyen, M. (2026, October 23). What does the insights industry actually struggle with? Notes from 19 recorded conversations in 2026. 19 recorded conversations with 17 people and organizations, April 1 to October 20, 2026. The four conversations counted as skeptical of synthetic respondents were with a measurement leader at a global research firm (September), the chief product officer of a synthetic-data startup (May), the founder of a synthetic-survey startup (September) and the founder of a US market-research agency (September). The appendix of that article lists all 19 by month, role and context. https://www.nguyendigital.co/blog-posts/what-strategists-and-insights-teams-struggle-with
[2] ND Research. Nguyen, M. (2026, October 23). What is feed exposure intelligence? The missing layer in consumer research. https://www.nguyendigital.co/blog-posts/what-is-feed-exposure-intelligence





