Synthetic Audiences Reshape Market Research

Market research has always involved a trade off between confidence and speed. A large study can produce strong evidence, but recruiting participants, conducting interviews, analyzing responses and turning findings into recommendations can take weeks. For teams making smaller product, messaging or positioning decisions, that timeline is often difficult to justify against the pace the rest of the business moves at. Synthetic audiences offer another option. Instead of recruiting people for every question, researchers can model likely users based on real world audience characteristics and use those synthetic users to explore decisions quickly. The important question is not whether synthetic research can replace human research. It is where synthetic research can provide useful evidence in the space between larger studies.

Defining synthetic audiences in market research

Synthetic audiences are simulated groups designed to represent specific customer segments, user types or ideal customer profiles. They are generated from behavioral and demographic context rather than recruited as individual participants for each research session, which is the detail that separates a genuine methodology from a novelty.

That distinction matters more than it first appears. A synthetic audience is not simply a collection of generic AI responses produced on demand. A useful system needs to model how a particular audience might react to a question, message, product concept or experience, drawing on patterns specific to that segment rather than a generic simulated consumer.

Consider a software as a service (SaaS) company weighing two landing page headlines. It could use synthetic users representing its target customer profile to identify differences in perceived value, clarity, likely objections and purchase motivation between the two options. The result is not proof that customers will behave exactly that way once the page goes live. It is evidence a team can weigh before committing budget or launching a broader study, which is a meaningfully different claim than the one synthetic research vendors sometimes imply.

Harvard Business Review's research on generative AI in early-stage market research frames the technology in similar terms, describing it as a tool for simulating customer responses to product concepts that can produce conclusions similar to a survey or focus group with far less time and expense. 1 That framing sets a useful expectation from the outset. Synthetic audiences are a research input, not a verdict.

Validating synthetic research against human benchmarks

The strongest case for synthetic research comes from validation rather than speed alone. Articos, one vendor operating in this space, reports that its methodology has been peer reviewed and reached 86 percent human accuracy across 46 studies, with benchmarking against research standards associated with the Baymard Institute and Nielsen Norman Group. Baymard's own methodology illustrates what a credible benchmark looks like in practice:

its research draws on more than 4,400 moderated usability sessions and 54 rounds of manual benchmarking across 344 ecommerce sites, producing hundreds of thousands of weighted performance scores. 2

Comparing a synthetic methodology against that kind of rigor is a meaningfully higher bar than comparing it against internal accuracy claims alone.

That distinction matters for insights teams evaluating any synthetic research vendor. Synthetic research should be evaluated against human research and established research benchmarks, rather than assuming an AI-generated answer is automatically representative of a real audience simply because the underlying model is sophisticated.

The validation also helps define a more useful role for synthetic audiences going forward. They can be treated as an additional research layer for questions where directional evidence is valuable, while higher stakes decisions can still warrant direct human validation before a team commits significant budget or reputation to the outcome. Framed that way, validation is not a one-time credential a vendor earns and moves past. It is an ongoing discipline research teams should keep applying as methodologies and models change.

Using synthetic audiences between major studies

The most practical use of synthetic audiences is often the space between formal research projects. Large studies may answer important strategic questions, but teams make dozens of smaller decisions before and after those studies and most of those smaller decisions never get any research attention at all under a traditional timeline.

Marketing teams might test alternative headlines, ad copy or value propositions before a campaign locks. Product marketers can explore positioning before investing in a full campaign build. Product teams can compare landing page concepts or identify potential objections before shipping a feature, catching a problem while it is still cheap to fix rather than after launch.

An AI user research platform can make these checks faster by simulating an audience around a defined ideal customer profile (ICP) and returning structured findings. Articos, for example, says its research can produce results in under 30 minutes, giving teams a way to gather audience evidence without waiting for a traditional recruitment and interview cycle to run its course.

This does not eliminate the need for larger studies and no credible vendor in this space claims that it does. Instead, it can help teams decide which questions deserve deeper investigation and which decisions already carry enough directional evidence to move forward without further delay.

Why speed changes what teams choose to research

Speed changes what teams are willing to research in the first place, which is a different effect than simply getting an existing answer faster. If every question requires several weeks and a substantial budget, research naturally becomes concentrated around major launches and strategic decisions, leaving everything smaller to instinct.

Faster methods make smaller research questions practical in a way they simply were not before. A team can test an early positioning hypothesis before finalizing a campaign, investigate an unexpected customer objection as it surfaces, or compare multiple concepts before choosing one to bring into human testing. Consumer adoption of generative AI tools has moved quickly enough that this shift is no longer confined to research specialists. Forrester's research on consumers and generative AI found that 38 percent of United States (U.S.) consumers report having used the technology within three years of its mainstream introduction, with weekly use now common among adopters. 3

The value of faster research is therefore not simply getting an answer sooner. It is creating more opportunities to use evidence before a decision becomes expensive to change, which compounds across a year of smaller product and messaging calls in a way a single fast study cannot on its own.

The limits of simulated audience behavior

Synthetic research carries an important limitation that no benchmarking exercise fully resolves. Simulated behavior is not the same as observing real people and a synthetic audience can reproduce patterns identified by its underlying methodology without guaranteeing how an actual customer will behave in a new market, under unusual circumstances, or once real money and real world constraints are involved.

Nielsen Norman Group has been direct about where it stands on this question.

Synthetic users are fake users generated by AI. While there may be a few use cases for them, user research needs real users

That position is worth taking seriously rather than dismissing as institutional caution. It makes human validation particularly important for high risk decisions, novel products, sensitive topics and situations where subtle behavioral context matters more than the pattern a model has learned to reproduce. Synthetic findings are better understood as evidence that can inform a decision, not as an unquestionable prediction of how customers will actually behave once a product or message reaches them.

Combining synthetic and human research effectively

A sensible approach is to use each method for the questions it handles best rather than treating the choice as a single decision that applies to every project a team runs. Synthetic audiences can support rapid exploration, early hypothesis testing, message comparisons and smaller decisions that might otherwise receive no research at all given normal timeline and budget constraints.

Human research remains valuable when teams need direct observation, deeper qualitative context or validation of important findings with actual customers. A synthetic result can even help improve that process by identifying which hypotheses or questions deserve attention in a larger study, effectively acting as a filter before the more expensive method gets deployed.

Industry standards are catching up to this layered approach. The revised ICC/ESOMAR Code, the market research industry's governing code of conduct, now requires researchers to disclose when synthetic data or AI contributed to a study, along with the extent of human oversight applied. 4 That kind of disclosure requirement gives clients and stakeholders a way to weigh synthetic findings appropriately rather than treating every result as equally certain regardless of its source.

For insights teams, the emerging model is therefore less about choosing between AI and human research and more about building a layered research process. Major studies establish strong evidence on strategic questions, while synthetic audiences provide faster feedback on the decisions that surround them. The result is a more continuous approach to market research:

fewer decisions made entirely on instinct, more opportunities to test assumptions before they become expensive and a clearer basis for deciding when a question still needs a human in the room

• 1Using gen AI for early-stage market research
• 2Baymard's UI research methodology and data
• 3The state of genAI and consumers for 2026
• 4Why the ICC/ESOMAR code will matter more than ever in 2026