If Synthetic Research Cannot Be Challenged, Can It Be Trusted?
13 October 2026

Authored by David Brett, Business Development Director, THG Fluently. THG Fluently was recently recognised as Best Research Services Partner at the 2026 MRS Operations Awards.

An AI-generated insight can look entirely plausible and still misrepresent the people, behaviours or market it is supposed to describe.

That is the challenge now facing market research. As synthetic respondents, AI personas and generated datasets find their way into live insight programmes, the industry must look beyond how quickly data can be produced and ask whether it can withstand scrutiny across languages, cultures and geographies.

According to Qualtrics’ 2025 MR Trends Study, 71% of global researchers believe synthetic data will account for more than half of all data collection within the next three years. If that prediction is even close to reality, synthetic research will soon influence important commercial decisions across sectors and markets.

Speed and scale will be central to its appeal. But neither can compensate for insight that is culturally distorted, linguistically inaccurate or poorly representative.

Synthetic research can appear convincing while carrying hidden assumptions. A translated question may be technically correct but fail to reflect how people in a particular market actually speak. An AI persona may look realistic but overlook local behaviours, terminology or social conventions. A synthetic output may perform well in one geography while producing misleading results in another.

These risks are difficult to identify from inside the systems that create the data.

The industry therefore needs to move beyond a simple question of whether synthetic research is accurate. Accuracy is only one part of the picture. We also need to ask whether outputs are culturally authentic, relevant to the market, representative of important subgroups, consistent across countries and reliable enough to support prediction.

There will not be one universal approach to answering those questions. Some organisations may compare synthetic outputs with established benchmark studies or live sample data. Others may use internal review panels, holdout testing, qualitative debriefs or triangulation with traditional methodologies. The right approach will depend on the use case, the market, the level of risk and the decision being supported. What matters is that the rationale is clear and the evidence is proportionate.

Synthetic research should not be viewed as a replacement for established research methods. Its value will depend on how responsibly it is tested alongside them.

AI can produce the output. Can you demonstrate why it deserves to be trusted?

That points to a broader need for independent validation, combining appropriate methods with independent human expertise to assess whether synthetic research is culturally authentic, market-relevant and fit for purpose. This is why THG Fluently have introduced Synthetic Data Validation

It should not be treated as a final compliance exercise or a one-off confidence check. It should be built into the research lifecycle. Synthetic outputs can be reviewed by native experts for cultural plausibility and market-specific realism. AI personas can be tested for behavioural credibility and local terminology. Outputs can be compared with real-world benchmark studies using prediction registration, cross-market comparison and variance analysis.

This creates a more meaningful evidence base than simply asking whether a model’s output appears believable.

The same principle applies to the data used to build and evaluate these systems. Human-authored reference responses, culturally relevant annotations, multilingual testing assets and cross-language comparability checks can help reveal where an AI model is learning the wrong lesson, reproducing bias or failing to reflect how respondents behave in real markets.

Independence matters throughout. Organisations developing synthetic respondents or AI models should not be the only ones responsible for judging whether those systems are working as intended. An external layer of evaluation can challenge assumptions, identify blind spots and strengthen governance for AI-enabled research.

Market research has spent years developing standards for sample quality, questionnaire design, translation and data integrity. Synthetic research deserves the same level of scrutiny. Trust cannot be created by the sophistication of a model alone. It must be earned through transparency, challenge and evidence.

The organisations that lead in synthetic research will not necessarily be those that generate the most data or do so the fastest. They will be those able to demonstrate why their insight deserves to be trusted.

That is the standard the industry should now be working towards.

Find out more about THG Fluently’s approach to Synthetic Data Validation Services →