Practical support for better synthetic-persona research.

The promise, and the catch

Synthetic research platforms promise something brand and product teams have wanted for a long time: consumer feedback in hours instead of weeks. Instead of recruiting a panel, you describe a target group, the platform generates AI-based personas, and you run interviews, surveys or concept tests against them. The first results arrive before a traditional study would have finished its screener.

The market has responded quickly. There are now dozens of platforms, each with its own approach, pricing logic and claims about accuracy. For a brand team, that creates a new problem. The hard part is no longer getting synthetic answers. It is knowing which platform fits the question, and knowing whether the answers deserve to be trusted.

Most organisations don’t have the in-house capability to judge that. Large insight functions sometimes employ people who can examine model output critically, test it for bias and put it next to real-world data. Most brand teams, marketing departments and mid-sized manufacturers don’t. That is the gap PolySight was built to close.

What these platforms do and where they differ

Synthetic research platforms broadly fall into four groups:

• Persona study platforms generate synthetic respondents and let you run structured interviews, surveys and concept tests. They suit early discovery, message testing and concept validation.
• Audience and market simulators model larger populations or stakeholder systems. They are built for scenario questions: how might different segments react to a price change, a campaign or a new category?
• Synthetic UX testers simulate users working through a website or prototype to surface usability and comprehension issues.
• Synthetic data infrastructure produces artificial datasets rather than interactive respondents. It sits underneath research workflows rather than replacing them.

Within each group, platforms differ on the points that matter most. What are the personas grounded in: population statistics, survey data, the client’s own CRM data, or mainly the language model’s general knowledge? What evidence does the vendor offer that its output matches real human responses? Which model, or models, generate the answers? And how easily can a study be repeated to see whether the result holds?

Five questions to ask before choosing

1. 1. What decision will this research inform? A copy test, a pricing hypothesis and a market-entry scenario need different tools. The research question should choose the platform, not the other way round.
2. 2. What grounds the personas? A persona grounded in census distributions behaves differently from one built on your own customer data. Both behave differently from a persona that is essentially a well-written prompt. Ask the vendor to explain the grounding for your target group specifically.
3. 3. What validation evidence exists, and for which context? Accuracy claims are often measured on one type of study in one market. A benchmark from US software users says little about German premium kitchen buyers.
4. 4. Can you review the research setup? Ask how the platform records the brief, audience context, assumptions and limitations. Consistent-looking answers are not evidence of accuracy. You need enough context to judge whether the output is relevant to your decision.
5. 5. What happens to your data? Grounding personas in your own customer data means sharing personal data with the party that runs the platform. Before anything is uploaded, it should be clear who processes the data, where, and under which contractual terms. PolySight clarifies this with you during study design and, wherever possible, works with anonymised or aggregated inputs so that no personal data has to leave your organisation. The assessment of your data protection obligations remains with your data protection officer or legal counsel.

The missing capability: reading synthetic output forensically

Synthetic answers can sound plausible without being well supported. Critical examination means checking the reasoning against the brief, the audience context and available evidence:

• Assumptions: what has the response assumed about the audience, product or purchase situation?
• Evidence: which statements are supported, and which are plausible but unverified?
• Missing perspectives: what objections, constraints or alternative explanations have been left out?

Use this review to identify questions for your team and priorities for real fieldwork. A fluent response or a repeated theme does not establish consumer preference, market demand or factual accuracy.

What PolySight does

PolySight is a client-facing service that combines proprietary software with consulting. Its purpose is to help organisations get the most out of the synthetic research platforms available, without having to build a data forensics team first.

The software. PolySight’s own platform sits alongside the research tools you use and works independently of any single vendor. It:

• uses a focused research brief and defined audience context to explore synthetic persona perspectives
• draws on a curated persona library organised around target groups, so studies don’t start from scratch
• helps teams review themes, objections and assumptions in the responses
• keeps the brief, audience context and responses available for review and export
• keeps studies re-runnable, so a revised concept can be tested against the same panel and compared directly

The consulting. Software alone doesn’t make a study useful. PolySight’s consulting covers the full path from question to decision:

• Platform selection: which type of synthetic research tool fits your question, data situation and budget, and which one doesn’t
• Study design: framing the research question, selecting and grounding personas, and documenting the audience context and research setup
• Forensic review: checking output for bias, implausibility and overconfident claims before anyone acts on it
• Interpretation and recommendations: turning findings into concrete implications for product, positioning, pricing and communication
• Fieldwork triggers: identifying the points where synthetic research has done its job and human research should take over

PolySight’s consulting covers research, product, brand and communication questions. It does not include legal or financial advice. Where a finding touches regulated claims, data protection obligations or investment decisions, we flag it for review by the appropriate qualified adviser.

How an engagement works

A typical engagement starts with a fit check: what decision you’re facing, what data you have and whether synthetic research is the right tool at all. It moves on to study design and a first study run, followed by forensic review and a findings session with your team. Many clients then continue with recurring studies, such as concept tests across a product pipeline, message testing before campaigns or competitive perception tracking. The same panels are re-used, so results stay comparable over time.

One proposal, one invoice. Every engagement is priced on a proposal tailored to your requirements. Platform usage and PolySight’s consulting are combined in a single invoice, in line with the engagement plan agreed with you.

What synthetic research can’t do, and why we say so

Synthetic personas aren’t a representative sample of a real population. Their output is directional. It is good for generating hypotheses, narrowing options and catching problems early, such as copy that reads as cold or technical to the audience it is meant to win. It doesn’t forecast demand, prove causality or replace human research for high-stakes decisions on capital allocation, safety or regulated claims.

PolySight treats that boundary as part of the method, not as fine print. Every PolySight report states it, because a synthetic finding is only useful when the people acting on it know what it can and can’t carry.