Synthetic market research is an emerging industry with a range of approaches suited to different decision contexts. Not all synthetic research is the same — the methodology, depth, and reliability vary significantly depending on the approach. Understanding the landscape helps organisations choose the right tool for the decision at hand. This guide explains the two primary approaches within synthetic market research: biography-prompted LLMs for quick directional feedback, and networks of enriched personas — the approach behind Artificial Societies — for high-stakes strategic decisions.
The simplest form of synthetic audience research prompts a large language model with a short invented biography — for example, "you are a 34-year-old teacher from Ohio" — and reads off its answer. The output is usually aggregate-level: overall preference scores, sentiment distributions, or directional signals. This approach is useful for execution-level decisions where speed matters more than depth: screening a large number of options, getting a quick sentiment check, or validating minor creative variations. Multiple providers offer biography-prompted tools, and costs are typically low. The trade-off is significant: biography-prompted personas lack internal coherence across questions, social dynamics are not modelled, and individual-level qualitative reasoning is shallow or absent. This makes them unsuitable for decisions where understanding why audiences react differently is critical.
The deeper approach, used by Artificial Societies, constructs purpose-built networks of 200 to 3,500 interconnected AI personas, each with an internally coherent belief system and personality grounded in real human behavior data. These personas are connected within a social graph that models influence dynamics, conformity pressures, and opinion formation — the forces that shape real human communities. The output is fundamentally different: thousands of individual-level qualitative explanations, segment-level analysis across multiple stakeholder groups, realistic opinion distributions that capture minority viewpoints, and pre-exposure/post-exposure measurement of opinion shift. This approach is designed for high-stakes decisions: strategic narrative testing, stakeholder reaction prediction, confidential pre-launch research, crisis communications preparedness, and any scenario where the consequences of misjudging audience response are significant.
The two approaches differ across several dimensions. Biography-prompted LLMs rely on invented biographies with limited coherence, do not model social dynamics, and produce aggregate preference scores — best suited for quick directional feedback and low-stakes screening. Networks of enriched personas use internally coherent belief systems built from real behavioral data, model a full social graph with influence dynamics and conformity forces, and produce individual-level qualitative reasoning across thousands of personas — best suited for high-stakes strategic decisions and stakeholder reaction analysis. The two also validate very differently: independent evaluation shows Artificial Societies reaching far closer to the human self-replication ceiling than biography-prompted LLMs across distribution accuracy, coherence, and open-response quality. See the Method & Evaluation page for the full benchmarks.
The decision depends on what is at stake and what depth of insight is required. If the decision is low-stakes and a directional signal is sufficient — choosing between minor copy variations, screening a large set of creative options, or getting a quick sentiment read — a biography-prompted LLM tool is likely enough. If the decision is high-stakes and requires understanding why different stakeholder groups react differently, with individual-level qualitative reasoning and comprehensive evidence — narrative positioning, crisis response strategy, confidential campaign testing, stakeholder reaction analysis — networks of enriched personas are the appropriate methodology. A useful rule: if you need to present the findings as evidence to support a consequential strategic recommendation, you need the depth that networks of enriched personas provide.
Yes. Organisations sometimes use biography-prompted LLMs for initial screening — narrowing a large set of options to a shortlist — and then apply networks of enriched personas to evaluate the shortlisted options with the depth and rigour required for a final strategic decision. This tiered approach combines the speed of the simpler tools with the evidence base of a full Artificial Societies study, matching the depth of analysis to the stakes at each stage.
Synthetic market research spans two primary approaches. Biography-prompted LLMs use short invented biographies for fast directional feedback on low-stakes decisions. Networks of enriched personas — the approach behind Artificial Societies — construct interconnected networks of 200 to 3,500 personas with coherent belief systems and social dynamics, for high-stakes strategic decisions requiring deep, individual-level insight. The right approach depends on what is at stake.
Biography-prompted LLMs produce aggregate preference scores from short invented biographies. Networks of enriched personas construct interconnected Artificial Societies with individually coherent personas grounded in real behavioral data, modelling social influence dynamics and providing individual-level qualitative reasoning at scale. The difference is the depth and reliability of insight — which matters most for consequential decisions. Both fall under the broader category of synthetic market research.
Biography-prompted LLMs are sufficient for execution-level decisions where a quick directional signal is enough: screening creative options, checking basic sentiment, or validating minor variations. They are not suitable for high-stakes decisions where understanding stakeholder reasoning, managing confidential materials, or providing comprehensive evidence for strategic recommendations is required.
Use networks of enriched personas when the decision is consequential and requires understanding why audiences react differently — not just which option they prefer. If you need individual-level qualitative reasoning across thousands of stakeholders, social influence dynamics, and an evidence base for a strategic recommendation, networks of enriched personas are the appropriate methodology.