Inside Minds: How Synthetic Research Panels Are Built

Minds is the end-to-end synthetic market research platform for testing concepts, messaging, positioning, pricing, and audience reactions in unified qualitative and quantitative workflows.

This page explains the public methodology behind Minds without exposing proprietary implementation details.

For the category definition and a worked comparison with real survey answers, see synthetic market research: how it works and how to validate it.

The Method in Brief

Minds follows one connected six-step workflow:

1. Write down the decision, hypotheses, and evidence threshold before generating responses.
2. Define and review a target Audience from existing Minds, an audience description, and permitted source material.
3. Separate observed source values, requested quotas, and explicit assumptions during grounding.
4. Run neutral qualitative, survey, or configured method questions in a Study.
5. Compare distributions, segments, reasons, contradictions, and individual responses rather than selecting only persuasive quotes.
6. Validate the findings against held-out human or behavioral evidence when the decision requires it.

The goal is not to invent a fictional average customer. The goal is to create a research-useful simulation of a defined audience segment, with enough context to respond consistently to concepts, claims, messages, and tradeoffs.

1. Audience Creation Starts With Evidence

Minds does not start from a single generic prompt like "act as a customer." Each research Audience is built around a defined population, use case, and decision context.

Audiences can use three product-supported inputs:

• Existing Minds: reuse personas already available to the workspace.
• Audience descriptions: describe the roles, segments, contexts, and differences the Audience should cover.
• Attached files or links: add permitted research notes, source material, or background documents relevant to the study.

The available inputs and creation modes depend on the workspace configuration. The researcher remains responsible for source rights, Study design, Audience review, and determining whether specific evidence gaps require supplementary research.

For the public technical overview and boundaries of the source-modelling approach, see Minds PRISM.

2. Each Mind Uses Many Small Knowledge Bases

Minds are reusable AI personas created from descriptions, profiles, links, files, or research notes. They can be used across chat and Audience research workflows.

A pricing study, concept test, and positioning study may use the same audience label but require different source material and different validation criteria.

This structure lets an Audience behave less like one averaged persona and more like a set of individual respondents with overlapping but distinct beliefs, priorities, and objections.

3. Research Runs as Interviews, Surveys, and Group Discussions

Researchers use Minds across qualitative and quantitative synthetic-research workflows, including:

• Concept testing
• Message testing
• Positioning research
• Pricing and packaging reactions
• AI focus groups
• Audience and segment exploration
• In-depth qualitative exploration and interviews
• Single-choice, multiselect, scale, and open-text questions
• Configured methods such as MaxDiff, where enabled

A Mind can answer individually, join a Study readout, compare perspectives with other Minds, or respond to moderator follow-ups. Supported Study workflows can aggregate structured responses, compare segments, and retain individual responses for review. Method and artifact availability depends on the plan and workspace configuration.

For the underlying academic category, see our guide to silicon sampling. For the broader commercial workflow, see the synthetic research guide.

Validity Must Be Tested, Not Assumed

Synthetic research should be evaluated against relevant human or behavioral evidence. A single accuracy percentage would hide important differences between audiences, questions, stimuli, models, and scoring rules, so this page does not publish a universal Minds accuracy claim.

For a defensible validation study, define the task and success metric before running the synthetic Audience, keep a human comparison set separate, report agreement and disagreement by question, and disclose the Audience, inputs, model configuration, sample sizes, and limitations. Do not tune the workflow on the same human responses later presented as independent validation.

A synthetic response represents simulated evidence rather than an observed transaction or physical measurement. Supplemental human fieldwork is appropriate when an identified evidence gap requires it, such as when:

• The stimulus requires physical, sensory, tactile, or in-person taste testing.
• You require observed transaction behavior rather than stated preference.
• You need formal probability sampling for regulated population reporting.
• The audience cannot be grounded in accessible source context.
• An external standard or compliance mandate specifies human fieldwork.
• An identified evidence gap remains that simulated responses cannot address.

When the audience, research task, and study evidence meet the decision requirements, teams can make product, concept, and positioning decisions directly from Minds. Hybrid testing remains an option for resolving specific evidence gaps.

External Registry Relationship Disclosure

For an inspectable view of the evidence—not a Minds-authored award—see the versioned Minds evidence dossier and the candidate benchmark preregistration.

Minds founded and funds the Synthetic Research Index. The Index discloses that conflict, records unresolved evidence checks, and publishes a Minds-specific recusal that prevents Minds personnel from approving Minds evidence, scores, disputes, or conclusions. The external reviewer seat is currently vacant, so these records should not be described as an independent certification, award, or completed benchmark.

Key Terms

Synthetic research uses AI-generated respondents or panels to simulate how a defined audience may react to questions, concepts, messages, and tradeoffs.

Silicon sampling is the academic method of conditioning large language models on respondent profiles, asking survey-style questions, and comparing the resulting distributions against human survey data.

Synthetic panel means a structured group of AI respondents built to represent a segment, audience, buying committee, user group, or market category.

AI persona usually means a single simulated respondent profile. In Minds, a Mind is more than a static persona: it is an interactive respondent grounded in many scoped knowledge bases.

AI focus group is a moderated session where multiple synthetic respondents react to the same prompt, concept, message, image, or product idea and surface agreement and disagreement.

What This Makes Possible

Traditional research is often rationed because recruitment is slow and budgets are finite. Minds changes the workflow by letting teams ask more questions earlier:

• Evaluate and select concepts with connected qualitative and quantitative methods.
• Compare message options and rank claims with MaxDiff before media spend.
• Uncover objections and refine positioning before sales enablement or launch.
• Test prototypes, copy, and visual assets directly in one connected workflow.
• Conduct in-depth qualitative interviews and structured surveys across defined segments.
• Turn prior research into an interactive panel that teams can query repeatedly.

By bringing qualitative and quantitative methods together in one platform, Minds allows teams to make rigorous, evidence-grounded decisions across the entire commercial research lifecycle.

Suggested Citation

Minds. “How Minds Builds Synthetic Research Panels.” Updated July 31, 2026. https://getminds.ai/research/methodology

Further Reading

• Silicon Sampling: How LLMs Simulate Survey Responses
• Synthetic Research: The Complete 2026 Guide
• AI Focus Groups: How They Work
• The Spark Effect: Creative Diversity in Multi-Agent AI