Synthetic Research FAQ
What synthetic research is, how AI personas work, what they're accurate for, and where you still need real humans.
Synthetic Research FAQ
The basics of what synthetic research is, how it differs from traditional methods, and where it fits in your stack. For deeper context, see the blog posts on synthetic user research, what is customer simulation, and synthetic vs recruited panels.
What it is
What is synthetic research?
Synthetic research uses AI personas, trained on behavioral patterns and domain knowledge for specific audiences, to answer the same questions a real research panel would. You ask a Mind or a panel of Minds about a campaign, product, message, or concept and get structured responses in minutes instead of weeks.
It does not replace every form of research. It replaces the rapid, exploratory, "we need a directional answer this week" work that used to be skipped entirely because traditional research was too slow.
What is a synthetic audience?
A synthetic audience is a group of AI personas that simulate a target segment. Examples: Gen Z students in Berlin, mid-market SaaS CMOs, suburban parents in the US Midwest, agency creative directors in the UK. You query the audience as a panel and get aggregated answers across the segment.
What is an AI persona?
An AI persona is a simulation of one specific role, archetype, or named individual. A persona has a worldview, communication style, knowledge domain, and decision criteria. You can ask it questions, show it stimuli (PDFs, screenshots, video), and it responds in character.
What is a synthetic user?
A synthetic user is an AI persona built specifically to simulate end-user behavior in a product context. Used heavily in product research, UX testing, onboarding flow validation, and feature pre-testing.
How it works
How are AI personas built?
Minds builds a Mind from public-web research (LinkedIn profiles, websites, PDFs, articles, public statements) and runs the evidence through psychological models for personality, values, motivations, and buying behavior. The result is a profile that matches real human data with 80 to 95 percent accuracy and is reusable for unlimited queries.
You can also build a Mind from a plain text description, raw notes, or your own internal research. Most teams start with public-web research because it is the fastest path to a validated persona.
What kind of data trains a Mind?
Public-web sources by default: LinkedIn, company sites, podcast transcripts, public articles, conference talks, public social posts. You can extend a Mind with your own knowledge base (interview transcripts, customer call recordings, internal research). Private uploads stay in your workspace.
How is a synthetic panel different from a single AI persona?
A panel queries 8 to 100 personas at once, in parallel, and aggregates the responses. You see the distribution (60 percent agreed, 30 percent pushed back, 10 percent had a question), the clustered themes from open-ended responses, and the percentage breakdowns from multiple-choice questions. A single persona only gives you one perspective.
Can I show personas images, videos, and documents?
Yes. Drop in PDFs, landing page screenshots, pitch decks, product images, packshots, mocks, competitor ads, interview notes, raw transcripts, and short videos. Every Mind in the chat sees the attachment and reacts.
Accuracy and validity
How accurate are AI personas?
Minds reports 80 to 95 percent accuracy against historical human data benchmarks. Accuracy is highest for:
- Opinions and preferences
- Language patterns ("how would they describe this?")
- Reactions to creative and messaging
- Objections and pushback
- Category framing and segmentation
Accuracy is lower for:
- Predicting actual purchase behavior in unfamiliar categories
- Long-term retention and churn (real cohorts win here)
- Novel product categories with no public training signal
How is accuracy measured?
By comparing AI panel responses to historical real-human research data on the same questions. The 80 to 95 percent range is the band most use cases fall into. We publish methodology details on request.
Is synthetic research peer-reviewed?
Synthetic personas as a research method are increasingly cited in academic and industry publications. Minds published the Spark Effect paper on creative diversity in multi-agent AI research, available at /research/spark-effect-creative-diversity-multi-agent-ai. The category is moving fast; treat any single accuracy claim with healthy skepticism and validate against your own historical data.
Can synthetic research show me net-new insights?
Yes, with caveats. Synthetic research is excellent at surfacing the language patterns, objections, and framings that exist in public data about your target. It is less good at revealing entirely net-new insights that nobody has ever said publicly. For breakthrough insight, combine synthetic (speed, breadth) with a small targeted set of real interviews (depth).
Limits
When should I NOT use synthetic research?
Three cases:
- Regulatory or legal evidence requires real-human consent and audit trails. Pharma label claims, financial product disclosures, regulated advertising claims.
- Longitudinal tracking of real customer cohorts. Synthetic personas do not actually buy your product, churn, or refer friends. Real customers do.
- Novel categories with no public training signal. If your persona is "the first 100 buyers of a category that does not exist yet," there is nothing to train on.
For exploratory, directional, and pre-test work, synthetic is faster and cheaper than the alternative.
Does synthetic research replace real customer research?
No. It replaces a specific subset: the rapid, exploratory, "we need a directional answer this week" work. Most teams use synthetic to triage which questions deserve a real-human study and which don't, then use real-human studies for the questions where the cost is justified.
Do synthetic responses count as "real" data?
They are real signal, not real behavior. Treat them as a high-fidelity prior on what real humans would say, then validate with real humans for high-stakes decisions. The blog post synthetic vs recruited panels covers when to use each.
Is synthetic research the same as a digital twin?
Related but not identical. A digital twin usually refers to a simulation of a specific real person or system updated continuously from live data. A synthetic persona is typically a representative simulation of a segment built from public data. Minds supports both patterns; see digital twin platform for business.