Merciv

Merciv vs synthetic audiences: test concepts against personas built from your own research

Merciv builds personas from the respondent-level research your team already owns and shows the sources behind each one, so you can screen concepts, ads, and claims early and send the strongest to real respondents.

Updated

The short answer

General synthetic audiences are the simulated respondents that research vendors build from their own panels and data. They are a strong choice for teams with little research of their own that want early reads at vendor scale. Merciv Personas are built for insights teams at consumer brands and retailers that own respondent-level research and want early reads in their own segments, because Merciv grounds every persona in that research and lists the studies behind it. Screen with either, then take go/no-go launches, major pricing commitments, and claims you will publish to real respondents, as vendors advise for high-stakes decisions .

Written by the Merciv team. We make Merciv. This page compares Merciv with general synthetic audiences and fieldwork using vendors’ public documentation, independent research, and industry guidance as of November 2026; see the method note at the end.

On this page

01Where each fits

At a glance

Choose a general synthetic audience if…

  • You have little respondent-level research and want an audience built from a vendor’s data.
  • You want to screen up to 100 early concepts at once, as Kantar ConceptEvaluate AI does⁠[3].
  • You want a vendor-published validation program behind its audiences.

Choose Merciv Personas if…

  • Your team owns segmentation studies, concept tests, or other respondent-level research.
  • You need early reads in your own segments, with each persona’s sources in view.
  • You want persona work beside your studies, syndicated data, and live consumer conversation.

Choose fieldwork with real respondents if…

  • The decision is a go/no-go launch, a major pricing commitment, or a regulatory submission.
  • You plan to publish a claim about what consumers prefer.
  • You need a dependable read on a small subgroup or on less common views.

02How they differ

Where do Merciv Personas and general synthetic audiences get their answers?

General synthetic audiences answer from the data each vendor holds. Qualtrics fine-tunes an open-source language model on “millions of rows of anonymized survey responses, additional licensed research training data, and publicly available data”⁠[1]. NIQ builds BASES AI Screener on “transaction data and language models trained on real consumer behavior”⁠[4]. That scale is the approach’s strength. Ipsos PersonaBot and Electric Twin also work from a client’s own research⁠[5, 6].

Merciv Personas answer from the studies your team has already run. Merciv builds each persona from that respondent-level research, works inside your taxonomy, segmentation, and definitions, and shows where each persona comes from. Every claim cites its source at page level, and every insight carries a confidence score. When supporting evidence is missing, Merciv says so instead of producing a plausible answer.

Picture a brand team weighing three pack claims. It puts each one to Merciv Personas built from last year’s segmentation study and reads each reaction beside the studies behind that persona. Then it takes the two strongest into fieldwork. Merciv cleans, structures, and connects every study a team has produced, so last year’s study can answer this year’s question. Welch’s built one library across its internal studies and syndicated data in Merciv, and grew its use of Merciv from Insights into Brand, R&D, and Category Management.

03How they differ

When do Merciv Personas and synthetic audiences need real respondents behind them?

Qualtrics says synthetic panels “perform best with perceptions, preferences, and intent-based questions”⁠[7]. NIQ positions BASES AI to help teams “iterate on ideas ahead of final qualification with human consumers”⁠[2]. Merciv builds Personas to test concepts, ads, and claims before you commit, and high-stakes decisions still need real respondents.

Independent research supports that line. Pew Research Center built AI respondents for its own panelists. Their answers differed from the human results by an average of 12 percentage points, with “clear errors for all major subgroups”⁠[8]. A 2026 Science Advances study found that digital twins gave less varied answers than the people they modeled on 154 of 164 outcomes⁠[9]. Pew tested public-opinion questions, so consumer concept tests sit outside its scope.

A beverage team with twelve flavor concepts uses Merciv Personas to narrow the list to four, and a concept test with real respondents picks the launch. Merciv adds that study to the team’s research, so the next screen starts from what real consumers said. Merciv also reads real consumer voice directly. Athleta used it to compare 10,000+ reviews attribute by attribute and found the product gap competitors had filled and Athleta hadn’t: secure zippered pockets.

04Side by side

Merciv Personas vs synthetic audiences vs fieldwork, side by side

  • Available
  • Partial support
  • Not available
  • Not publicly documented
Merciv Personas vs synthetic audiences vs fieldwork, side by side
What you needMerciv PersonasGeneral synthetic audiencesReal respondents (fieldwork)
What is each read built from?
Context and scope.
Your research. Each persona comes from your respondent-level studies.
Context and scope.
Vendor data: survey archives, panels, transaction data, or past concept tests⁠[1, 10, 4, 3].
Context and scope.
Real people. Each answer comes from your sample.
Does it work in my segments and definitions?
Yes.
Merciv runs inside your taxonomy, segmentation, and definitions.
Context and scope.
Varies by vendor. Ipsos PersonaBot and Electric Twin work from a client’s research⁠[5, 6].
Yes.
You set the sample and questions.
Can I see what a read is based on?
Yes.
Per persona. Merciv shows each persona’s sources and cites each claim’s page.
Context and scope.
At model level. Qualtrics and Kantar document their training data⁠[1, 3].
Yes.
Every answer traces to a respondent record.
How do I know how far to trust one read?
Context and scope.
Scored. Every insight carries a confidence score, and Merciv flags missing evidence.
Context and scope.
Varies by vendor. Simile tags every result with a predicted accuracy level⁠[11].
Context and scope.
Sample-based. Sample size and design set confidence.
Can I test with no research of my own, or in a new market?
No.
Merciv Personas need your respondent-level research.
Yes.
Toluna’s personas draw on 79 million+ consumers across 17 markets⁠[10].
Yes.
Fieldwork creates the research.
Is there published validation against real people?
Not publicly documented.
Yes.
Qualtrics, Toluna, and Electric Twin publish results against human responses⁠[12, 10, 6].
Context and scope.
Not applicable. Real people are the benchmark.
Can it carry a go/no-go launch or a major pricing commitment?
No.
High-stakes decisions still need real respondents.
Context and scope.
Less suited, per Qualtrics, which lists both among high-stakes final decisions⁠[1].
Yes.
Real respondents supply the evidence.
Can it substantiate a claim we will publish?
Context and scope.
Screening. Use Merciv Personas to pick which claims to field.
Context and scope.
Screening. Toluna positions Claims Test Instant for early-stage work⁠[10]; the National Advertising Division requires AI testing validated for the specific claim⁠[13].
Yes.
With a representative sample that can support the claim⁠[14].

Comparison based on publicly available information as of November 2026.

05Sources

Sources and method for synthetic audiences and Merciv

We make Merciv, and we wrote this comparison of general synthetic audiences, Merciv Personas, and fieldwork. We used: - vendors’ public documentation, validation reports, and pricing descriptions; - research from Pew Research Center and Science Advances; - guidance from MRS, ESOMAR, the FTC, the National Advertising Division, and CAP; - Merciv’s own product. Vendor figures appear as each vendor reports them. We checked every source on November 4, 2026, and we will update this page when Merciv or a named vendor changes. To request a correction, contact contact Merciv.

Found a factual error? Send Merciv a correction with the claim and a current source.

  1. Qualtrics, “Synthetic data FAQ,”
  2. NIQ, BASES AI announcement,
  3. Kantar, “ConceptEvaluate AI,”
  4. NIQ, “BASES AI Screener,”
  5. Ipsos, PersonaBot launch,
  6. Electric Twin, “FAQs,”
  7. Qualtrics, “Creating a Synthetic Panel,”
  8. Pew Research Center, “How well synthetic samples replicate public opinion,”
  9. Science Advances, “Digital twins are funhouse mirrors,”
  10. Toluna, Claims Test Instant,
  11. Simile, homepage,
  12. Qualtrics, “Fine-tuned AI outperforms general-use LLMs,”
  13. National Advertising Division, Starkey decision,
  14. CAP, survey claims advice,
  15. NIQ, “The rise of synthetic respondents,”
  16. FTC, “Policy Statement Regarding Advertising Substantiation,”
  17. Qualtrics, X4 2026 announcement,
  18. ESOMAR, “20 Questions to Help Buyers of AI-Based Services,”
  19. MRS, “Guidance on Using AI and Related Technologies,”
  • Use a general synthetic audience when your team has little respondent-level research. It also fits when you need an early read in a market your studies have not covered. The vendor brings the data: NIQ’s BASES AI draws on consumer-permissioned data from hundreds of thousands of households . Merciv Personas reach only the consumers your own studies reach.

See Merciv work on your own data

We’ll securely connect your data, run your hardest question end to end, and show you every source, before you commit to a contract.

Book a demo