Audiences — reusable synthetic research samples

An Audience is a reusable group of Minds representing the people you want to understand. Researchers may also call this a synthetic sample or panel, but Minds uses Audience consistently in the product.

An Audience is not the research activity itself. The work happens in a Study, where you can use one Mind or one or more Audiences with the method that fits your decision:

• an in-depth interview with one Mind
• qualitative exploration at scale across an Audience
• a questionnaire with directional quantitative readouts
• a concept or message test
• a segment comparison
• a mixed-method Study that combines these approaches

Build an Audience in two ways

Select existing Minds

1. Turn on multi-select in the Minds list and choose the Minds you need.
2. Select Create Audience in the action bar.
3. Review the Audience, give it a clear name, and save it for reuse.

You can also open New Audience and add existing Minds there.

Describe the people you need

When the right Minds do not exist yet, describe the segment and let Minds draft it:

1. Select New Audience in the sidebar.
2. Describe the people you want to understand—for example, “working parents in Germany aged 30–45 with children in school” or “B2B CMOs at Series B companies.”
3. Minds allocates the provisional profiles before writing their biographies. Review the applied composition in the distribution cards and inspect their supporting sources.
4. Request changes through Rework, then select Create to use the accepted review.
5. Select Create. The Minds continue building in the background if you leave the screen.

An uploaded completed study provides a reference sample. Respondent characteristics define its composition; supplied satisfaction, purchase-intent, and other outcome answers stay separate as historical findings. Counts, percentages, missingness, and valid survey-weighted findings are calculated from the rows. Employment, occupation, and income retain observed relationships where feasible. Names, emails, and respondent IDs are excluded from synthetic persona content.

Review the applied composition in the distribution cards. Uncollected profile fields remain modeled or unknown; missing and not-asked states stay distinct. Composition uses unweighted sample counts. Valid weighted findings and original sample shares remain in the source evidence. Requested quotas take precedence. A customer sample does not establish national population proportions.

For new research with a similar audience, historical outcome answers are withheld from persona generation and retrieval. Explicit reconstruction may use historical aggregates, labeled as informed by existing answers. Change the research purpose through Rework. Synthetic Minds are not the original respondents, and reproducing supplied answers does not validate a prediction. Existing saved Audiences retain their composition.

Choose the Audience design

Under Additional settings, choose how Minds should shape the draft:

• Balanced — creates a compact Audience while representing the grounded distributions supported by the source data.
• Segment coverage — covers more supported segmentation criteria and relevant subgroups.
• Benchmark depth — uses deeper benchmark-style coverage when the decision requires more granular cells.

These settings affect the Audience design; they do not create a separate research method.

Start a Study

Open an Audience and select Start Study, or start a new Study from the sidebar. Add the question, brief, website, image, ad, video, or document you want to explore.

Choose how much setup you need:

• Quick — analyzes the request and sources, recommends relevant saved or new Audiences, and asks for confirmation before starting.
• Custom — opens an AI-guided setup where you define the research goal, sources, Audiences, method, questions, and response formats before anything runs. Existing Audiences and NEW drafts are selected in the Audience step; NEW drafts remain uncreated through review and use the grounded Quick Audience-creation pipeline only after final confirmation. The Study opens when those Audiences are ready.
• Blank — opens an empty Study so you can assemble it yourself.

“Custom” describes the setup experience. It is not a research method. The Study itself can use an interview, questionnaire, qualitative exploration, concept test, segment comparison, or a combination of methods.

Review questions and response formats

Before a planned Study runs, review the proposed questions and response format. A question can use open text, choices, a standard 1–5, 1–7, 1–10, or 0–10 scale, or a custom integer scale. The confirmed format stays attached to the Study through processing and aggregation.

API and MCP clients receive the same per-question response contract, so the confirmed format remains consistent across interfaces.

Quick planning, generated Custom questions, plan revisions, v1, and MCP use the same research-planning policy and fast planner model. When the confirmation selects an available method—Custom research, Focused question, Questionnaire, Qualitative exploration, or a calculator-backed method such as MaxDiff, NPS, top/bottom box scoring, key driver analysis, TURF, Gabor-Granger, Van Westendorp, Kano, ranked preferences, or segment comparison—the exact method contract stays attached through the normal stream or durable questionnaire processor. Experimental and planned methods remain visible in the plan, but the Run action stays blocked; they are never silently converted to Custom research.

Minds can detect multiple questions in a prompt or uploaded questionnaire. You can distinguish your original questions from suggestions, edit the plan, and confirm the final sequence before it runs.

Add methodological complexity only when it helps

Most requests do not need a named method. Start with the decision, main source or asset, and the questions the user wants answered. Add methodological complexity only when the user requests it or when it materially changes the evidence.

• Available methods run through the current Study runner. For example, MaxDiff builds balanced forced-choice tasks, and NPS, top/bottom box scoring, key driver analysis, TURF, Gabor-Granger, Van Westendorp, Kano, ranked preferences, and segment comparison each design their questions server-side, then calculate deterministic overall and per-Audience results after collection. Segment comparison includes pairwise significance tests across the answering Audiences.
• Experimental methods can be represented and reviewed but are rejected at execution.
• Planned methods are represented for forward compatibility but are not executable.

The question processor remains method-agnostic: it collects exact task answers and owns retries, quota, ordering, and persistence. Versioned server adapters design method-specific tasks and calculate artifacts, so future built-in or client-specific methodologies can be added without rewriting the processor. The result summary receives those deterministic artifacts as authoritative evidence and explains them without recalculating the scores. Conjoint is also available: its adapter prepares choice tasks, estimates part-worths, returns validation diagnostics, and simulates preference shares. It requires explicit advanced-method opt-in and a valid configuration.

The Study loader uses persisted server acceptance and processor-start timestamps. Reopening or reloading the Study therefore continues the same elapsed counter instead of starting again at zero.

Compare Audiences in one Study

Add two or more Audiences to the same Study to compare segments side by side. Ask the same question and examine where their individual responses, distributions, themes, and synthesized findings converge or differ.

This is a segment comparison within a Study, not a separate “panel” object.

Go from breadth to depth

If one response stands out, open that Mind for an in-depth interview. Probe the reasoning, show another stimulus, or ask what would change the response. The original Audience Study remains saved, so you can return to the wider evidence at any time.

Understand the evidence

Depending on the question and configured response format, a Study can show:

• individual Mind-level responses
• distributions and comparisons
• clustered qualitative themes
• synthesized findings grounded in the collected responses
• available alignment indicators for the Audience and its grounding

Treat quantitative outputs from synthetic respondents as directional evidence, not automatically as population estimates. The strength of the result depends on the Audience definition, grounding, question design, and available validation evidence.

Validate an Audience against real surveys

Audience Validation checks an Audience against real, published surveys. Minds finds surveys whose respondents best match the Audience, asks the Audience's Minds the same questions, and scores how close their answers are to the published answers. Open an Audience and select the Validation tab to see its results or run a new validation.

Automatic validation of new Audiences

A new private Audience is validated automatically, once, as soon as all of its Minds have finished training, if it has at least 10 Minds and you have an included validation left that month. Automatic validations never spend synthetic responses and send no email. Audiences that existed before automatic validation was introduced are validated only when someone asks.

Run a validation yourself

You can validate any Audience you can edit, at any time and as often as you like. The Audience needs at least 10 ready Minds. In the Validation tab, choose what to check it against:

• the published surveys that fit the Audience best, found by Minds
• up to five listed surveys that you pick
• your own survey: the questionnaire and its results, as PDF, Excel (XLS, XLSX), CSV, Word (DOCX), or TXT
• a respondent-level dataset, on the Team plan
• a Study run you already did, which is scored for free and asks nothing new

All chosen surveys are asked in one run, in a Study named "Validation run: ". That Study is the platform's working copy: it does not appear in your Study list and does not count toward your Study limits. Validation runs queue behind the Studies people start, so they never slow your own research. A run usually takes 10 to 60 minutes, and you can stop it before any survey is asked.

When a validation you started runs longer than a few minutes, you receive an email when it finishes. Its View the validation button opens the Audience on its Validation tab.

Which questions are asked

Surveys are ranked by who answered them compared with the people who define the Audience: the Audience's own people, a part of them, a neighbouring population (including the same kind of people surveyed in another country), or the wider population they belong to. Surveys of the Audience's own people come first; evidence about the wider population supports a result but never outweighs more specific surveys.

Before any Mind answers, every question is checked for this Audience, one by one. A question is left out when:

• the Audience's people cannot answer it, for example because it presupposes something they don't have
• the published answers are not a fair target for this Audience, because the respondents differ in a way that would change the answer; the general public's eating-out habits, for example, are no target for site workers
• the same question is already asked from a better-fitting survey

Where a survey publishes a question's answers by age or region, the target is narrowed to the Audience's own mix. Every left-out question is listed with its reason. One validation asks at most 60 questions from at most 12 surveys, and at most 30,000 answers in total. It needs at least 8 fitting questions; with fewer, nothing is asked and nothing is spent.

Read the score and its range

For each survey, the Validation tab shows a score and its 95% range, who the survey asked compared with the Audience, the number of questions, and a link to the source. The score is 100 minus the average absolute difference, in percentage points, between the Minds' answers and the published answer shares: 100 means identical answers. The 95% range reflects sampling of both the Minds and the survey's respondents, and any adjustment to the Audience's composition. A run with several surveys also shows its combined score.

The top of the tab shows the Audience's overall validity across every survey it was checked against. Each survey counts once, by its latest check, weighted by the number of questions scored.

What it costs

Plans include validations that cost no synthetic responses: 3 runs a month on the Individual plan, 3 per seat on Team (pooled across the team), a number set in the contract on Enterprise, and none on Free. One run counts once, however many surveys it asks. A run that ends without a score, and without spending any included answers, gives its slot back.

Beyond the included runs, a validation is billed like any Study: one synthetic response per Mind per question. When Minds chooses the surveys, it fits them to the responses you have left.

Validation is also available through the Audiences API and the MCP tools validate_audience and get_audience_validation.

Tips

• State the decision you need to make, not only the topic.
• Use a single Mind for depth and an Audience for breadth or comparison.
• Attach the actual stimulus when testing a claim, concept, page, image, or video.
• Use explicit choices or scales when you need comparable directional readouts.
• Follow up on surprising responses before relying on a synthesized pattern.
• Compare different Audiences only when the difference is relevant to the decision.

Audience is who you study. Study is the saved research workspace. Method is how you learn.