Set up a study
- 1Build the questions in display order
Put questions everyone sees in the survey once. Put each cell's concept questions together, with any same-concept comparison after its first-pack questions.
- 2Open Random assignment and quotas in the builder
In the advanced JSON editor, enable randomAssignment and define 2–50 cells. Give each cell a unique ID, label, and questionIds list. A question absent from every cell list is shared; a question may belong to more than one cell.
- 3Set only approved caps
Optional quota on a cell caps completed responses plus active reservations in that cell. totalQuota caps the whole study. Leave them out if targets are not yet approved.
- 4Add audience targets when needed
Use audiences for cohorts with separate total caps and optional per-cell caps. Put their shared screener questions before every cell question and turn question randomization off. Numeric ranges can classify birth years; values match authored answer options exactly.
- 5Mark mutually exclusive checkbox answers
On a checkbox screener, set question.config.exclusiveOptions to an array of exact authored option labels, such as ['None of the above']. Selecting one clears other choices; selecting another choice clears it. The assignment and submit APIs reject contradictory combinations.
- 6Check each cell in Brand Preview
Brand Preview runs the audience screeners before showing the selected cell. It does not reserve a place, save a response, or increase a completion count. Check the order and images on a phone-sized screen as well as desktop.
What each cap counts
| Setting | Counts | When it fills |
|---|---|---|
| Cell quota | Completed responses plus active, unexpired reservations in that cell | A new respondent is assigned to another available cell. |
| Study totalQuota | Completed responses plus active, unexpired reservations across all cells | New arrivals see the study-full result. |
| Audience quota / cellQuota | Completed responses plus active reservations in that audience, optionally per cell | A qualifying respondent sees a full result before any cell-specific question appears. |
| Answer quota | Completed responses matching all of its question predicates | A matching in-flight respondent can see the full result at submit; no completion or panel reward is issued. |
Resume and analyse
A browser returning with its saved response ID resumes in the same cell. A respondent entering from Genie Perks uses a signed panel identity and can resume across devices. An expired reservation may reclaim its original cell if room remains; otherwise the survey is full for that respondent.
Results filters can cut by assigned cell, audience and survey answers. Individual response exports include assigned cell and audience IDs and labels, plus assignment time. Balancing uses current capacity, so abandoned presentations are released and cumulative impressions may diverge.
Assignment settings and question order lock when the survey is published. A different cell design needs a duplicate survey. Question randomization is a separate feature and should stay off when the order of concept evaluation matters.
The existing answer-option shuffle does not save which A/B package appeared left or right. If position effects matter, use separately authored cells and encode the position in each cell ID or label; this editor does not yet rotate a paired image automatically.
Each answer quota sets a maxCompleted value and one or more question predicates. All predicates must match; a predicate can list several accepted answers. These are absolute completed-response caps. They do not target unverified profile demographics, percentages, or scheduled quota resets.