Definition
A cohort is a segmentation contract: personas in the group should be similar on the dimensions relevant to the research question and intentionally different from other cohorts.
Cohorts improve interpretability by making comparison explicit. They are the bridge between individual persona narratives and population-level signal.
Operational model
- Segmentation layer: define inclusion criteria tied to the research objective.
- Representation layer: ensure enough persona diversity inside the cohort to avoid caricatures.
- Comparison layer: evaluate response deltas between cohorts, not just within one group.
- Revision layer: adjust cohort definitions when signal quality drops or overlap grows.
Execution workflow
1. Define
Write explicit inclusion criteria (demographic, behavioral, or attitudinal) aligned to your study hypothesis.
2. Assemble
Create or select personas that match criteria without overfitting to a single stereotype.
3. Validate
Run a short calibration skill and verify the cohort responds distinctly from neighboring cohorts.
4. Refine
Adjust cohort membership when overlap, drift, or low-signal responses appear.
Quality checklist
- Inclusion criteria can be audited and explained in plain language.
- Cohort size is large enough to avoid one-person outlier dominance.
- Personas represent a range of experiences within the target segment.
- Segment boundaries are tested against at least one adjacent cohort.
- Calibration results are stored before launching a full study.
Failure modes to avoid
- Defining cohorts with vague labels that cannot be measured or audited.
- Overfitting cohorts to a narrative and losing internal variability.
- Allowing heavy overlap between cohorts, which blurs comparative results.
- Treating one successful run as permanent validation and skipping rechecks.
Related platform APIs
These APIs are the main building blocks used when this concept is put into production workflows.
/personasList persona candidates for segment assembly.
/personas/:personaIdReview fit against cohort criteria.
/personas/promptRun calibration prompts against cohort members.
/personas/prompt/sessionsCompare calibration outputs over time.