Reviewing and selecting segments

Top of the page

The header shows your project name, the audience type (B2C or B2B), and the regions you pinned during setup. These are quick reference; the actual decisions happen below.

The critic banner

Right under the header you’ll usually see a quality check banner. It runs a separate agent over the assembled audience and reports something like Quality check: 6/7 passed. The banner is green when everything passes, amber when there are warnings, red when something is blocking. Click to expand and you’ll see the individual checks.

Common checks include segment plausibility, population shares, and OCEAN consistency. Each carries a verdict: pass, warn, fail, or skipped. A grounding score (the % of claims tied directly to evidence) is often shown here too.

Treat warn as a nudge to look closer at that segment before continuing. Fail is a stop sign. The system won’t let you generate personas while a blocking failure is unresolved.

Segment cards

Each segment is a meaningful subgroup inside your audience, defined by a primary differentiating dimension. That’s the specific behavioral or psychographic criterion that explains why members of this segment make different choices from others. It’s not a demographic split. Two same-aged buyers in the same city can sit in different segments because they evaluate decisions differently.

On each card you’ll see:

Segment name — the human label, e.g. Product-focused early adopters.
Population share — a percentage like 28% showing what fraction of your final personas this segment will produce. If the share was derived from data rather than estimated, it’s tagged as data-grounded.
Description — a few sentences on who this segment is and what makes them distinct.
Supporting signals — an expandable footer showing how many evidence signals back this segment, broken down by provenance.

Reading the supporting signals

This is where the depth lives. Click the supporting-signals footer and the segment expands to show every signal that fed it. For each signal you get:

The implication — the claim itself.
A theme badge — behavioral, pain point, attitude, constraint, goal, belief, preference, or decision rule.
A provenance dot — colored marker showing how the signal was sourced. Grounded means it traces to a single specific source. Inferred means it synthesises across multiple sources. Calibrated means it draws from validated behavioral distributions. Hypothesis is an explicit working assumption with no source backing yet.
Confidence — a percentage shown when below 100%, useful for spotting weakly-evidenced claims.
Source badge — a violet pill if the signal came from a document you uploaded, a sky-blue pill if it came from web evidence (with a link out to the source).

If a segment’s signals are mostly hypothesis with low confidence, that’s the kind of segment to scrutinise before including. The personas you generate from it will be necessarily speculative.

The right sidebar

A sticky panel on the right summarises the whole audience. It shows the audience type, total segment count, region, and two science-grounded sections: OCEAN priorities (which Big Five traits matter most for this audience, ranked high/medium/low) and primary biases (the cognitive biases the audience exhibits most prominently). These are the levers that’ll show up in personas.

The selection gate: choosing which segments build personas

Once the critic check is resolved, the page surfaces a panel titled Build Personas. This is the gate. You decide which segments turn into personas.

Each segment has an include toggle. By default, every segment is included unless its evidence is too thin (Candor flags those as low-confidence and starts them off-by-default). When you toggle a segment off, the population shares of the remaining segments renormalise. Personas you don’t generate now can’t be added later inside the same project, so make sure your selection covers the audience you actually want to interview.

You need at least two segments included to proceed. Hit the Build Personas action and a confirmation dialog tells you how many archetypes and personas will be generated and how long it should take (typically 8-12 minutes).

How to make a good selection

Three rules of thumb:

Cover the spread. If you only include segments that already agree with your hypothesis, your interviews will tell you what you already think. Include dissenters.
Skip empty segments. A segment built on two hypothesis signals isn’t useful evidence; it’s speculation. Skip those, or add more uploaded research and regenerate the audience.
Don’t over-include. Six segments produce more personas, more interviews, more synthesis to read. Three to four well-evidenced segments usually beat eight thin ones.