The Right Sample Size for AI-Moderated Interviews

The right sample size for AI-moderated interviews is not a number. It is a logic. one that most researchers borrow from the wrong context and apply without thinking.

Key Takeaways

• Saturation logic is real but segment-specific: n=20 may be enough within one coherent segment, but not across five.
• AI-moderated interviews shift the economics of scale, making segment-level saturation affordable where budgets previously forced approximation.
• The value of larger samples is not depth on a single audience. it is coverage across multiple segments, geographies, and longitudinal waves.
• The same saturation principles that govern human-moderated qual govern AI-moderated qual; what changes is how much it costs to reach them.
• Bad recruitment at any sample size produces bad data; scale amplifies recruitment error, not just signal.

The Saturation Principle Still Holds

The received wisdom. you saturate at around 20 interviews per segment. is grounded in real research. Guest, Bunce, and Johnson's foundational saturation work found that most thematic content emerged within the first 12 interviews, with rare themes appearing beyond 20. That finding is methodologically sound. What happened next is where the field went wrong: n=20 became a budget number dressed up as a methodology number.

Saturation is segment-specific. Within a single, coherent, well-recruited segment. one audience, one research question, good probing. 20 deeply conducted interviews will likely saturate. But a study labeled "national" that has 20 interviews spread across five consumer segments has four interviews per segment. That is not saturation. That is sampling anxiety managed by averaging.

The saturation logic does not change for AI-moderated interviews. What changes is the cost of reaching saturation per segment.

What Scale Actually Buys You

The weak argument for running more AI-moderated interviews is that they are cheaper. The strong argument is that scale changes what research can honestly claim.

Consider a brand team testing reactions to a repositioning across three distinct customer cohorts. loyalists, lapsed users, and new-to-category considerers. Human-moderated qual at traditional pricing might deliver 8 interviews per cohort. AI-moderated qual at a fraction of the cost can deliver saturation-level depth across all three. The finding is no longer an average across an implicit blend; it is a segment-specific insight with its own thematic structure.

Scale also enables two things traditional qual budgets rarely permitted: geographic honesty and longitudinal tracking. A study fielded only in Tier-1 metros produces Tier-1 insights. Enumerate's asynchronous AI moderation makes it economically viable to include Tier-2 and Tier-3 markets without proportional cost increases, because there is no moderator travel, no scheduling coordination, and no per-interview labor. And when cost-per-wave falls, running a second wave six weeks later becomes a research decision, not a budget exception.

For agencies running studies across multiple client segments, and for in-house teams responsible for covering multiple personas, this is the practical unlock: qual at scale is not more interviews on the same audience. it is honest coverage of the audiences that were previously skipped.

The Numbers That Actually Matter

A working heuristic for setting sample size in AI-moderated work:

• Per segment, per distinct use case: aim for saturation, which means at least 15-20 well-probed interviews within a coherent group. Fewer is a judgment call that should be documented, not a default.
• Across segments: multiply. If your research question spans three segments, three geographies, or three product lines, each warrants its own saturation logic. not a shared pool.
• For longitudinal work: smaller per-wave samples are defensible if the question is directional change rather than first-time saturation. A wave designed to detect theme shift can be lighter than a wave designed to establish themes.

The bottleneck was never the interview count. It was the cost of reaching the right count across every segment that mattered. That constraint has changed. The methodology underneath it has not.

Curious how this applies to your current study design? Book a demo with Enumerate and bring your brief.