FAQ & methodology
How the analysis works, why the evidence holds up, and how we treat your data.
Methodology
How does the analysis actually work?
Once enough interviews are in, the agent transcribes and translates them, then runs thematic analysis across all of them at once. It groups the data into themes and insights, and anchors every insight to specific transcript moments. When a study design exists, themes map back to your objectives and guide sections. Analysis output is in English regardless of the interview language.
Why can I trust the quotes?
Every insight is anchored to at least one timestamped moment in a real interview. The strongest quotes are cropped verbatim from the original audio, never synthesized or re-voiced. If an insight cannot be anchored to a span, generation fails loudly rather than inventing a quote. Quotes are shown in the original language with an English translation beneath, and the audio is always the original.
What is “honest prevalence”?
We state how many participants a finding actually came from (“7 of 10 participants”), not vague words like “most.” Disconfirming voices are surfaced alongside supporting ones, and each insight carries a confidence rating based on how many participants raised it, how it spreads across segments, and whether anything contradicts it.
What research methods do you support?
At this stage, 1:1 in-depth interviews. That is the method behind most qualitative work, and focusing on it keeps transcription, diarization, and analysis quality high. Focus groups and other methods may follow.
Human and AI interviewing
Human interviewer or AI interviewer?
Human-led is our default, because engagement and depth are highest when a person is genuinely listening. We offer a fully-AI interviewer as an option, but we only recommend it in two cases: when no researcher is available to moderate, or when subjects would be more honest with a machine than a person, which can happen with sensitive topics.
Does the AI replace the moderator?
No. It co-moderates: a private panel gives you a live transcript, pacing against your guide, and suggested probes. You keep the conversation, the judgment, and the client relationship. The participant sees a human, not a bot.
Do I have to change how I run interviews?
No. Keep interviewing on Zoom, Meet, or Teams. QualDive works alongside the tools you already use; there is no workflow to rip out.
Data & trust
Does my data train AI models?
No. Your client data is never used to train AI models. It is encrypted in transit and at rest, consent-aware by design, and erasable on request.
How do you handle consent and deletion?
Consent is tracked per participant, and erasure is a built-in path rather than a manual favour: on request, a participant's data is removed and the cascade reaches the derived analysis. Participants' voices belong to them.
What languages can we field in?
Fieldwork can run in many languages; transcription and translation handle the original recordings, while analysis and the report are delivered in English. The original voice is always preserved in the audio.
Getting started
What do I need to bring?
Your interview recordings or transcripts. If you also design the study in QualDive, the analysis maps back to your objectives automatically; upload-only studies still analyze.
What is live today, and what is coming?
Study design and the analysis-to-report loop are in beta now. The live interview co-pilot is in development. We tag each stage honestly rather than implying everything is shipped.
Still have a question?
Ask us anything, or see the evidence model in a finished report.