Executive Summary
The market research industry is a $90 billion global enterprise built on a methodology that has barely changed in fifty years: human interviewers asking questions of human respondents, one at a time, at significant cost and considerable delay. That model is being disrupted — not gradually, but abruptly — by the convergence of large language models, conversational AI, and adaptive probing systems capable of conducting expert-quality qualitative interviews at machine speed and population scale.
This white paper synthesises the evidence from forty peer-reviewed papers spanning 2019 to 2026 — the complete arc of the AI interviewing literature. The findings are striking in their consistency: across multiple independent research groups, using diverse populations and methodologies, AI-conducted interviews now match or approach human interviewer performance on the metrics that matter most, while reducing cost by an order of magnitude.
"In 75% of matched pairs, doctoral researchers in sociology rated the AI interview transcripts as more informative than the self-written open-text responses." — Geiecke & Jaravel, 2025
What You'll Learn
From the historical tension between standardised and active interviewing, to the information-theoretic case for adaptive probing, to the ethics of AI-mediated research — this paper covers the complete science behind AI interviewing.
The Paradigm Shift: From Surveys to Conversational AI
AI-moderated interviewing creates a third position in the long-standing debate between standardised and active interviewing — achieving perfect standardisation while simultaneously implementing adaptive probing with a consistency no human can match. This section traces the evidence from Xiao et al. (2019) through Geiecke & Jaravel (2025) and explains why this is a genuinely new mode of data collection, not an incremental improvement on surveys.
Architecture, Rapport & the Machine Heuristic
Rapport is not a monolithic quality — it is a set of designable behaviours. This section covers architectural best practices for building effective AI interviewers, the counterintuitive finding that participants disclose more to AI in sensitive contexts (the 'machine heuristic'), and the current limitations that remain an active research agenda rather than permanent barriers.
Information Theory & Adaptive Probing
Sun et al. (2025) found LLM interviewers elicited 0.90 unique correct information items per question versus 0.46 for human interviewers — nearly double the efficiency. This section presents the information-theoretic foundations for why adaptive probing works, the empirical case across multiple independent studies, and design principles for maximising information yield per question.
Scale, Quality Parity & the Authenticity Challenge
From N=20 to N=2,000: the evidence for quality parity across five independent research groups, the practical implications of scaling qualitative research, and the provocative authenticity challenge posed by Cox, Shirani & Rouse (2024). Plus: participant experience data showing 53% strictly prefer AI over human interviewers.
82%
of participants rated AI interview experience positive (Chopra & Haaland, 2023)
142%
more words written in AI interviews vs. open text fields (Geiecke & Jaravel, 2025)
10×+
cost reduction vs. human-administered qualitative interviews
Excerpt from the Report
"The paradigm shift is not merely technical. It is methodological. For half a century, qualitative researchers have debated the tension between standardised survey interviewing and the 'active interviewer' model — between reliability and depth, between scalability and richness. AI-moderated interviewing dissolves this tension. A well-designed LLM interviewer achieves standardisation more perfectly than any human — never fatigued, never inconsistent, never suggestive — while simultaneously implementing active listening, adaptive probing, and conversational responsiveness with a consistency no human interviewer can replicate."
"The most striking finding across multiple studies is that participants often prefer AI interviewers to both human interviewers and traditional survey formats. 53% strictly preferred the AI interviewer over a human — a remarkable finding for a technology still in relative infancy. The conditions under which AI interviewers are preferred vary by interview domain, topic sensitivity, and participant characteristics — a design parameter, not a fundamental limitation."
Continue reading in the full report…
Get the Full White Paper →Who This Report Is For
UX & Product Researchers
Understand exactly where AI interviewing matches human performance, where it falls short, and how to design AI interview instruments that produce academically defensible data — grounded in 40 peer-reviewed papers.
Product Managers & Founders
Get the evidence base you need to confidently replace slow, expensive human interviews with AI-moderated research — and understand which use cases are ready today versus which require caution.
Market Research Leaders
Assess the competitive threat and opportunity that AI-moderated interviewing represents to the $90 billion market research industry — backed by the complete literature from 2019 to 2026.
AI Builders & Platform Teams
Access the design principles, architectural patterns, and ethical frameworks needed to build AI interviewing systems that work — including the emerging standards for quality evaluation and authenticity monitoring.
Download the White Paper
Enter your email and we'll send it straight to your inbox.
Check Your Email
We've sent the white paper to .
The download link should arrive within a few minutes. Check your spam folder if you don't see it.
While you wait
Explore Deutero.AI →