Mind or Machines: Exploring AI moderation and when to use it

In the rapidly evolving world of market research, AI moderators offer revolutionary scalability and efficiency. AI-powered moderators can conduct thousands of interviews 24/7, offering unparalleled scalability, whether working at home, on-site or on the go.

But can a machine truly replicate the nuances, empathy and insights of a human moderator? In this paper, we address this crucial question. Through a series of rigorous, multi-country pilot studies, our experts compared the performance of standard AI bots, enhanced AI systems and expert human moderators.

The challenge: AI lacks the human touch

Initial research confirmed what many suspected: while standard AI moderators excel in terms of engagement and scale, they perform poorly in certain key areas. In blind tests, expert human moderators rated AI-led interviews poorly in the following areas:

• Building rapport with participants.
• Improvising and deviating from the script to probe deeper.
• Adapting to a respondent's unique communication style.
• Detecting unarticulated nuances and emotional cues.

The solution: teaching AI to be more human

We hypothesised that the key to improving AI moderators lies not only in technology, but also in psychology. A standard AI bot was enhanced by embedding human-like expertise, including:

• Subject and context mastery: The AI was given in-depth knowledge of the research topic and cultural context to help it ask more relevant follow-up questions.
• Behavioral frameworks: Using models such as COM-B, the AI was trained to better identify the discrepancies between what people say and what they do.
• Emotional intelligence: Integrating the proprietary Ipsos Emotion Framework enabled the AI to interpret emotional states based on valence (pleasantness), arousal (intensity) and control.

It’s not mind or machines, it’s mind AND machines 

The results were clear: each additional layer of human-centric knowledge significantly improved the AI moderator’s performance, resulting in higher-quality data and a dramatic increase in the identification of emotional themes. However, human moderators still had the edge when it came to uncovering deep, unexpected insights by following 'emotional breadcrumbs' and adapting with true empathy. 

The future of qualitative research lies in powerful collaboration between humans and AI, not in choosing between them. The most effective approach is a hybrid model in which AI acts as a co-moderator. 

AI-led research is ideal for answering large-scale tactical questions where speed and efficiency are paramount. However, human-led research remains essential for in-depth strategic analysis requiring profound empathy and nuance. The 'best of both worlds' approach involves human experts designing, guiding and concluding the research journey while leveraging AI to achieve scale, ensuring depth and quality. 

This Ipsos Views paper is predominantly based on an extended paper (of the same name) presented at ESOMAR North America 2025.