Before launching a website redesign, optimising an e-commerce conversion funnel, or repositioning an offer, user interviews are at the heart of what UX experts do. They allow us to understand how users of a product actually think: what holds them back, what builds their trust, what makes them take action or not. These structured conversations are conducted to gather what data doesn't show. Today, AI tools promise to do the same thing, faster and at a lower cost. So, can user interviews be replaced by AI?
Pourquoi faire des entretiens utilisateurs ?
Web analytics tools measure what happens: how many people clicked, at what stage they dropped off, what content was viewed. This is quantitative analysis. What these tools don't tell you is why. Why this form is off-putting, why this page generates poorly qualified leads, why a logical user journey on paper doesn't match how your customers actually think.
These are precisely the questions that user interviews exist to answer. These are structured conversations, conducted with people representative of the target audiences, to understand their needs, their barriers, and the logic that guides their decisions. The aim is not to collect opinions, but to observe how they think in the face of a real situation and to gather qualitative analysis.
- Why would a qualified visitor leave without making contact
- What vocabulary do the targets use to describe their problem and does it match that of the website?
- At what point in the journey is trust built or lost?
- What decision criteria are not addressed in the content?
Les entretiens utilisateurs s'inscrivent généralement dans les premières étapes d'un projet, souvent pendant la phase de découverte ou de recherche. Ils sont essentiels pour comprendre les besoins, les motivations, les comportements et les points douloureux des utilisateurs finaux. Ces informations aident à définir le problème, à identifier les opportunités et à guider la conception et le développement du produit ou du service.
User interviews typically take place before a significant decision: before a redesign, before the launch of a new service, or before reworking a conversion funnel. They allow you to ask the right questions before committing resources. Their role is to reduce the risk of building something well-designed but poorly oriented.
They can also intervene after a launch, to understand why the results are not as expected despite a technically sound system. In this case, they complement the quantitative analysis: the data shows where the problem lies, and the interviews explain why.
What AI can and cannot do in user interviews
The study Navigating the Jagged Technological Frontier (Dell’Acqua et al., Harvard Business School / Boston Consulting Group, 2023), a study of 758 consultants, highlights that AI significantly improves performance on formalised tasks – summarising, drafting, structured analysis – but impairs the quality of judgement on tasks requiring a nuanced contextual understanding. For the latter, participants using AI produced results that were 19% less reliable than those working without it. The authors refer to this phenomenon as the «Jagged Technological Frontier»: AI excels where the task is clearly defined and fails where it is not precisely defined.
Thus, AI tools can today analyse large volumes of verbatim feedback, group recurring themes in customer returns or generate summaries from transcripts. These uses are relevant and AI excels at them: they speed up the processing of what has already been collected.
However, AI cannot and should not conduct an interview in place of a human because it fails in this fine reading. This is the case with the physical and social experience that unfolds during a user interview: vocal intonation, gestures, hesitations, silences. The same statement, depending on the intonation, can be interpreted as negative or positive by a human, whereas AI will not perceive the nuance.
Jakob Nielsen, a pioneer in web usability, speaks of a gap and the AI's inability to capture users« behavioural nuances. The AI lacks understanding of the physical and social experience that allows a human researcher to grasp what is happening in the room, not just what is being said. A participant who replies »yes, that's very good" while crossing their arms and looking away communicates something that no AI language model will understand.
In the case of digital accessibility, Replacing user interviews with AI would mean missing the point. It is in usage and interaction that we can understand how a person with a disability navigates and the issues they may encounter. For example, two people using the same assistive tool might have radically different navigation strategies. These differences in usage only become apparent through interaction.
How to effectively articulate AI and user interviews?
AI and user interviews can absolutely coexist in a research process, provided each is assigned its appropriate role.
Before the fieldwork, AI can help structure the preparation: formulating an interview guide, generating hypotheses from existing data, identifying unexplored angles in previous research. This is useful shaping and priming work, but it does not replace the researcher's judgement to validate what is worth exploring.
On the ground, AI has no place as an interlocutor. However, certain automated note-taking or real-time transcription tools can relieve the UX expert of the burden of transcription, allowing them to remain fully available during the exchange.
After the field, it is perhaps where AI brings the most: transcribing, grouping verbatim responses, identifying thematic recurrences in a volume of data that a human alone would process more slowly and less systematically. Here, the gains are real. The Harvard/BCG study mentioned above confirms this: for structured and well-defined tasks (summarisation, categorisation, rephrasing), AI significantly improves processing speed and quality.
The challenge is not to pit AI against user research but to identify what can be automated and what requires contextual human interpretation. Using AI to group existing verbatim feedback is one thing. Replacing interviews with AI is another. In the first case, we speed up a process. In the second, we eliminate the source of information itself and make decisions based on assumptions that we believe we have verified but have never been tested against reality.
Overall, AI is a productivity tool for UX researchers and experts: it compresses processing time, reduces repetitive work and allows for larger volumes to be covered. However, it operates downstream of the collection process, not in its place. The UX expert remains the guarantor of what is collected, of the interpretation, and of the resulting recommendations.
User interviews allow us to understand why your targets behave as they do. They inform structuring decisions in digital projects to understand usage and avoid all assumptions.
AI finds its full place in this approach, but downstream and as a productivity tool: for transcribing, grouping, and identifying recurring patterns in what has already been collected. On these well-defined tasks, the gains are real and documented. What it cannot do is replace the encounter itself or capture what plays out beyond the words. The challenge is therefore not to choose between AI and user research, but to distinguish what can be automated and what requires the interpretation of a human situation. This distinction directly influences the quality of the decisions made in a digital project.