Reviewed and fact-checked: 21 July 2026.
User interviews help product teams understand goals, language, context and reported experience. They are not the same as usability testing: what a participant says they do may differ from what they do when completing a task. An AI voice recorder for user interviews can reduce note-taking pressure and make exact moments easier to revisit, but it cannot replace informed participation, observation, sound moderation or disciplined analysis.
Quick verdict: define the product decision and research question, obtain informed agreement for each recording type, capture speech and behaviour as separate evidence, correct the transcript before coding and link every reported insight to checked source material.
Commercial disclosure: NERALVO sells the Halo AI voice recorder. This guide describes a general UX-research workflow and does not claim that AI can create valid insights, approve product decisions or determine the lawful and ethical process for a specific study.
Match the method to the research question
| Question | Suitable method | Main evidence |
|---|---|---|
| How do people understand this problem? | In-depth interview | Language, goals, history and context |
| Can people complete this task? | Moderated usability test | Observed behaviour, errors and task outcome |
| What happens in the real environment? | Contextual research | Workflow, artefacts, interruptions and constraints |
| How common is a pattern? | Survey or quantitative analysis | Measured distribution across an appropriate sample |
| Which design performs better? | Controlled comparison or experiment | Defined outcome measures |
Do not use an interview alone to claim that a design is easy to use. Participants may describe preferences without revealing whether they can complete the task successfully.
Define the research decision
Before recruiting, write:
- The user group and context.
- The decision the team must make.
- The behaviour, understanding or friction being investigated.
- What evidence would support or challenge the current assumption.
- Which product stage or prototype is involved.
- How findings will be recorded, shared and retained.
Weak question: “Do users like our new checkout?”
Stronger question: “Can first-time mobile users identify delivery cost, correct an address and understand when payment will be taken?”
Recruit participants relevant to the decision
Participants should reflect the people, situations and access needs the product decision concerns. Consider:
- Experience level and frequency of use.
- Device, connection and assistive-technology context.
- Language and literacy needs.
- People who have abandoned or avoided the service.
- Different routes through the task.
- Users who need help online or face exclusion.
- Relevant organisational, security or purchasing roles.
A convenient sample of colleagues can support an early rehearsal, but it should not be presented as representative user evidence.
Obtain informed agreement for each capture method
GOV.UK user-research guidance says participants should understand who is doing the research, its purpose, the data being collected, what will happen, how results will be used and shared, recording arrangements, retention and their ability to stop or withdraw within the stated process.
Audio, video, photography, screen recording, prototype data and remote observation are separate activities. Explain each one rather than using a vague blanket statement.
| Capture | Explain |
|---|---|
| Audio | Who will hear it, transcription method and retention |
| Video | Whether face, environment or body movement is included |
| Screen recording | Which windows, notifications and personal data may appear |
| Remote observers | Who is watching and whether they can record |
| Quotations or clips | Where and to whom extracts may be shown |
| AI processing | Which approved processor handles the recording or transcript |
How NERALVO Halo may support user research
NERALVO Halo is an ultra-slim, phone-mounted AI voice recorder with NOTE mode for suitable in-person interviews, supported CALL mode for lawful and disclosed remote sessions, 64GB local storage, up to 35 hours of recording and Bluetooth sync with the DOWAY app. DOWAY can generate transcripts, summaries, templates, translations, mind maps and exports. One year of DOWAY Max is included.
Halo captures audio, not the participant’s screen or complete physical behaviour. When interaction matters, link audio timestamps to authorised observation notes or screen recordings. The research organisation must approve the complete device, processor, access and retention workflow.
Prepare a neutral discussion guide
| Purpose | Neutral prompt |
|---|---|
| Understand context | Tell me about the last time you needed to do this |
| Explore goals | What were you trying to achieve? |
| Observe interpretation | What do you expect will happen when you select that? |
| Probe friction | What are you looking for now? |
| Clarify behaviour | What made you choose that route? |
| Explore workaround | What would you normally do if this failed? |
| Test meaning | What does that message mean to you? |
| Close | What was most difficult or unexpected? |
Avoid praising, correcting or teaching during a usability task unless the protocol requires an intervention. Assistance changes the evidence and should be recorded.
Capture speech, behaviour and context separately
| Evidence type | Example |
|---|---|
| Participant statement | “I expected delivery to be free.” |
| Observed behaviour | Participant scrolled past the delivery-cost panel twice |
| Task outcome | Address correction completed after one failed attempt |
| Environmental context | Participant used a small screen in bright outdoor light |
| Researcher interpretation | Delivery information may lack prominence |
| Design implication | Test earlier cost disclosure in the next prototype |
These categories should not be collapsed. One participant saying a label is clear does not outweigh repeated observed confusion.
Use time-linked observation notes
Audio alone may not show which button was selected or when the participant hesitated. Record:
- Timestamp.
- Task or screen.
- Participant action.
- Spoken comment.
- Outcome or error.
- Moderator intervention.
- Context or accessibility issue.
- Question for later analysis.
This creates a traceable connection between the transcript and the observed interaction.
Correct the transcript before coding
Check:
- Participant and researcher speaker labels.
- Product, screen and task names.
- Dates, amounts and references.
- Negation and conditional wording.
- Whether a comment was prompted by the moderator.
- Exact quotations selected for reporting.
- References to visible behaviour that require observation notes.
- Unclear or overlapping speech.
Do not mistake moderator language for user evidence
If the moderator asks, “Was the button too small?”, the resulting answer is influenced by the question. Record leading prompts, hints and interventions so the analysis does not present the response as unprompted evidence.
Build an evidence table before writing insights
| Finding field | Required content |
|---|---|
| User and context | Relevant participant characteristics without unnecessary identity |
| Task or goal | What the person was trying to achieve |
| Evidence | Observed behaviour, outcome and checked quotation |
| Frequency | How many relevant sessions showed the pattern |
| Severity | Effect on completion, trust, safety or access |
| Contradiction | Evidence that does not fit the pattern |
| Interpretation | Researcher explanation, clearly labelled |
| Design implication | Decision or experiment the evidence supports |
| Confidence | Limits from sample, method and data quality |
Preserve outliers and negative evidence
AI summaries often emphasise repeated language. A single outlier may still expose a serious accessibility, safety, legal or service-failure risk. Review:
- Participants who could not complete the task.
- People using assistive technology.
- Unexpected workarounds.
- Contradictions between stated confidence and observed behaviour.
- Segments excluded by the design.
- Evidence that challenges the team’s preferred solution.
Separate insight from recommendation
Evidence: four of six first-time mobile participants opened the help page before finding the identity-check requirement.
Insight: the requirement is not visible when users decide whether they can complete the application.
Recommendation: test showing the requirement before the application begins.
The recommendation is a product choice, not something automatically proved by the transcript.
Use AI-generated themes as a starting point
AI can group repeated terms, produce first-pass summaries and locate quotations. It may ignore screen behaviour, overcount repeated moderator language, miss sarcasm or combine different user groups. Researchers should check every material theme against corrected transcripts, observations and the research plan.
Protect participant identity
Treat notes, recordings, screenshots and prototype entries as research data. Use participant identifiers in working files, keep contact details separately, remove unnecessary personal information and restrict access. Exact role, location, organisation and distinctive quotations can identify a participant even when their name is removed.
High-risk research needs specialist arrangements
Research with children, vulnerable adults, emotionally sensitive subjects, health information, financial difficulty, safeguarding concerns or high-risk services may require specialist recruitment, consent, safeguarding, accessibility and escalation procedures. A routine product-interview workflow should not be used without appropriate organisational expertise.
A controlled end-to-end workflow
- Define the decision, research question and method.
- Recruit relevant participants and plan accessibility.
- Provide understandable information and obtain required agreement.
- Test the recording and observation setup.
- Run the session with neutral moderation.
- Secure audio, notes, screen material and consent records.
- Generate and correct the transcript.
- Link speech to behaviour and task outcomes.
- Analyse consistently across sessions.
- Preserve contradictions and high-severity outliers.
- Create evidence-based findings with limitations.
- Link findings to product decisions and future tests.
- Share only authorised, minimised outputs.
- Apply participant requests, retention and deletion rules.
Frequently asked questions
Can AI create UX insights automatically?
It can organise text and suggest patterns. Valid insights require context, observation, method and researcher judgement.
Is a user interview the same as usability testing?
No. Interviews explore reported experience and meaning; usability testing observes people attempting tasks with a design.
Can remote interviews be recorded through CALL mode?
Supported apps may work, but compatibility varies. Recording must follow the approved lawful and transparent process.
Should every participant quotation appear in the report?
No. Use quotations that accurately illustrate a supported finding and preserve context.
Does removing a name anonymise the research?
Not necessarily. Indirect identifiers and distinctive details can still reveal identity.
Should the whole product team access raw recordings?
Only where the research plan permits it. Share the minimum authorised evidence necessary for the decision.
Keep behaviour connected to the product decision
A reliable interview record helps teams understand people rather than merely collect opinions. Informed participation, neutral moderation, observed behaviour, corrected transcripts and disciplined analysis turn recordings into defensible product evidence.
Explore NERALVO Halo for approved user interviews, transcription and structured research notes.
Ready to capture meetings properly?
View the NERALVO Halo AI voice recorder with 64GB local storage, meeting capture, compatible phone-call recording workflows and one year of DOWAY Max included.
View NERALVO Halo