AI in user research: Building ethical and trustworthy insights

Over the past year, we’ve seen a noticeable shift in how organisations approach user research. Across both our own delivery work and conversations with clients, AI is increasingly being explored as a way to speed up analysis, manage growing volumes of research data, and deliver insights more efficiently.

At the same time, we’ve also seen growing uncertainty around where AI adds value, where it introduces risk, and how researchers can use it responsibly without weakening the quality or integrity of their work.

According to Maze’s Future of User Research Report 2026, nearly 70% of user researchers now use AI in their workflows, up 19% from last year. Used thoughtfully, AI can help teams manage growing volumes of research data, speed up activities such as transcript tagging and synthesis, and create more space for interpretation, critical thinking, and engagement with participants.

However, without appropriate guardrails, AI can also introduce risk, erode trust, and weaken research integrity. Responsible use requires AI to be legal, transparent, and accountable, with researchers remaining responsible for validating outputs and making final decisions. This article explores five key considerations for using AI in research responsibly.

1. Protecting research data

Uploading raw research data to commercial AI systems can introduce immediate risk, particularly where organisations have limited visibility or control over how data is stored, processed, or reused. Interview transcripts and research notes often contain personal, sensitive, or commercially confidential information. Once shared with external tools, this data may be stored or processed in ways that are not fully visible, including being transferred across jurisdictions or used to improve underlying models. This can make the information difficult to track, manage, or remove.

Data protection obligations still apply, regardless of the tools used. This includes personal data as well as commercially sensitive material such as client data or strategic insights. Even when direct identifiers are removed, indirect identifiers, such as job roles or specific experiences, can still make individuals identifiable.

Before using any AI tool, ask:

• Does this include personal or sensitive information?
• Could someone be identified directly or indirectly?
• Do I know how this tool stores and processes data?

If there is any uncertainty, avoid uploading the data in its original form. Anonymise transcripts and limit inputs to only what is necessary.

2. Maintaining transparency and accountability

AI can make it harder for others to understand how insights were produced and how data was handled. Without clear communication, stakeholders may not know where AI has influenced outputs, making findings harder to evaluate or challenge.

For this reason, AI use should be disclosed whenever it plays a role in analysis or reporting. It should be clear where and how AI has supported the research process, including which activities were automated and which were reviewed, interpreted, or decided by human researchers.

Research participants do not necessarily need to be informed about every use of AI in the research process, just as they would not normally be told about all other software tools researchers use. However, where AI plays a significant role in analysing participant data or influences interpretation and decision-making, organisations should be transparent about its use, as well as how data is anonymised, protected, and overseen by researchers.

Internally, teams should document which tools were used, for what purpose, what data was involved, and what checks were applied. This creates a clear audit trail and helps ensure the research process can be explained if needed.

3. Ensuring insights stay grounded

High-quality research depends on how closely findings reflect participants’ accounts, how well they align with the research objective, and how clearly insights can be traced back to evidence.

AI-generated outputs can produce confident-sounding summaries that are not fully grounded in the source material. They may smooth over nuance, prioritise dominant patterns, and overlook less prominent perspectives, introducing bias into the research process.

For example, a summary of interview transcripts might highlight the most frequently mentioned issue while missing a critical concern raised by a smaller group of participants. In some contexts, this can have implications for fairness, inclusion, or safety if decisions are made based on incomplete findings.

A 2025 Corpora UK Research Report, based on 500 UK researchers, analysts, and professors, found that only 58% regularly verify AI outputs before using them. This indicates that unvalidated results are already being incorporated into research workflows.

To mitigate this risk, outputs should be checked against the original data, with a clear link maintained between evidence and interpretation. Researchers should review underlying evidence rather than relying solely on AI-generated summaries, particularly in research that could affect services, policy, safety, or fairness.

4. Maintaining human judgement

AI can improve efficiency, but it can also change how researchers engage with their data. When outputs are generated quickly and presented clearly, they are more likely to be accepted without scrutiny.

This creates a risk of overreliance, where AI-generated summaries are accepted without sufficient validation. Over time, this can weaken familiarity with the data and reduce time spent interrogating transcripts and field notes, weakening the link between raw evidence and final insight.

Researchers still need direct engagement with the raw data. Returning to transcripts and field notes helps maintain context, challenge assumptions, and ensure conclusions remain grounded in evidence.

5. Applying greater care in sensitive research

Some research contexts require a higher level of care, particularly when working with vulnerable individuals or sensitive topics. In these cases, the consequences of getting things wrong can be more severe, and a higher ethical threshold should be applied.

Errors in these contexts are not just analytical. They can lead to real-world harm, such as services, policies, or products being designed in ways that exclude, misrepresent, or disadvantage the people they are intended to support.

For this reason, AI should only be used where there is a clear and justifiable benefit. Researchers need to consider whether its use could introduce risk, particularly when participants may be less able to challenge or correct the interpretation of their experiences.

Additional care is required in obtaining consent and handling data, with a strong emphasis on participants’ rights, dignity, and well-being. This includes being transparent about how data will be processed and ensuring that its use does not compromise trust.

In some cases, the most appropriate decision is not to use AI at all.

Moving forward

AI is a powerful tool, but it does not guarantee good research. Used well, it can improve efficiency and support better outcomes. Used poorly, it can expose data, distort findings, reinforce bias, and erode trust.

The ability to use AI in research does not automatically mean it should be used. Researchers must apply careful judgement based on the context, sensitivity, and potential ethical impact of the work, particularly where fairness, safety, or vulnerable groups are involved.

The value of research still depends on human judgement, context, and accountability. AI can support that process, but it cannot replace it. When used thoughtfully and responsibly, it can help researchers work more efficiently while maintaining the quality, integrity, and trust that meaningful research depends on.