Blog

How Lens uses AI in UX research (and how it handles your data)

July 8, 2026 · 6 min read

AI in research tools tends to arrive as either hype or a black box — a magic button that summarizes your work and asks you to trust it. Lens takes the opposite approach: the AI is deliberately narrow, transparent, and under your control. It helps you see patterns in your own research faster, and nothing more.

The principles behind it

• Grounded, not trained — the model only sees a snapshot of the relevant repository data at request time. Your research is never used to train a model.
• Assists, never decides — every AI output is clearly labelled and fully editable. It supports your judgment; it doesn't replace it.
• Your key, your provider — bring your own Anthropic (Claude) or OpenAI key. Nothing runs without it, and you choose the model.
• Off where it counts — every feature has a switch, output is editable, and viewer accounts get no AI at all.

What the AI actually does

Four grounded features, each optional and each pointed at your repository rather than the open web:

• Ask your repository — a chat grounded in your research that answers questions and references the insights, interviews, and themes behind its answers.
• Surface patterns — scan your data for duplicates, emerging themes, persona-coverage gaps, and opportunities worth a look.
• Summarize interviews — turn a transcript into a summary, key observations, and (if you enable it) an overall sentiment — all editable.
• Cluster affinity notes — group sticky notes into candidate themes to kick-start analysis, which you then accept or rearrange.

What data is sent — and what isn't

Only the context a request needs is sent: a snapshot of your repository for the assistant and suggestions, the interview material for a summary, or the notes for clustering. With Redact PII on — the default — emails and phone numbers are stripped from that context before it ever leaves the server. The provider runs inference only: it is a sub-processor, not a training pipeline, and Lens sends nothing in the background. AI runs when you ask it to, on the data in front of you.

Why grounding matters

The failure mode of AI in research is confident invention — a tidy summary of findings that were never actually said. Grounding the model strictly in your own data keeps its answers traceable: you can follow a claim back to the interview behind it. And when the data doesn't cover something, a grounded assistant says so and suggests what research would close the gap, rather than filling it with plausible fiction.

You stay in control

Every feature is gated by a toggle in Settings, so an admin can switch off any capability — or all of them — for the whole workspace. Access follows role: researchers, designers, and product owners get AI; viewers don't. Provider, model, and keys are admin-only. Nothing is on that you didn't turn on.

How this fits the bigger picture

Treating AI as a sub-processor with redaction and least-necessary data is part of how Lens treats research as personal data throughout. For the full picture, see GDPR-compliant user research and the use cases for how teams put it into practice.

← All posts