The new normal

A few years ago, analysing ten user interviews took me days. I listened to recordings, wrote notes and moved sticky notes around until patterns appeared. Today, AI tools can transcribe, tag and group those interviews in minutes. This is not a small change. In Lyssna's survey of UX researchers for 2026, almost nine in ten named AI-assisted analysis and synthesis as the biggest trend in the field.

But faster is not the same as better. After using AI in my own research for more than a year, I see a clear line between the tasks where it helps and the tasks where it quietly makes things worse.

Where AI really helps

Preparation. Before I talk to users, I ask AI to help me learn the vocabulary of a new market, draft interview questions and spot leading questions in my script. It is like having a patient colleague who reviews my plan.

Transcription and first-pass tagging. AI is very good at turning hours of audio into searchable text and suggesting first themes. I still read the transcripts, but I start from a structure instead of a blank page.

Connecting studies. When a product has many small studies, AI can look across all of them and remind me that users said something similar six months ago. This is where research becomes a memory, not a pile of reports.

Where I stop: synthetic users

"Synthetic users" are AI personas that answer questions as if they were your customers. The idea is attractive: no recruiting, no scheduling, results in seconds. Nielsen Norman Group tested this by repeating studies they had already run with real people. Their conclusion was that the AI answers were often too shallow to be useful. Synthetic users tend to like every concept and list many needs without knowing which ones really matter.

The problem is simple. A synthetic persona is built from general patterns. Real research is valuable because it finds the things nobody expected: the strange workaround, the emotional reason behind a drop-off, the feature people say they want but never use. AI cannot find what is not already in its training data.

My rule of thumb

I use one simple test before any AI-generated "insight" goes into a decision: Would we act on this without checking with a real person? If the answer is yes, I am using AI as a research participant, and I stop.

So in practice:

• AI helps me prepare and process.

• Real people give me the answers.

• I make the decision, and I own it.

What this means for designers

AI makes research cheaper, which is good news for small teams that used to skip it. But it also makes fake research cheaper. The designers who will stand out are the ones who still go and watch real people use real products, and then use AI to move faster from what they saw to what they build.