Introducing Visual Insights

What people say and what people do are two different datasets

You ask Gen Z how they feel about AI customer service agents. "I'd rather talk to a person," someone tells you, and they mean it. Then you watch them in a support chat. Two buttons: talk to a human with a six-minute wait, or talk to an AI agent now. They click the AI agent in three seconds.

Both things are true. The first one is what they believe about themselves. The second one is what they do. Great research captures both.

The tradeoff

If you run usability or screen-share research, you've been forced to make a tradeoff.

Moderated: a researcher sits in, watches the screen, catches the contradiction, and steers the participant back when they wander off. It works. It also doesn't scale, costs a lot, and eats your team's calendar.

Unmoderated: it scales, it's cheap, and a session is only as good as what the participant says and whether they follow directions. The say-do gap goes uncaught. The off-task session gets thrown out. And the only way to find any of it is to have someone re-watch hours of recordings, which rarely happens.

So you end up reporting what people claimed instead of what they did, and then defending findings without full confidence.

What we built instead

Visual Insights lets Listen Labs' AI Interviewer see the participant's shared screen and act on what it sees. On-screen behavior becomes data you can query, not footage you have to re-watch.

It works in two main ways.

It acts during the interview. When someone does the opposite of what they said, it catches it and asks follow-ups in real time. The question is based on what they actually did, not a script written in advance. And when a participant shares the wrong window or lands on the wrong tab, it redirects them so you don't lose the session.

It does the synthesis after. Say-do gaps, off-task moments, and the reasoning behind on-screen choices show up across sessions without anyone re-watching anything. And you can keep asking for various cuts in plain language. 

For example, if you ask the chat "What's the median and 90th percentile time people spent with the AI agent before asking for a human?" you get the numbers. If you ask "Pull a reel of people who said they prefer humans but chose the AI agent" you get relevant clips.

The pushback

The obvious objection is the right one: an AI that watches the screen sounds like a black box, and you’re not going to put your name on a finding you can't back up.

Fair. That’s why every finding is traceable. Click any on-screen metric and jump to the exact moment behind it (00:38, selected AI agent over "Talk to a person"). Export the raw data. Under the hood, a video model writes a timestamped second-by-second log that tags every meaningful change, so you have a record of what happened on screen the way a transcript is a record of what was said. If the model misread something, recode it, and the correction flows into the report. 

Is this just screen recording? No. A recording is something you re-watch. This is behavior the AI moderator probes live and you can query after, traced back to the moment it happened.

How customers are using it

Catch the say-do gap. Probe contradictions while they're happening and get the real reason, instead of missing them or digging through recordings three weeks later.

Quantify friction. Measure per-subtask time, medians, 90th percentiles, or click counts across participants instead of citing one memorable session.

Run usability research at scale. Get live probing and redirection across more sessions than a moderator could sit through, without dropping to shallow unmoderated tests.

Understand real shopping behavior. Watch what shoppers do when they navigate a site or react to a concept, not just what they tell you afterward.

See for yourself

Visual Insights works with the rest of Listen Labs. With Emotional Intelligence, you now know what they did and how they felt while doing it. And with Research Agent, fold the findings into the charts, slides, and reports you need to share your findings.

Visual Insights is live. If you're not a Listen Labs customer yet, reach out to our team to get started.

What people say is one dataset. Now you've got the data on what they actually do.