What it is
Visual Insights gives Listen a read on what participants actually do on screen, like where they click, what they navigate to, what they skip, instead of relying only on what they say out loud and a passive screen recording requiring extensive manual review.That on-screen signal does two jobs: it sharpens the interview, and it carries into analysis. In the moment, the Listen Interviewer can ask follow-ups based on what was said and done together, including where the two don’t line up. After the interview, that same on-screen behavior becomes a signal the report and analysis can read across every session and surface trends, gaps, and contradictions between what people said and what they actually did, and becomes completely queriable with the Research Agent chat. It’s steered by your study goals throughout. Whatever you tell it to care about, it’s more likely to watch for live and look out for it in the Report. Say your goal flags that you care about search methods: if a participant types a weirdly specific query, the Interviewer is more likely to ask why they searched it that way in the moment, and that behavior is more likely to surface when the report gets built.
What it does + when it’s useful
In the interview:- Probing on observed behavior. A participant uses some filters but not others. The moderator asks why they chose the filters they did, and they explain they use hard filters for the things they’re strict about (size, price) and leave the rest open to browse on vibes. Steered by your study goals, so if you flagged that you care about how people navigate or filter, it’s more likely to dig in there.
- Catching say-do gaps. A participant says early on that they care a lot about reviews when picking a product. Then on screen, they pick a pair of trainers without ever opening the reviews. The moderator catches the disconnect and asks why (it turns out there weren’t enough of them to feel trustworthy). That’s an insight you’d completely miss if Screen Observation didn’t catch the action and probe on it.
- Recovering off-track sessions. Tale as old as time in unmoderated/AI-moderated testing: someone gets ahead of the instructions or lands on the wrong thing, and the whole session becomes useless. Now, if a participant is doing a running-shoes task but goes to Nike and picks Air Jordans, the moderator says “Can you find running shoes instead?” and gets them back on track. It also handles the small stuff like “I don’t know if I’m screen sharing right now,” and keeps people moving.
The report and chat agents can use Screen-observed behavior in the same way they use a normal spoken transcript: you can query it, quantify it, and pull it into the report any way you like, giving researchers many ways to explore, compare, and validate their data.
An example: Music streaming app study
The task was to build a playlist of five songs for someone they love.
87% of participants rated the task “easy” or “very easy.” But their on-screen behavior told a different story: across sessions, only a handful actually added all five songs in the task window despite all saying they completed it, and 28% displayed meaningful navigation frictions (opening and closing menus, backtracking, etc.) along the way. For those that did complete the full task, it took 6-10 minutes, much longer than the expected ~4. And it’s all traceable back to what people actually did. From the “28% experienced navigation frictions” metric, you can drop into the behavior underneath it: which participants it happened to, what the snags were, and from there jump to the exact timestamp, open that moment in their screen recording, and watch it happen for yourself (the same UI you use across the Listen analysis suite today). You can also ask the Research Agent chat about their on-screen behavior, things like “how many people hit navigation confusion,” “create a highlight reel of people who said X but did Y,” and it’ll pull it together.