Key takeaways
- →Research published in Nature found that pairs working over video generated significantly fewer creative ideas than pairs in the same room.
- →The same research found no penalty when it came to selecting the best idea from a set, so the deficit is specific to divergent generation.
- →The proposed mechanism is narrowed visual attention: the screen holds your gaze, and a narrow field of focus narrows the associations you reach for.
- →The practical response is to move idea generation off the shared video window and into silent, individual, low-stimulus time before the call.
- →Anonymous written submission also removes production blocking, which suppresses output in in-person brainstorms just as reliably.
- →Judge the change by counting distinct ideas per person per round rather than by how lively the call felt.
The video calls creativity question has an unusually clear answer for a workplace debate: a large study published in Nature in 2022 found that pairs generating ideas over video produced fewer creative ideas than pairs doing the same task in the same room. It held in a controlled lab setting and again in a field experiment with engineers across five countries.
The result that gets left out of the headline is just as important. When those same pairs had to pick the best idea from a set, the video pairs did as well as, and in some measures better than, the in-person pairs. The deficit sits in generation, not in judgement.
That distinction is what makes the finding usable. You do not need to fly people in to run a workshop. You need to stop asking a room full of video windows to generate ideas together in real time, and move that one phase somewhere the screen is not holding everyone's eyes.
What the study actually did
Brucks and Levav ran a laboratory experiment with roughly six hundred participants, paired up and given a creative idea-generation task, half face to face and half over videoconference. The virtual pairs generated fewer creative ideas. They then repeated the design as a field experiment with around fifteen hundred engineers across five countries, and found the same pattern in real working conditions.
The researchers also tracked where people were looking. Video participants spent substantially more time looking directly at their partner's face on the screen and less time letting their gaze wander around the room. That narrowed visual attention is the mechanism they propose for the narrowed idea generation.
It is worth being precise about the size of the claim. This is a well-powered study with both lab and field components, which is rare and makes it more credible than most workplace research. It is also a study of pairs performing short idea-generation tasks, not of quarterly innovation output in distributed companies. Treat it as strong evidence about one specific activity.
What the finding does not say
It does not say remote teams are less creative. It says a particular ninety-second-to-five-minute activity, performed live over video, yields fewer ideas than the same activity performed live in a room. Most of what a distributed team does is not that activity.
It does not say video calls are bad for decisions. The selection result points the other way, and there is a plausible reason: a narrowed focus is exactly what you want when the job is to compare four options against criteria rather than to reach sideways for a fifth.
And it does not say that in-person brainstorms are good. Group brainstorming in a room has its own long-documented problem — production blocking, where only one person can speak at a time and the others lose their half-formed ideas while waiting. In-person generation beats video generation in this study. Neither beats silent individual generation followed by group evaluation.
Why the screen narrows what you come up with
The visual-focus explanation is intuitive once you have watched yourself on a call. The video window is a small, bright, high-value rectangle containing a face that appears to be looking at you. Your eyes stay there. Your peripheral field, the ceiling, the window, the mess on the desk, all the incidental input that seeds unexpected associations, quietly stops arriving.
There is a second, less discussed cost: the social monitoring load of being on camera. You are watching your own face, checking whether you look engaged, timing your turn to speak so you do not talk over anyone. That is working memory spent on presentation rather than on the problem.
Both explanations point at the same intervention. Take the eyes off the grid and take the performance requirement away for the part of the session where people are supposed to be thinking loosely.
✗ The default remote brainstorm
- •Sixty minutes, everyone on camera the whole time
- •Facilitator asks for ideas and waits for volunteers
- •One person talks at a time; the rest hold their idea or lose it
- •The first three ideas set the direction for the remaining fifty minutes
- •Ends with a shared document that mostly records the loudest two people
✓ The same hour, re-sequenced
- •Ten minutes solo and off-camera, everyone writing their own list
- •Submissions collected privately so nobody sees the room's answers first
- •Fifteen minutes on camera clustering and questioning what came in
- •Twenty minutes evaluating against criteria, where video does no harm
- •Ends with a ranked set and a visible count of distinct ideas per person
A format that keeps divergence intact
Send the prompt before the session (24 hours ahead)
One question, no more than two sentences, with no example answers attached. Examples anchor the range of what people bring. Ask for a minimum count — five ideas each — because a floor produces far more than an open invitation does.
Run the generation phase off-camera (8-10 minutes)
If it has to happen live, tell everyone to turn video off, look away from the screen, and write on paper or in a private document. Say explicitly that silence is the activity, or someone will fill it within ninety seconds.
Collect privately, reveal all at once (2 minutes)
Use a word cloud or an open-text activity where responses stay hidden until the round closes. Nobody should be able to read the first submissions while still writing their own, or the round drifts towards whatever landed first.
Cluster on camera (12-15 minutes)
Now video earns its place. Group the submissions live, ask the author of anything ambiguous to explain it in one sentence, and merge duplicates. Keep the low-vote outliers visible rather than deleting them at this stage.
Select with structure (15 minutes)
Score against two or three named criteria, or run a hidden-tally vote. This is the phase the research suggests video handles perfectly well, so use the time you saved earlier to do it properly.
Close with the count (3 minutes)
Show how many distinct ideas the group produced and how many people contributed at least one. Those two numbers are your baseline for judging whether the next format change helped.
When video is genuinely the wrong room
There are sessions where the generation phase is the whole point and the stakes justify the travel: an early-stage product concept sprint, a rebrand, a strategy reset where nobody yet knows what the options are. If you get people in a room twice a year, spend that time on divergence and do the evaluation remotely afterwards.
The inverse is more common and more wasteful. Teams fly in to sit in a room and review a deck, which is precisely the task video does not degrade. If the agenda for your offsite is mostly presentation and decision, the offsite is expensive theatre.
For distributed teams that cannot meet at all, the asynchronous version works better than most people expect. Give the prompt a working week, collect written submissions throughout, and use the live call only for clustering and choosing. You get more ideas per person than a live call produces, and time zones stop being a constraint.
✓ Pros
- •Selection, comparison and structured evaluation hold up fine over video
- •Written submission removes production blocking, which affects in-person groups too
- •Anonymous collection surfaces ideas that juniors will not say out loud on camera
- •Asynchronous generation gives people their own best thinking hours
- •You can run more, shorter sessions without travel cost
✗ Cons
- •Live on-camera idea generation produces fewer creative ideas than the same task in a room
- •The video grid narrows visual attention and adds self-monitoring load
- •Building on someone else's half-formed idea is harder without shared physical space
- •Long video sessions accumulate fatigue faster than equivalent in-person time
- •Silence on a call feels like failure, so facilitators cut thinking time short
Practical adjustments that take no extra time
Top Tips
- Give the generation prompt before the meeting rather than during it. The single cheapest change, and it converts dead call time into private thinking time.
- Ask for a specific number of ideas per person. A floor of five produces roughly double what "share anything that comes to mind" produces from the same group.
- Let people turn cameras off during silent work and say so explicitly, so nobody worries it looks disengaged.
- Keep submissions hidden until the round closes so early answers do not narrow the later ones.
- Run the wildest-idea round separately and last. Mixed into a normal round, wild ideas get self-censored; given their own ninety seconds and a clear licence, they arrive.
- Never evaluate in the same breath as you generate. The moment someone says "that would never get approved", the generation phase is over regardless of what the agenda says.
Measuring it on your own team
You can test this in two sessions without any instrumentation. Run your next idea session in the usual live format and count two things: distinct ideas after de-duplication, and the number of participants who contributed at least one. Then run the following session with a pre-sent prompt and silent private submission, and count the same two numbers.
Count distinct ideas rather than total submissions, because the live format inflates volume with restatements of whatever was said first. The de-duplicated count is the honest measure and it is usually where the gap shows up.
Track the participation number just as carefully. A format that produces the same idea count from twice as many contributors has improved something real, even if the headline total looks flat, because it means you are no longer running a session that only three people are actually in.
Count after de-duplication
Merge restatements before you compare formats. Live rounds inflate totals with rephrased versions of the first idea.
Watch the contributor count
Ideas per session matters less than how many people put at least one forward. Three-person sessions disguised as twelve-person sessions are the common failure.
Time-stamp the submissions
If most responses arrive in the first thirty seconds, people are reporting cached opinions rather than generating. Extend the silent phase.
Keep the outliers
The idea nobody voted for is the one worth revisiting in a fortnight. Archive the full list in the recap rather than only the winners.
How confident should you be in this?
More than usual for a workplace claim. The combination of a controlled laboratory study and a multi-country field experiment addresses the standard objection that lab findings do not survive contact with real work. The effect appeared in both.
The honest caveats are about scope rather than quality. The task was short-form idea generation in pairs. Whether the same mechanism holds for a six-person group over three hours, or for the slow kind of creativity that happens between meetings, is not something this study can tell you.
The good news is that the recommended response is worth doing regardless of who is right. Silent individual generation before group discussion outperforms live group generation in person as well, for entirely separate reasons. You are not making a bet on one paper — you are adopting a format that wins under both explanations.
What to change in your next session
Take your next remote idea session and move exactly one thing: the generation phase. Send the prompt a day early, open the call with collection rather than discussion, and keep responses hidden until everyone has submitted. Everything else on your agenda can stay where it is.
In Session Flo, that is an open-text or word cloud activity with results hidden until the round closes, followed by a ranking or voting activity once the ideas are on the board. The recap afterwards gives people the full list, including the ideas that did not win, which is where the next session usually starts.
The measure of success is not a better-feeling meeting. It is a longer, stranger list of distinct ideas from more of the people who were in the room.
Frequently asked questions
Does this mean we should stop running remote workshops?
No. The evidence points at one phase, not the format. Selection, evaluation, planning and decision-making show no penalty over video, and those make up most of a typical workshop. Move idea generation to silent, private, off-camera time and the rest of your remote agenda is fine as it is.
Is audio-only better than video for creative work?
The study compared video with in-person rather than with audio, so there is no direct answer in the evidence. The proposed mechanism — a screen holding your visual attention — suggests that not staring at a grid of faces should help, which is why the practical advice is to let people look away and work on paper during generation rather than to ban video outright.
How long should the silent generation phase be?
Eight to ten minutes for a substantive prompt, three to four for a quick divergent round. Shorter than three minutes and you only collect the cached opinions people arrived with. Longer than twelve and output per minute drops sharply, so it is better to run two separate prompts than one very long round.
What about hybrid sessions with some people in the room?
Hybrid is the hardest case, because the in-room group will start generating out loud and the remote participants will be listening rather than thinking. Run the generation phase identically for everyone: everybody silent, everybody submitting through the same tool, nobody speaking. The in-room advantage disappears the moment half the room is talking.
Does anonymity change idea quality?
It changes who contributes, which usually changes the range. Anonymous submission reliably surfaces input from people who will not volunteer an unformed idea in front of their manager. Attach names later during clustering if you need to ask an author what they meant, and be transparent about that up front.
Keep reading
- why silent brainstorming beats shouting ideas out — The generation format this research points towards, in step-by-step form.
- facilitating a brainstorm that produces real ideas — The full session structure around the generation phase.
- choosing between synchronous and asynchronous workshops — When to move generation out of the call entirely.
- an honest comparison of in-person and hybrid meetings — Which activities actually justify getting people in a room.
- run a hidden-results idea round in Session Flo — Open text and word cloud activities with responses hidden until the round closes.