psychological safety and Google's Project Aristotle: what the research says about teams that speak up
A hospital study that seemed to show the best nursing units making the most mistakes. A factory study of 51 teams. A Google project on 180 of its own teams. Together they made "psychological safety" one of the most quoted ideas in management. The core idea is well supported. Some of what gets bundled with it is not.
By Brightside English editorial team
Reviewed by a Brightside senior coach · · 5 min read
Yes, Google's internal research put psychological safety first among five team dynamics. But the idea came from Amy Edmondson's peer-reviewed work in the 1990s, and Google's study was internal and not published in a journal. Claims about equal turn-taking and "social sensitivity" come from a separate 2010 study on group intelligence, whose central idea is still debated.
The study that started it: errors on hospital wards
Amy Edmondson, an organisational researcher, studied eight units in a university hospital. She surveyed staff about how each unit was led and how people worked together, and an observer who did not know her hypotheses watched the nursing teams at work.
The result looked backwards. Units with stronger direction and coaching from the nurse manager, and better relationships, recorded significantly more medication errors, not fewer. Edmondson's interpretation, published in 1996 in the Journal of Applied Behavioral Science, was that a detected error rate measures two things: how many errors happen, and how willing people are to report them. In the better units, people were more willing to say "I made a mistake." In the others, mistakes were more likely to stay hidden.
It is worth being clear that this was an interpretation of a small study, not a proof. But it pointed to a question that shaped the next decade of her work.
The study: 51 teams in a factory
In 1999, Edmondson published Psychological Safety and Learning Behavior in Work Teams in Administrative Science Quarterly. She defined team psychological safety as a shared belief that the team is safe for interpersonal risk taking: asking a question, admitting an error, raising a concern, without fear of being embarrassed or punished.
She studied 51 work teams in a manufacturing company, combining several research methods. Teams with higher psychological safety showed more learning behaviour, the everyday work of noticing problems and adjusting. That learning behaviour, in turn, was linked to team performance. A team's confidence in its own ability (team efficacy) did not predict learning once psychological safety was taken into account. Structural factors, such as coaching from the team leader and support from the organisation, also mattered.
Project Aristotle: what Google reported
Researchers at Google later studied 180 of its own teams: 115 engineering project teams and 65 sales teams, ranging from three to fifty people. According to Google's re:Work guide, first published in 2015, the researchers held confidential conversations with leaders, analysed more than 250 items from engagement surveys and ran over 35 statistical models.
They named five dynamics that set effective teams apart, in this order: psychological safety, dependability, structure and clarity, meaning, and impact. Psychological safety was the strongest. The guide credits Edmondson for the concept and uses her definition. It concludes that what mattered was "less about who is on the team" and more about how the team worked together. Several things did not significantly predict effectiveness at Google, including whether people sat together, team size, seniority, tenure and how extroverted members were.
Did it hold up?
Edmondson's idea: yes, broadly. Psychological safety has since become one of the most studied ideas in team research. A 2017 meta-analysis in Personnel Psychology by Frazier and colleagues pooled 136 independent samples, covering more than 22,000 people and nearly 5,000 groups, and examined what builds psychological safety and how it relates to outcomes such as task performance. That is a large and varied base of evidence. Most of it is correlational, though: safe teams and good teams tend to go together, which is not the same as proving that one causes the other.
Project Aristotle: treat with care. Google published its findings as a practical guide on its own website, not as a peer-reviewed paper. The data came from one company with an unusual workforce, and outside researchers cannot check the analysis. The guide itself says that factors which did not matter at Google may matter elsewhere. It is a useful case study that agrees with the academic research, not independent proof of it.
The turn-taking claim: genuinely debated. Many popular accounts of Project Aristotle add that the best teams share speaking time equally and are good at reading each other's feelings. Those points come from a different study: Anita Woolley and colleagues, writing in Science in 2010, tested 699 people in groups of two to five. They reported a general "collective intelligence" factor, which they called c, that predicted group performance across many tasks. It was linked to members' average social sensitivity, to how evenly people took turns in conversation, and to the proportion of women in the group, and only weakly to members' individual intelligence.
That finding has been challenged. Bates and Gupta, reporting three studies in 2017, found that members' individual IQ accounted for most of the differences in group IQ. Credé and Howardson also argued in 2017, after reanalysing existing data, that the original analysis overstated the case. A 2021 meta-analysis by Rowe, Hattie and Hester found only a moderate link between c and group performance across 857 groups, with results varying widely between studies, and advised against embracing the idea too early. On the other side, Woolley and colleagues pooled 22 studies covering 5,279 people in 2021 and again found a robust collective intelligence factor, predicted by members' social perceptiveness and by how well the group collaborated. That analysis did not, however, test equal turn-taking directly.
So: social perceptiveness has decent support. Equal airtime as a key to team intelligence is an open question.
How it gets misused
- "Psychological safety means being nice." Edmondson's teams were safe for risk taking, including disagreement and bad news. A team where nobody challenges anything is comfortable, not safe in her sense.
- "Google proved it." Google's guide agrees with earlier peer-reviewed research. It did not settle the question.
- Running a stopwatch on who speaks. Turn-taking may help, but the evidence is mixed. Forcing equal minutes can feel staged and misses the point: people need to believe they can speak.
What it means for you
If you lead a team, the research suggests your words carry extra weight. Google's guide, drawing on Edmondson, offers three habits: frame the work as a learning problem, admit your own fallibility, and model curiosity by asking questions. These are communication habits, and you can practise them.
Frame the work as uncertain, so that raising problems is part of the job:
"We have not done a launch like this before. I need you to tell me early when something looks off."
Go first with a mistake:
"I got the forecast wrong last quarter. Here is what I missed. What else should I be watching?"
Respond to bad news in a way that makes the next piece of bad news more likely to reach you:
"Thank you for flagging this now. What do you need from me?"
This applies to a site lead in Pune with a construction crew, a ward sister in Doha, a founder in Kochi with six engineers, and a parent at the dinner table. People watch what happens to the first person who speaks up.
Sources
- Amy C. Edmondson, Learning from Mistakes is Easier Said Than Done: Group and Organizational Influences on the Detection and Correction of Human Error, Journal of Applied Behavioral Science 32(1), 1996, summarised by AHRQ PSNet. psnet.ahrq.gov/issue/learning-mistakes-easier-said-done-group-and-organizational-influences-detection-and
- Amy Edmondson, Psychological Safety and Learning Behavior in Work Teams, Administrative Science Quarterly 44(2), 350, 1999, record at Harvard DASH. dash.harvard.edu/entities/publication/13a7b031-0fdd-45ec-a7e0-2b80e2bc679f
- Google re:Work, Guide: Understand team effectiveness (Project Aristotle), first published 2015. rework.withgoogle.com/intl/en/guides/understand-team-effectiveness
- M. L. Frazier, S. Fainshmidt, R. L. Klinger, A. Pezeshkan and V. Vracheva, Psychological Safety: A Meta-Analytic Review and Extension, Personnel Psychology 70(1), 113-165, 2017, record at Old Dominion University. digitalcommons.odu.edu/management_fac_pubs/13/
- A. W. Woolley, C. F. Chabris, A. Pentland, N. Hashmi and T. W. Malone, Evidence for a Collective Intelligence Factor in the Performance of Human Groups, Science 330(6004), 686-688, 2010. doi.org/10.1126/science.1193147
- L. I. Rowe, J. Hattie and R. Hester, g versus c: comparing individual and collective intelligence across two meta-analyses, Cognitive Research: Principles and Implications 6, 26, 2021 (discusses Bates and Gupta 2017 and Credé and Howardson 2017). pmc.ncbi.nlm.nih.gov/articles/PMC8019454/
- Christoph Riedl, Young Ji Kim, Pranav Gupta, Thomas W. Malone and Anita Williams Woolley, Quantifying collective intelligence in human groups, PNAS 118(21), 2021. pmc.ncbi.nlm.nih.gov/articles/PMC8166150/
Published by The Full Stop, the journal of Brightside English, in the The Experiments series. Every factual claim links to a source below; where accounts or findings are disputed, we say so. Sources checked on 26 Oct 2026. Reviewed by a Brightside senior coach. Corrections: see our editorial standard.