Psychological safety shifted from a research finding to a slide sometime between 1999 and now.
The original work was specific. Amy Edmondson defined it as the belief that team members can take interpersonal risks without facing punishment. She also created a seven-item tool to measure this concept. Google's Project Aristotle studied 180 teams to see what made the best ones succeed. They expected to find smart people or the perfect mix of skills. Instead, they discovered that psychological safety was the key factor.
Both of those are real. Neither survived contact with the corporate adoption cycle intact.
What remains is a phrase, a single line on the yearly engagement survey, and an offsite exercise. In this exercise, everyone shares something personal, then returns to silence.
I have run that offsite. I've spent a big part of my thirty-seven-year career reading the room. People often saw my insights as the room's condition, and that's what I want to discuss. Defining psychological safety is simple. Many organizations with more authority than I have already tackled it. It’s often more helpful to name what it is not. Many failures I’ve seen come from a specific substitution.
1. It is not comfort
This is the most common substitution and the most expensive.
Comfort is the absence of friction. Psychological safety is the presence of enough trust to generate friction on purpose. They point in opposite directions. A team where nobody disagrees is not safe. It is quiet. And from the front of the room, those two states are indistinguishable.
Edmondson has been explicit that the construct was never about niceness. It was about candor — whether the cost of saying the difficult thing is survivable. A team can be extraordinarily pleasant and extraordinarily unsafe. Most of the dysfunctional teams I have worked with were pleasant.
2. It is not something you can declare
"My door is always open" is not a policy; it is a description of a door.
We infer safety, but we do not announce it. People form their views based on evidence. They consider what happened to the last person who raised a tough issue. They think about whether the concern from March influenced any decisions later. They also look at whether the person who pushed back in the meeting received a promotion. The announcement does not touch any of that.
It does do one thing, though. It moves the burden. When a leader says the door is open, silence shifts from being a system issue to an employee's failure. That is worse than saying nothing.
What actually works is demonstration, and it is unglamorous. When I told my team I was autistic, other people started telling me things about themselves. Not because I had declared a safe space, but because I had gone first and nothing bad happened to me. The energy those people had been spending on masking became available for the work. No poster produces that.
Psychological safety isn't proclaimed; it's designed.
3. It is not evenly distributed
This is often overlooked. It's a measurement issue first, not a culture issue.
Psychological safety gets reported as a team-level score. Teams are not homogeneous. That score is an average. But averages aren't good for conditions that change a lot at the edges.
We now have data on this rather than just intuition. A 2024 study by Bahadurzada, Edmondson, and Kerrissey tracked health-care workers from 2019 through the pandemic. They found that psychological safety acted as a lasting resource. It helped people during tough times and predicted retention years later. The finding that matters here is the distribution. The protective effects were strongest for physicians, women, and people of color. Which is to say: strongest for the groups already carrying the most burnout.
Psychological safety is worth the most to the people who have the least of it. Those are also, structurally, not your median survey respondent. A team scores 4.2 out of 5. One member knows that raising costs more than it returns.
This is the general shape of what I call the 25/25 problem. Roughly a quarter to a third of your employees are highly perceptive. They spot problems before they surface in the data. Meanwhile, a meaningful share of leadership operates from control rather than collaboration. The exact percentages matter less than the collision, which is not occasional. It is structural and it recurs.
Put it another way. Part of your building is on fire and 25 percent of your employees are smoke detectors. When the smoke detectors sound, management says they're too sensitive to fire. They suggest trying to be less flammable. The smoke detectors search for buildings that aren’t burning. Meanwhile, management worries about retention, all while standing in a building on fire and holding matches.
If you want to know where your own team actually sits, the Perception Advantage Quiz takes about ten minutes and doesn't ask your manager.
4. It is not the absence of complaints
Low complaint volume can mean two things: everything is fine, or the reporting system is seen as a trap.
These produce identical data. If you measure with complaint volume, you can't spot a healthy team. A team that has completed its learning looks better on the dashboard. The metric improves as the condition worsens.
I know how this works because I have been on the wrong end of it. Years ago I sat in weekly meetings with consultants running an ERP implementation. My body was sending warnings. The lead consultant’s voice rose slightly. The junior consultants avoided eye contact. Our IT staff struggled to stay calm. I had never implemented an ERP system. These people billed more per hour than most people make in a day. So I said nothing, week after week, and wondered whether I was being too sensitive.
The system went live. Sales teams went to paper orders. Customers couldn't get product. Every scenario my gut had flagged in the first meeting arrived on schedule. The consultants charged millions for a disaster I spotted for free. They didn’t even mention it.
That project generated zero complaints. It was, by the dashboard, a well-run engagement.
A psychologically unsafe workplace has signs that often go unnoticed. Here are some examples that don't appear in complaint logs:
-
People check who is on the invite before deciding what to say
-
Concerns arrive as questions rather than statements, so they can be withdrawn
-
The real talk happens in the fifteen minutes after the meeting, with a smaller group.
-
Bad news travels upward slowly and arrives pre-softened
-
Someone has a reputation for being difficult and nobody can name a specific incident
-
New hires stop raising things somewhere around month four
There's a relabeling pipeline to recognize. It's very consistent and worth learning. People who can't be intimidated become "difficult." People with long memories become "negative." People who build genuine bonds across the org become "not a team player." People who accurately read that something is off become "too sensitive."
Every one of those looks like a functioning team from above. That is not a coincidence. That is the design.
5. It is not measurable by the person holding the power
Here is the mechanism underneath the other four.
Joe Magee and Adam Galinsky studied power distance. They found that having power affects how you see others. Subordinates feel more distant and less real. This change lowers your ability to empathize and see things from their perspective. This is not a character flaw in particular executives. It is what the position does to perception, reliably, to most people who occupy it.
Now consider how psychological safety usually gets assessed. A leader forms an impression of whether their team feels safe. They create it from a safe place where talking feels low-risk. Their view is shaped by a system that power has already distorted, making it less accurate regarding the specific people involved. They notice they feel fine. They notice the meeting went well. They conclude the team is fine.
These assessments usually match the views of the person who ordered them. When the tool agrees with the person in charge of the budget, it’s not measuring what it should.
This is also why communication workshops don't fix it. Power distance comes from one's position in a structure. To change it, we need to change the structure. You cannot train your way out of a geometry problem.
What it is
Without the workshop, each team member constantly evaluates psychological safety. This thought often runs below their awareness. What does it cost me to say this. What happened last time. What happens to people like me here.
You do not change the answer by telling people the answer is different. Change it by altering what happens when someone speaks the hard truth. Do this visibly and often, focusing on those who need to hear it most.
If you want a number to manage against, use Edmondson's original seven-item instrument. It was published in 1999 and has been tested across thousands of teams. I use it as is. When something is that well-validated, you don’t need to change it. Run it quarterly. Weight it in leadership evaluation at something that hurts to miss.
MIT Sloan's research shows that a toxic culture predicts employee turnover 10.4 times better than pay. That is the cost of getting this wrong, and it does not show up as a line item anywhere.
The rest is slower than an offsite. It is also the only version that survives the next reorganization.
Sources
- Amy C. Edmondson, The Fearless Organization (Wiley, 2018)
- Charles Duhigg, "What Google Learned from Its Quest to Build the Perfect Team," NYT Magazine, Feb 25, 2016
- Bahadurzada, Edmondson & Kerrissey, "Psychological Safety as an Enduring Resource Amid Constraints," International Journal of Public Health 69 (2024)
- Magee & Galinsky, on social hierarchy and power distance
- Sull et al., "Why Every Leader Needs to Worry About Toxic Culture," MIT Sloan Management Review, Mar 16, 2022
Find out what your team isn't telling you
The Perception Advantage quiz measures the thing engagement surveys average out. Free, ten minutes, no login.
Take the quizOr read the book this came from: The Perception Revolution.