Not harsh to the AI.
Harsh to the people around you.
And you do not notice it happening.
A peer-reviewed study published in PNAS Nexus, one of the most rigorous scientific journals in the world, just proved that spending time with an AI chatbot changes how you judge other humans. Harshly. Measurably. Automatically.
This is not a theory about what AI might do to workplace culture someday. This is a measurement of what it is already doing today, in organizations that have no idea it is happening.
The Study
The paper is called “People Judge Others More Harshly After Talking to Bots.” Written by researchers from the University of Pennsylvania, the University of Hong Kong, and the University of Florida. Two preregistered experiments. 1,261 participants total.
Here is exactly how it worked.
Participants were paired with a partner to complete a creative task. Half were told their partner was human. Half were told it was an AI.
Then both groups were asked to evaluate the work of a third person, a purported human.
The result, participants who had just worked with an AI rated the human’s work significantly lower than participants who had just worked with another human.
The people who had been in AI interaction mode evaluated human work more harshly. Statistically significant. Replicated in a second study.
Then the researchers tested whether this was just about standards, maybe participants graded more strictly because they wanted to be consistent across interactions.
They ran Study 2 with a critical twist, participants were told their evaluation would never be shared with the person being evaluated. The harsh judgment could not possibly be about signaling fairness or maintaining standards.
Study 2 replicated the effect anyway. The harshness was not strategic. It was not conscious. It was automatic, a behavioral residue of the AI interaction that bled directly into the next human encounter, even when it had no social function.
What the Language Analysis Revealed
The researchers did not stop at behavior. They analyzed the language people used while working with their AI partner versus their human partner.
The pattern was consistent, people working with AI were more demanding, more instrumental, and displayed less positive affect than people working with human partners.
This makes intuitive sense once you see it. When you interact with AI, you do not soften your requests. You do not check in. You do not say thank you out of genuine gratitude. You do not read the room. You task. You prompt. You demand and refine and prompt again.
That mode, the AI interaction mode, is efficient. It is also the structural opposite of how high-trust human relationships are built and maintained.
And the study shows it does not stay contained to the AI interaction. It bleeds.
Count the AI Interactions in a Typical Workday
Think about how many AI interactions happen before noon in 2026.
ChatGPT for a research question. Claude for a document draft. Copilot for code review. An AI scheduling assistant for a meeting. A customer service chatbot for a vendor issue. Another prompt. Another refinement. Another instrumental demand with no social warmth required.
Each one is training you, subtly and cumulatively, to be more demanding and less charitable.
And then a colleague sends you their work for feedback.
The researchers called this a “potentially worrisome side effect of the exponential rise in human-AI interactions.” Not worrisome for AI, AI has no feelings to hurt. Worrisome for us. For the humans on the receiving end of judgments shaped by hours of transactional AI interaction.
Why This Is a Leadership Problem
Most organizations adopting AI are focused on productivity metrics. How much faster? How many tasks automated? What is the ROI?
Those are legitimate questions. But they are measuring the output side of the ledger while ignoring the cultural side entirely.
If your team is spending increasing hours each day in AI interaction mode, and the data says they are, and that mode makes them measurably harsher in their next human interaction, and the data says it does, then the adoption of AI tools is producing a cultural side effect that no one is measuring, no one is managing, and almost no one is aware of.
The trust and collaboration erosion that follows will not show up in your productivity dashboard. It will show up in engagement scores, in turnover data, in the quality of feedback conversations, in the health of the team relationships that make execution possible.
By the time it is visible, it will already be expensive to fix.
This Has an Even More Personal Dimension
The researchers identified one implication that extends beyond the workplace entirely.
Consider a parent who spends the workday in back-to-back AI interactions, prompting, demanding, refining, evaluating with the transactional precision that AI tools reward. Then they come home.
Their child shows them a drawing. A homework assignment. A project they worked on.
The parent evaluating that work is not in neutral. They are carrying the behavioral residue of hours of AI interaction mode. The study’s findings suggest they may be measurably harsher in that evaluation than they would have been on a day without the AI interactions.
This is one of the most important implications in my forthcoming book Raise Them Distinct, the idea that how we interact with AI at work does not stay at work. It shapes the human beings we bring home.
What Awareness and Discipline Look Like
The study does not suggest stopping AI use. That is neither realistic nor the right conclusion.
What it suggests is that AI adoption without behavioral awareness is incomplete. The organizations and individuals who navigate this well are the ones who recognize the mode shift, who notice when they have been in AI interaction mode for hours and consciously shift back before the next human conversation.
That is not a soft skill. That is a leadership capability. And in 2026, it is one that almost no organization is deliberately building.
Three practical starting points:
Name the mode. Give your team language for “AI interaction mode” versus “human interaction mode.” The simple act of naming it creates awareness that reduces the automatic bleed.
Build a transition. Before a feedback conversation, a one-on-one, or any high-stakes human interaction, create a brief intentional pause after AI work. Even 90 seconds of reorientation changes the starting posture.
Measure what matters. If you are measuring AI productivity gains but not measuring trust, collaboration quality, or feedback effectiveness, you are getting an incomplete picture of what AI adoption is actually doing to your organization.
The Question for Every Leader
The AI tools your team is using are not neutral. They are shaping the mode your people bring to every subsequent human interaction.
The question is not whether this is happening. The study says it is.
The question is whether you are building the awareness and the discipline to know when you are in AI mode, and to switch back before the next human conversation.
Those prepared need not fear the forces at work.
Distinct or Extinct is available now on Amazon. Download Chapter 1 free at realmikeevans.com
Take the Kryptonite Scorecard at realmikeevans.com/scorecard, does your organization have awareness of this effect?
Source: Tey, Kian Siong; Mazar, Asaf; Tomaino, Geoff; Duckworth, Angela L.; Ungar, Lyle H. “People judge others more harshly after talking to bots.” PNAS Nexus, September 2024. DOI: 10.1093/pnasnexus/pgae397.