The Empathy Outsource

What Happens When the Care Was Real but the Author Was Not

Issue 270, June 25, 2026

A manager at a client organization told me about a message she received from her director on the morning after her father’s funeral. It was warm, specific, and unexpectedly moving. It acknowledged her loss, named the particular strain of returning to work too soon, and told her to take whatever time she needed without guilt. She read it twice. And then, in the way that the mind sometimes betrays us, she noticed something in the cadence of the third sentence, a smoothness, a balance, a certain rounded perfection that did not match how her director actually spoke. She could not prove it. But the suspicion arrived and would not leave: this message, the one that had just comforted her, had probably been written by a machine. She told me that the comfort did not just fade. It inverted. The same words that had made her feel seen now made her feel like she was processed. And she could not decide which was worse, that her director hadn’t written it, or that she would now never be sure.

This is the dynamic I want to dive into because it is becoming one of the most quietly consequential developments in how human beings relate to one another inside organizations (and perhaps even personally in day-to-day life), and almost no one is talking about it openly and honestly or taking a critical thinking moment to consider that what they may have just received was written by something other than a human.

We have entered an era in which everyone can delegate the language of care to AI. In an organizational context, that means our co-workers, our immediate supervisors, and yes, the leaders of an organization. The recognition note, the performance feedback, the condolence message, the difficult conversation prepared in advance, all of it can now be drafted, in seconds, by a system that produces warmth on demand. Either via a chat interface, an agent, or via the ever-growing use of an “AI-Twin”. The question is not whether the output is good. The unsettling research finding is that the output is often better than what we would have written ourselves, more thoughtful, clearer, warmer, and seemingly more personal. The question, however, is what happens to the recipient, and to the relationship when the care was real but the author was not. Let me let that land with you for a minute because, as I mentioned, I don’t think it is something we have yet pondered.

I am calling this The Empathy Outsource, and the human costs to this new development and behavior operate in a place that no productivity metric will ever capture: emotions, thoughts, self-conceptions, and perceptions of the humans who are in receipt.

The Newsletter’s Continuing Thread

A note for readers who have followed the last several issues of this newsletter. We have spent recent weeks inside the human dynamics that AI is surfacing in organizations, from the shadow adoption examined in The Permission Paradox to the cognitive depletion driving people to these tools in the first place. This issue continues that arc, and I want to restate what I said a few weeks ago, because it matters even more here. This is not an AI newsletter, and Ideas and Innovations has not become one. The subject has always been the human factor. What AI does is hold up a mirror to behaviors and needs that were present all along, as we now have case study after case study and situation after situation bringing elements of our very humanness forward that we likely have been dismissing or overlooking for some time.

The Empathy Outsource is one of the clearest examples because the thing under examination is not artificial intelligence at all. It is the oldest currency in human relationships, the sense that another person took the time to care, to write a note, send an email, or give a card, and what happens to that currency when its source becomes uncertain.

We as humans know the people around us, whether at work or in our personal lives. We learn how they communicate: the words they use, the information they bring forward, and even the tone, facial expressions, and body movements that encompass their communications and interactions. This knowing is perhaps one of the best “tells” we have to assess whether what we have received is written by someone we know or by a machine. There is a counterpoint worth acknowledging here. Most of us do not write the way we speak, and I have heard this observation about my own communication, where my writing carries more formality and complexity than my day-to-day speech. That gap between spoken and written voice adds another layer to the challenge of discerning whether what we have received was authored by a person or by a machine, and it is a layer that AI’s increasingly fluent output is making harder to navigate by the day.

The Research Is Strikingly Consistent, and Strikingly Counterintuitive

The most important thing to understand about AI-generated empathy is that it works, right up until the moment the recipient knows where it came from (if they are even able to determine where it came from). This isn’t speculation. It is one of the most consistent findings in the emerging research on human-AI interaction that is showing how easily manipulated we can be, particularly in situations where we are vulnerable.

A study published in the Proceedings of the National Academy of Sciences in 2024 examined whether AI could perform the deeply human function of making someone feel heard. The researchers found that AI-generated messages actually made recipients feel more heard than human-generated messages did, and that AI was more effective at detecting the emotions in what people shared. Then came the reversal. When recipients learned that a message had come from AI rather than a human, the feeling of being heard dropped sharply. The exact same capacity that made the message effective became the thing that hollowed it out, the instant its origin was known.

This pattern has now been confirmed across a remarkable body of research work. In a study published in Nature Human Behaviour in 2025, a research team, including Amit Goldenberg of Harvard Business School, conducted nine studies involving more than 6,000 participants across multiple countries to test what they came to call the human empathy premium. Their finding was more than blunt. People consistently rate identical messages as less empathic the moment they believe AI produced them. The researchers named the inverse effect the AI Penalty: label a message as AI-generated, accurately or not, and the recipient downgrades its value immediately. The content of the message doesn’t change. The words are identical. Only the perceived author changes, and that alone is enough to collapse the message’s emotional worth.

The studies found something even more specific. When AI was tuned to deliver cognitive empathy, the simple acknowledgment of someone’s situation, people rated it about the same whether they thought it came from a human or a machine. A really interesting finding that shows that often words, regardless of where they came from, do indeed influence us.

But when the message was tuned for affective empathy, the sense that the sender genuinely shares your feelings, people strongly preferred the human source. We will accept a machine that understands our circumstances. We resist, at a deep level, the idea that a machine could care. Perhaps revealing some deeply programmed evolutionary filter that recognizes real human interaction that can, in many ways, run counter to how we humanize material items and our pets.

A 2025 study published in Frontiers in Psychology sharpened the stakes further by testing emotionally charged scenarios, the birth of a child, and the diagnosis of a terminal illness. The researchers manipulated who recipients believed had authored a message, ranging from a close friend to ChatGPT. The result was that attribution to an artificial source produced the steepest declines in perceived authenticity and moral respect, and that anthropomorphic cues, the design tricks that make AI feel more human, temporarily elevated the evaluation but collapsed entirely upon disclosure, particularly in the highest-stakes contexts. The more the moment mattered, the more the artificial source cost.

Why the Brain Reacts This Way

It would be easy to dismiss all of this as irrational prejudice, a bias against machines that will fade as the technology improves and people grow accustomed to it. I have already shared that many across the Generation Z population are interacting with AI on a personal level daily, and who consider the AI a friend and counselor.

But, I don’t think the research to date supports that weirdly comfortable conclusion, and a theoretical paper published in Trends in Cognitive Sciences in early 2026 explains why the reaction may be neither irrational nor temporary.

The argument is that empathy, in humans, is not primarily about the words. It is a predictive signal. I have dove into this topic time and time again via this newsletter and in my presentations to help others understand communication theory and technological determinism (and its various interpretations).

When someone extends genuine empathy to us, our minds register far more than the content of the message. We register evidence about that person’s future commitment to us, their likely cooperation, and their place in our social world. An empathic message from a colleague updates what the researcher calls our mental social map. It moves that person closer, strengthens the connection, and raises our prediction that they will show up for us again. This is what empathy is for, evolutionarily. It is a costly signal of relational investment, and we are exquisitely tuned to read it.

Now consider what happens when the message comes from AI or is suspected to. The warmth is still there in the words. But the signal underneath, the evidence of another person’s investment, is gone. A machine generated the care at near-zero cost, which means it tells us nothing about anyone’s commitment to us. The social map doesn’t update because there is no relational information to update it with. This is why the suspicion alone is so corrosive and leads to many “OMG” moments when the recognition dawns. The manager who received the condolence note did not need proof that her director used AI. The mere possibility severed the message from its function as a signal of investment, and once that link is cut, the words become just words, however beautiful. The same paper notes the darker implication. If empathic language can now be produced at almost no cost, people can learn to deploy it to appear caring without actually committing anything, effectively hacking the appearance of social closeness. A very scary thought in a world of humans craving human collaboration and connection. Our deep sensitivity to AI-authored empathy may be the mind’s defense against exactly that exploitation.

The Workplace Is Already Living This

This is not a future problem. It’s very much here in our lives today. The outsourcing of relational language is already so widespread in organizations and the data on how employees respond should give every leader pause.

Major employers, including JP Morgan, Citi, and IBM, have introduced AI tools to assist with performance appraisals and feedback. The efficiency case is obvious. The trust case is far more troubling. A 2025 survey of UK employees found that only 28 percent would fully trust a manager if AI had contributed to their feedback, and 30 percent said that AI involvement would actively damage their trust in that manager. Nearly half, 49 percent, believed that AI-assisted recognition or praise simply lacks authenticity. The skepticism was highest among both the oldest and the youngest workers, which undercuts the common assumption that resistance to AI empathy is merely a generational holdover that will disappear with time. The youngest workers, the ones most fluent with these tools, were among the most resistant to receiving care through them. Which is furthered by related research showing that Gen-Z is uncomfortable using AI in a professional setting unless directed to do so. Ryan Vet and I dove into this interesting development in an episode of The Human Factor Podcast.

Researchers Anthony Coman of the University of Florida and Peter Cardon of the University of Southern California drew the line precisely in their 2025 work. Employees consider a manager’s use of AI writing assistance appropriate and professional for informational and routine communication, the meeting reminder, the scheduling note, the policy summary. But for relationship-oriented messages, the ones requiring empathy, praise, congratulations, motivation, or personal feedback, employees believe these are better handled with minimal technological intervention. That said, handled by a human, not by AI.

There is a boundary in the employee’s mind, and it falls exactly at the line between information and relationship. AI may carry the information. The relationship, people deeply insist, must be carried by a real person.

There is a genuine complication here that intellectual honesty requires me to name, because it cuts against the argument I am building. Some research points the other way. Work referenced by performance management platforms has found that AI-generated feedback can improve employee performance more than human feedback does, but, and this is the essential qualifier, only when the employee does not know the feedback was generated by AI. The moment it is disclosed, performance gains decline and trust in the feedback’s quality drops. This finding is real, and it deserves to be sat with rather than waved away, because it surfaces the most uncomfortable question in this entire domain. If AI-generated care produces better outcomes precisely when its origin is concealed, then the efficiency argument for the Empathy Outsource is also, inescapably, an argument for deception. The tool works best in the dark, keeping us humans in the dark on the source of what we receive. And a leadership practice that depends on the people it touches, never finding out how it works, is not a practice any leader should be comfortable building a culture on. These are mind-boggling considerations, particularly given how fast our adoption of AI is across the entire world of humans.

The Asymmetry Leaders Miss

The reason intelligent, well-meaning leaders walk into the Empathy Outsource is that the math looks compelling from where they sit. They are depleted, a condition we examined directly two weeks ago. They have forty people to recognize, a dozen difficult conversations pending, and a tool that produces warmer, more articulate, more consistent language than they could generate at the end of an exhausting week. From the sender’s side, the trade appears to be efficiency at no cost to quality. The output is often better. Where is the harm?

The harm lives entirely on the other side of the exchange, and it’s invisible to the sender. This is the asymmetry that the research makes completely unavoidable. The leader experiences AI-assisted empathy as a quality-neutral efficiency gain. The recipient, if they ever know or even suspect, experiences it as the discovery that their moment of being cared for was, at least in part, automated. The sender saves ten minutes. The recipient potentially loses something far harder to restore: the belief that when this person reached out to them in a moment that mattered, it was actually this person reaching out. Those two experiences are not symmetric, and because the cost is borne by someone who isn’t in the room when the decision to use the tool is made, it’s almost never weighed. We have measured the time these tools save with great enthusiasm in both reality and hype. We have measured what they cost the person on the receiving end with almost nothing at all.

And there is a compounding effect that makes this more dangerous than a single disappointing message. Trust, once a suspicion like this takes hold, it doesn’t stay contained to one note. The manager who suspects one condolence message was automated will reread every warm message that the director ever sent her through the same lens. Was the birthday recognition real? The praise after the difficult project? The reassurance during the reorganization? Let that sit for a few moments.

The Empathy Outsource doesn’t just devalue the message that triggers the suspicion. It retroactively contaminates the entire history of care between two people, and it poisons the well for everything that comes after. This is the precise opposite of what relational language is supposed to accomplish.

What This Asks of Leaders

I am not going to argue that leaders should never let AI near a sentence, because that position doesn’t survive contact with how work is actually happening in today’s terms, and it is not what the research recommends. The boundary that employees themselves draw is more useful and more honest than a blanket prohibition. Use the tools for the informational and the routine, where they genuinely help and where no one expects a piece of your heart to be in the scheduling reminder. But the relational core, the moments that exist precisely to signal that one human being is invested in another, must remain unambiguously human. Not because human writing is always better, the research is clear that it often isn’t, but because in these specific moments the authorship is the message. The point of a condolence note was never the elegance of its prose. It was the evidence it carried that you stopped, you thought about this person, and you spent something of yourself on them. AI can replicate the prose. It cannot replicate the spending, and the recipient’s mind is built to know the difference.

This leads somewhere genuinely difficult, and I want to end in the difficulty rather than escape it. The deepest problem with the Empathy Outsource isn’t really about disclosure or detection or which messages are appropriate to automate. It’s about what we are practicing when we delegate care. Recognition, feedback, and hard conversations are not just outputs to be delivered. They are the actual substance of relational leadership, the work through which a leader becomes the kind of person whose care means something.

A leader who routes that work through a machine may produce better messages in the short term while slowly losing the capacity and the standing to mean them. The muscle that isn’t used atrophies. We have spent this issue examining what the Empathy Outsource does to the people who receive automated care. The harder question, the one worth carrying out of this newsletter and into your week, is what it does to the leader who sends it?

The question then is not whether AI can write a warmer note than you can. It almost certainly can. The question is whether you still want to be the kind of leader whose people never have to wonder.

Related Reading

The Empathy Outsource: What Happens When the Care Was Real but the Author Was Not (Issue 270)

Directly Referenced in This Issue

The Permission Paradox: Why Your People Are Hiding the Very Things You Say You Want

Directly referenced for its examination of shadow adoption, where employees adopt AI tools privately because the organization has neither permitted nor prohibited them. The Empathy Outsource extends this dynamic into emotional territory, where the tools being hidden are not productivity aids but expressions of care.

The Readiness Illusion: Why the AI Agent Era’s Loudest Claims Outrun the Evidence

The preceding issue in the AI adoption arc, which argues that capability and readiness are not the same thing. The Empathy Outsource builds on this distinction by asking what happens when AI becomes capable of producing language that reads as empathetic but the organization is not ready to govern its use in emotionally significant communication.

The Truth About Transformation: Revised and Expanded

Kevin Novak’s book, referenced for its central argument that transformation succeeds or fails at the human level. The Empathy Outsource asks what happens to the human level when the language of care is delegated to a machine, and the people receiving it cannot tell the difference.

Ideas and Innovations Newsletter Issues

The Algorithmic Mirror: What AI Reveals About How We Actually Think and Decide

Explores how AI functions as a diagnostic lens for organizational behavior. The Empathy Outsource applies this same framing to emotional expression, arguing that AI-generated empathy reveals how selectively and inconsistently we have been offering care in its absence.

Artificial Understanding: The Intelligence We Built and the Comprehension We Didn’t

Examines the gap between AI capability and genuine human comprehension. The Empathy Outsource extends this into the emotional domain, where the gap between producing empathetic language and genuinely comprehending another person’s experience is the central tension.

The Busyness Trap: Why We Wear Exhaustion as a Status Symbol

The cognitive depletion narrative in The Empathy Outsource traces directly to this earlier issue. When professionals are too depleted to write a thoughtful note, send a considered email, or compose a genuine expression of care, AI becomes the pressure valve for emotional labor that the organization never acknowledged as labor at all.

The Mental Overload of Modern Leadership: Why Today’s Executives Are Burning Out Differently

The cognitive architecture research cited in The Empathy Outsource connects to this earlier examination of how leadership demands have outpaced human processing capacity. Emotional expression is one of the first capacities to be shed under cognitive load, which explains why leaders are among the earliest adopters of AI-generated empathy.

When Change Champions Burn Out: The Hidden Cost of Driving Change and Transformation

Explores the cognitive burden on individuals who carry transformation forward, the same population most likely to outsource emotional communication when their capacity for genuine care expression is exhausted by the demands of their role.

The Relationship Decay Rate: Why Professional Connections Atrophy Without Intention

Examines how professional relationships deteriorate without sustained investment. The Empathy Outsource raises the question of whether AI-generated care counts as genuine investment or whether it accelerates the very decay it appears to address by replacing intentional human attention with scalable but hollow language.

The Loyalty Trap: When Commitment Becomes a Cage

Examines how organizational loyalty can suppress authentic expression. The Empathy Outsource explores a parallel dynamic where professional norms around emotional communication create pressure to express care in forms that are expected rather than felt, making AI delegation a natural extension of performative workplace empathy.

The Authenticity Paradox in Transformation Leadership

Explores the tension between strategic authenticity and genuine vulnerability in leadership. The Empathy Outsource heightens this paradox by introducing AI as a third variable: when leaders use AI to produce empathetic language, the care may be strategically authentic but experientially indistinguishable from the real thing.

Structural Silence: Why Organizations Train People Not to Speak

Examines how organizations systematically teach employees when not to speak. The Empathy Outsource describes a related pattern where organizations have never taught employees that emotional expression is a professional competency, creating the void that AI adoption fills.

Beyond Demographics: The Psychology of Communication That Actually Drives Engagement

Explores the psychological mechanisms behind communication that genuinely connects with people. The Empathy Outsource interrogates what happens when those mechanisms are replicated by AI, and the recipient responds to the communication rather than the communicator.

The Generational Fault Line: Why Your Change Initiative Lands Five Different Ways

Examines how generational psychology shapes responses to organizational change. The Empathy Outsource cites research showing that Gen Z’s comfort with AI-generated emotional content, including romantic relationships with AI, represents a generational shift in how authenticity itself is defined.

Survival Mode Leadership: The Hidden Costs of Managing by Fear

Explores how fear-driven environments push people into survival mode. The Empathy Outsource argues that when emotional communication becomes another obligation in a fear-based environment, outsourcing it to AI is not laziness but a rational allocation of depleted cognitive resources.

Measuring What Matters Series

Measuring Trust as a Behavioral Asset: How to Operationalize the Most Important Intangible in Organizations  (Issue 266)

The Empathy Outsource identifies trust as the variable most at risk when empathetic communication is delegated to AI. This Measuring What Matters article provides the framework for operationalizing trust measurement, including the behavioral markers that would reveal whether AI-mediated care is building or eroding relational capital.

The Identity Problem in Measurement

Examines how professional identity shapes what organizations choose to measure. The Empathy Outsource reveals a measurement blind spot: organizations track communication volume and response time, but have no mechanism for measuring whether the care expressed in those communications was genuinely authored by a human being.

Transformation Psychology

Why AI Adoption Resistance in the Workplace Is a Leadership Problem, Not a Technology Problem

The foundational transformation psychology article argues that AI resistance is rooted in identity and trust rather than technology. The Empathy Outsource extends this by examining what happens when AI adoption is not resisted but embraced for the most psychologically intimate dimension of professional life: emotional expression.

The Communication Paradox in Transformation

Explores why more communication often produces less understanding. The Empathy Outsource reveals a new dimension of this paradox: AI enables more empathetic language at a greater scale, but if the comprehension behind that language is absent, the volume of care may increase while the depth of connection decreases.

Emotional Exhaustion in Change Management: Warning Signs and Solutions

Examines the warning signs and consequences of emotional exhaustion during organizational change. The Empathy Outsource argues that the outsourcing of emotional communication to AI is itself a warning sign of exhaustion that most organizations are not equipped to recognize.

The Partnership Posture: A Working Frame for Humans and AI

Offers a constructive framework for the human and AI relationship. The Empathy Outsource raises the question of what the partnership posture looks like when the domain is not productivity or analysis but emotional care, and whether that boundary requires a different kind of governance entirely.

Human Factor Podcast Episodes

Season 2, Episode 021: When Generations Collide: The Generational Fault Lines of Organizational Transformation

Features Ryan Vet and explores generational dynamics in transformation. Directly relevant to The Empathy Outsource’s discussion of Gen Z’s comfort with AI-generated emotional content and the research showing that nearly three-quarters of Gen Z have engaged in some form of relationship with AI, representing a fundamental shift in how authenticity is defined across generations.

Season 2, Episode 015: The Contagion Effect: How Emotions Spread Through Organizations During Change and Transformation

Explores how emotions spread through organizations, drawing on research from Hatfield, Cacioppo, and Rapson as well as Sigal Barsade’s foundational work. The Empathy Outsource raises the question of what happens to emotional contagion when the original emotional signal was generated by AI rather than felt by a human being.

Season 2, Episode 017: The Algorithmic Mirror: What AI Reveals About How We Actually Think, Decide, and Deny

The companion podcast episode to the newsletter article. Explores AI as a diagnostic lens for organizational behavior, the same framing that informs The Empathy Outsource’s argument that AI-generated empathy reveals more about our capacity constraints than it does about the technology.

Season 2, Episode 019: Structural Silence: Why Organizations Train People Not to Speak

Examines how organizations systematically train employees when and how not to speak. The Empathy Outsource extends this into the emotional register, where organizations have never taught employees that emotional expression is a professional competency, creating the conditions under which AI delegation becomes the default.

Season 1, Episode 006: The Communication Paradox: When More Words Create Less Understanding

Explores why comprehensive communication strategies often backfire and how cognitive overload prevents genuine comprehension. The Empathy Outsource describes a new form of the communication paradox where AI enables more empathetic words while potentially producing less genuine understanding.

Season 1, Episode 007: The Vulnerability Advantage: Why Admitting Weakness Makes Organizations Stronger

Explores how measured vulnerability strengthens organizational trust. The Empathy Outsource raises the question of whether AI-generated empathy represents the opposite of vulnerability: a form of emotional expression that carries none of the personal risk that makes genuine care meaningful.

Season 1, Episode 009: Transformation Fatigue: When Your Organization Can’t Absorb More Change

The cognitive depletion argument in The Empathy Outsource builds on the transformation fatigue research explored here, where exhaustion becomes the precondition for delegating not just tasks but emotional expression to AI.

Sources

1. Yidan Yin, Nan Jia, and Cheryl J. Wakslak, AI can help people feel heard, but an AI label diminishes this impact, Proceedings of the National Academy of Sciences, 2024
2. Matan Rubin, Amit Goldenberg et al., Comparing the Value of Perceived Human Versus AI-Generated Empathy, Nature Human Behaviour, November 2025 (nine studies, 6,000+ participants)
3. Frontiers in Psychology, The illusion of empathy: evaluating AI-generated outputs in moments that matter, 2025
4. Empathy as a predictive signal: why we devalue AI empathy, Trends in Cognitive Sciences, 2026
5. UK employee trust survey reported by Raconteur, 2025
6. Anthony Coman (University of Florida) and Peter Cardon (University of Southern California), research on supervisor use of AI-assisted writing, 2025, reported by HR Dive
7. AI-disclosed performance feedback research referenced by Lattice, 2021