Leadership and AI6 min read
Stop Using AI to Write Your Empathy
Empathy is evidence that you noticed, not a writing task. Three tests for when an AI draft is legitimate, and what to write instead.
By Jerald Lee
In brief
A note that says the right thing while revealing that nobody was paying attention is worse than no note at all. Teams can tell.
Empathy is not a writing task. It is evidence that you noticed. A model can produce the sentence, find the register, and remove the awkwardness. It cannot supply the detail that proves someone was paying attention. That detail is the whole message. Without it you have a well-written note that could have been sent to anyone, which is another way of saying it was sent to no one. Three tests decide whether a draft is legitimate: did you see the thing, did you feel the thing, and would the person be surprised that a machine wrote it.
# Stop Using AI to Write Your Empathy
You have probably received one. A message after a bereavement, or a promotion, or a hard week, that says exactly the right thing and leaves you with nothing to hold.
The grammar is fine. The tone is considerate. It arrives on time, which is more than most people manage. And still, reading it, you have the faint sense that nobody was in the room when it was written.
That sense is usually correct, and it is worth taking seriously, because the person who sent it probably believed they were being kind.
The discomfort is the work
It is worth being honest about why anyone reaches for a model here.
Writing something difficult is uncomfortable. You sit down to write to someone who lost a parent, or who did not get the promotion, or whose project was cancelled after two years, and the blank page asks you to think about them properly. That is not a writing problem. It is the cost of attention, and it is supposed to cost something.
The model removes the discomfort. That is what it is good at. What it also removes is the thinking that the discomfort was prompting.
So the note goes out on time, and the sender feels that they handled it well, and the loop closes in their own mind. The person on the other end receives warmth with no evidence behind it, and now has to decide whether to be grateful for it.
What the model can do
It can produce the sentence. It can find the right register for a condolence, a congratulation, a difficult piece of feedback, or an apology. It can remove the awkwardness, tighten the phrasing, and give you a version that reads well.
That is genuinely useful when the problem is the writing. It is not useful when the problem is that you have not thought about the person yet.
The distinction is easy to miss because both situations feel identical from the inside. In both, you are sitting at a screen with something difficult to say. In only one of them have you already done the part that matters.
What it cannot do
It cannot notice.
It does not know that the person has been quiet in meetings for three weeks. It does not know that the project they were proud of got cancelled quietly, or that they asked for feedback twice and did not get it. It does not know that their father was ill in March and they never mentioned it again.
The specific detail is the whole message. Without it, you have a note that could have been sent to anyone.
Why people can tell
Not because the writing is bad. Because attention leaves marks.
A note that noticed something has a shape. It references a thing you said in a corridor, or a week you looked tired, or a decision you made that nobody acknowledged. It is slightly awkward in places, because the sender was trying to be accurate rather than smooth.
A drafted note is smoother than that. It is well proportioned. It contains no detail that could only have come from being there.
People cannot usually name the difference. They do not think "this was written by a machine". They think "this is kind" and then, a day later, they cannot remember anything in it.
The Empathy Test
Three questions before you send an AI-drafted personal note.
Did you see the thing? If the note contains no detail you observed yourself, it is a template with a name in it.
Did you feel the thing? If you are writing because it is the right thing to do, that is a reason to write. It is not the same as having something to say.
Would the person be surprised that a machine wrote this? If the answer is no, do not send it. If the answer is yes, you have written something that belongs to you.
The third test is the one that matters, and it is the easiest to fail while believing you passed, because the model is good at producing sentences that sound like they came from someone who was there.
Why the obvious objection is reasonable
Three objections come up, and each one is fair.
"I am not good with words." Often true, and it is a real problem. The answer is not to outsource the noticing. It is to write the detail first in plain language, then use the model to fix the sentences if you need to.
"I would have sent nothing otherwise." Also true, and worth sitting with. A drafted note is better than silence, but it is not the same as a note, and the person can tell which one they got. If the choice is genuinely between a draft and nothing, send the draft. Then ask why the choice was that narrow.
"Everyone uses it." Probably. That does not change what the recipient experiences. The standard is not what other leaders do. It is what the person in front of you needs in order to believe you were paying attention.
Where a draft is legitimate
There is a legitimate use, and it is narrower than most people assume.
It is legitimate when you know what you want to say and the writing is the obstacle. You have the detail, you have the feeling, and you want help finding the words. The model is a copy editor for something that already exists.
There is a second case. High-volume, low-stakes correspondence: scheduling, acknowledgements, routine follow-ups. Nobody is looking for evidence of attention in a calendar confirmation.
The test is not whether a machine wrote it. It is whether the person expects you to have noticed something. When they do, the noticing has to be yours.
Three practices
Write the detail first. One sentence about what you actually observed. Not the summary, the specific thing. The detail is the part the model cannot supply, and it is also the part that makes the rest of the message work.
Decide whether the note needs to be written at all. Some of the most useful responses to a difficult week are a conversation, not a message. A drafted paragraph can be a way of avoiding the room.
Write it badly first. The first version should sound like you, not like a model. You can edit for clarity afterwards. You cannot edit attention back in.
What this is really about
Here is the part that does not get said often enough about warmth at work.
Teams learn what is real by watching what leaders do when it costs something.
A leader who sends drafted warmth is teaching the team that warmth is a format. A leader who notices, and says the specific thing, is teaching something else. The second is slower and does not scale. That is the point.
The same logic applies to feedback. A drafted piece of feedback is usually a softened piece of feedback, and softening is often how the actual message gets lost. If the sentence would embarrass you to read aloud, it is probably the sentence that needed to be said.
The experiment
Before you open a model this week, write one sentence by hand about something you noticed in a person.
Not a compliment. An observation. "You have been quiet since the client review." "You looked relieved when the timeline moved." Something only you could have seen.
One sentence. One person. No prompt. That is the whole exercise, and it will tell you whether the noticing is still yours.
Where this connects
[Why I Left Google](/why-i-left-google) is about what changes when the structure that managed your attention disappears. [What Self-Leadership Looks Like After Corporate](/self-leadership-after-corporate) is about the part that does not automate: deciding what deserves your attention in the first place.
If the pattern is that you are managing more and noticing less, [our engagements page](/engagements) explains where that work sits.
Frequently asked questions
Can AI write an empathetic message?
It can write the sentence. It cannot supply the attention the sentence is meant to prove. If the note contains no detail you observed yourself, the reader will usually feel the difference even if they cannot name it.
When is it acceptable to use AI for a personal note?
When you already know what you want to say and the writing is the obstacle. It is not acceptable when the model is doing the noticing for you. The Empathy Test above is the short version.
Does this apply to feedback as well?
Especially to feedback. A drafted piece of feedback tends to be softened, and softening is often how the message gets lost. Write the specific observation first, then decide whether it needs help with the phrasing. ## A five-minute reflection What is one thing you noticed about someone this week that you have not said out loud? ---

Jerald Lee
Commercial Systems Advisor
Jerald Lee is a Hong Kong-based Commercial Systems Advisor, founder of Curiosity At Work and Operating Partner of The Growth Coach Hong Kong. He spent 13 years at Google APAC and helps founder-led companies build predictable revenue systems.