Something has started to feel different in communication lately and for a while I couldn’t quite put my finger on what it was. The emails aren’t bad. Actually, that is part of what makes it interesting. They are often really good: perfectly structured, friendly, professional, calm and empathetic. There is an “I really appreciate you sharing this”, an “I completely understand your perspective” and occasionally we are all “navigating this journey together”. Nothing is technically wrong, but sometimes you read something and still feel that something is missing.

At first I thought it was effort, but that isn’t quite it either. I genuinely don’t care whether someone spent twenty minutes typing an email or twenty seconds using AI to help write it. I use AI constantly myself and that is exactly what it should help us with. Please use it to make an email clearer, improve the grammar, structure messy thoughts or stop the angry 23:48 version from reaching someone’s inbox. We really don’t need to go back to staring at an empty Outlook window for half an hour because apparently suffering makes communication authentic. 😂

What I think we need to pay much more attention to is something else: relational effort.

Not the effort of producing the words, but the human effort around them. Did someone stop and think about the person receiving the message? Did they remember what was discussed before? Did they consider what might sit behind the question? Did they decide what they actually think? Did they consider how the answer might land? Did they decide whether an email was even the right response?

That is different from writing effort, and generative AI is starting to separate the two.

For most of our working lives, the words we sent each other carried little signals of human investment. A carefully written response usually required some time. A personal answer required context. Finding the right words meant thinking about what you actually wanted to say. None of that guaranteed empathy or care; humans have been perfectly capable of writing terrible emails without AI for decades. But there was at least some relationship between the effort required to create communication and the communication itself.

Generative AI changes that relationship. Anyone can now produce a beautifully written, apparently thoughtful, deeply empathetic 800-word message in seconds. AI can make it warmer, softer, more personal and more understanding. It can reproduce many of the linguistic signals we associate with someone taking care over a message while dramatically reducing the effort required to produce those signals.

And that isn’t necessarily bad. In many cases it is fantastic. But it changes the signal. 🙈

Economist Joshua Gans has examined this through the economics of signalling: when generative AI dramatically reduces the cost of producing high-quality communication, some of the signals we traditionally associated with effort and quality become harder to interpret. When almost everyone can cheaply create polished communication, polish itself tells us less about what happened behind it.

Maybe that is what we are beginning to feel. It isn’t necessarily “this was written by AI.” It is something more subtle: how much human attention sits behind these words?

That becomes even more interesting when we talk about empathy because we have almost turned empathy into a feature. Make this more empathetic. Click. Done. Except empathy didn’t happen. The language of empathy happened.

Real empathy requires context. What happened before this message? Why is this person asking this? What might they be worried about? What haven’t they said? What has already been promised? Do they need reassurance, recognition, clarity, a decision, an apology or simply five minutes of someone’s attention?

AI can write “I completely understand how frustrating this must be.” But those words alone don’t prove that anyone actually understood.

Research is starting to show how interesting this distinction is. In a series of nine studies involving more than 6,000 participants, researchers found that the same AI-generated empathic responses could be perceived differently depending on whether people believed they came from a human or from AI. At the same time, other studies have shown that AI-generated responses themselves can perform extremely well when people judge qualities such as compassion and empathy.

That creates a fascinating contradiction: AI can become incredibly good at producing the language of empathy while humans still value the human intention behind it.

Maybe empathy isn’t only contained in the words. Part of empathy may be knowing that another human made an effort to understand you. CAnd that matters enormously in leadership.

People don’t experience leadership mainly through strategy decks, job titles or values written on an office wall. They experience it through hundreds of tiny interactions: whether someone remembers what they said last week, notices that something isn’t right, asks the second question before immediately giving an answer, calls after a difficult meeting or recognizes that an apparently simple email has three months of history sitting behind it.

Those interactions communicate something bigger: I am here. I see you. I am paying attention. AI can increasingly reproduce the language associated with that presence. That is enormously useful, but it also means leaders need to become more intentional about where actual human presence matters.

There is another layer to this. AI doesn’t only reduce the effort required to create communication; it dramatically increases the amount of communication we can create. More emails, more summaries, more documents, more updates, more messages and more beautifully structured explanations of things that perhaps needed three sentences.

Imagine a leader uses AI to create a 900-word update in twenty seconds and sends it to 300 colleagues. If everyone spends four minutes reading it, the sender may have saved ten minutes while creating twenty hours of reading somewhere else in the organization.

Congratulations. Productivity. 😂

That is why AI productivity cannot only be measured from the perspective of the producer. AI makes words cheap. Human attention remains expensive. The question shouldn’t only be “How quickly can we create this?” but also “Does this deserve someone else’s attention?” And this is where priorities become important.

Most AI personalization today focuses on how someone works. Tone of voice, preferred language, formatting, writing style and recurring tasks. Useful, but the next level may be teaching AI what actually matters. What are this leader’s priorities? Which relationships deserve attention? Which decisions genuinely require them? Which commitments have been made? Where should their time deliberately not go?

Because AI is extremely good at optimizing whatever we put in front of it, but not everything in front of us deserves to be optimized.

A leader receiving 80 emails can use AI to answer all 80 faster. That looks productive. But perhaps only six deserve that leader’s personal attention. If AI helps process all 80 without distinguishing those six from the other 74, we haven’t necessarily transformed work.

We may simply have automated being busy. For leaders especially, typing speed isn’t the scarce resource. Attention is. So what can we actually do differently? I think there are five relatively simple shifts.

First, give AI priorities, not only a tone of voice. Don’t just tell an AI how to write. Give it context about what matters, which relationships deserve attention, what outcomes are important and what should deliberately not consume time. Then the question changes from “Can you answer this?” to “Given these priorities, does this deserve personal attention?”

Second, don’t ask AI to add empathy. Ask it to help create understanding. Instead of “make this more empathetic”, ask “Before writing anything, what might this person actually need from the sender?” Maybe it is reassurance, clarity, recognition, a decision or an apology. Use AI to broaden perspective before using it to improve sentences. Empathy should happen before the words, not be sprinkled over them afterwards.

Third, ask whether communication should exist at all. Before generating an important response, ask whether email is actually the right medium. When something involves trust, disappointment, performance, conflict or emotion, the best AI output may not be a beautifully written message. It might simply be: have the conversation.

Fourth, use AI to create less communication, not only more. Ask whether something can be three sentences, who genuinely needs to receive it and what can be removed without losing the meaning. AI shouldn’t only save time for the sender. It should protect attention for the receiver too. Saving ten minutes while creating an hour of work for everybody else isn’t productivity.

Fifth, let AI challenge the relational effort behind communication. Instead of only asking it to improve a message, ask it to question the thinking behind it. Does this actually answer the concern? What context might be missing? Does this sound caring without addressing the real issue? Is responsibility being avoided? Would this person be better served by a conversation than another message?

That is where AI becomes much more interesting than a writing assistant. It creates a little friction before action. And perhaps that is exactly what we need.

The question “Was this written by AI?” will probably become less useful anyway. AI will increasingly know our vocabulary, rhythm, preferences, context and communication history. Eventually the distinction between human-written and AI-assisted communication may become almost impossible to see.

There is a much more important question:

Were you actually present in what you sent? Did you understand the context? Did you decide what you believed? Did you consider how it would land? Did you take responsibility for the message? Did this moment receive the human attention it deserved?

All of those things can be true while AI helped write every single sentence. And that is why the answer isn’t less AI. We should absolutely let AI remove unnecessary effort. Let it summarize the 40-page document, prepare the meeting, draft the standard response, search for information, improve the first draft and remove unnecessary words.

But then use some of the effort AI gives back for something better. Think. Ask another question. Make a decision. Call someone. Listen. Spend time on the people and priorities that actually matter.

Because perhaps this is one of the biggest shifts we haven’t fully understood yet. For years, words carried signals of time, thought, attention, memory, judgment and care. AI can now reproduce many of those signals almost instantly, which means the value starts moving somewhere else. Perfect words become less scarce. Human attention becomes more scarce. Context becomes more valuable. Judgment becomes more valuable. Authenticity becomes more complicated. And relational effort may become one of the most important signals of leadership.