What exactly are we protecting when we try to keep artificial intelligence out of our email?
There is a sentence I keep hearing. I don’t sit inside these companies, but I hear it coming out of them: in conversations, on stages, in what people pass along once the topic turns to AI. It goes like this: “We can do AI. But it matters to me that our people still write their own emails.”
Sometimes it’s email. Sometimes it’s: the concepts have to come from people. Or: we write the important texts ourselves. It is always the same sentence. Only the object changes. I hear it so often because strategic communication is what I think about, so my ears prick up when it comes. Email is just the handiest example. What’s meant is more. It’s the whole stack a company thinks and speaks with: emails, texts, concepts, decisions. Still, let’s stay with email. You can show everything with it.
I like the sentence. I really do. It has a stance, and I prefer a stance to enthusiasm without a reason. And yet I can’t shake it. I suspect it decides more than the person saying it intends. So let me take both seriously. First the sympathy. Then the suspicion.
The occasion: Der Spiegel just made up its mind
I’m not writing this in a vacuum. Der Spiegel, Germany’s leading news magazine, has just decided that no AI may write or rewrite its texts anymore. Editor-in-chief Dirk Kurbjuweit did it by striking a single word from the newsroom guidelines. Until then, AI was allowed as long as a text came “predominantly” from humans. The German word was maßgeblich. It’s gone now.
What’s interesting is how he describes the debate behind it. Long. Heated. Two camps. On one side, the people he calls “car builders”: those who want to build their text themselves, front to back. On the other, the “inspection crowd,” named after the TÜV, Germany’s vehicle safety inspectors: people who would let AI write and then check and sign off on the result. Kurbjuweit sided with the car builders.
You can see why this makes me sit up. These are exactly the two camps a company faces when someone says their people should still write their own emails. Build it yourself, or check it and sign off. I’ll come back to that. First, though, I have to ask the question Der Spiegel answered for itself long ago.
First the question nobody asks often enough: what is an email, anyway?
Before I judge whether a machine may help write an email, I want to know what we’re talking about. And that isn’t obvious. An email isn’t one thing. It’s many.
There’s the email about two people. The private note. The birthday message. The love letter to your partner, if you like. I wouldn’t hand that to an AI unchecked. Not on principle. Because something other than information is at stake.
A gift works in two directions. It does something to the person who receives it. And it does something to the person who makes it. If I give away the letter to the person I love, no matter to whom, I’m not giving away the result. I’m giving away the fact that I did it myself. That was the actual proof. Here, origin counts. This isn’t a question of efficiency. It’s provenance.
I’ve written about this at more length elsewhere. In short: authenticity isn’t a property of the medium. It’s a relationship. Between source, context and what I have a right to expect. There never was a golden age of authenticity when everything was pure and untouched. There was only ever the question of what can rightly be expected at this point, in this moment. With a love letter, I expect a human hand. And rightly so.
Then there’s the other email. The business one. And there everything shifts.
Business is a different matter
Here I understand the reflex right away. Of course the information has to be correct. Of course nothing sensitive, nothing wrong, nothing half-baked can go out. That’s what we have people for, as the layer that checks. That’s right, and it stays right.
But between “a person checks” and “the machine may do nothing at all” lies a wide space. And I see us closing that space too quickly. Whoever keeps AI out of email on principle may be protecting themselves from one risk. They’re also giving something away. The chance to put a thought in clearer order. To think a little more from the reader’s side. To be a little less long-winded.
And let’s be honest about the state of our business mail. We ramble. We write long emails, longer documents and even longer concepts, with the actual point buried somewhere in the middle. And then something happens that hardly anyone talks about: the AI on the other end shortens the whole thing anyway. Whether secretly at someone’s desk or in the company’s approved tool makes no difference. My three paragraphs become three lines before the recipient has even read them.
For me, that’s the punchline. We argue about whether a machine was allowed to help write at the start. And we overlook that at the end, one is almost always reading along.
In more than twenty years of radio, I learned that you have to say what it’s about in the first few seconds. Otherwise the listener switches off, without a trace of guilt. The inbox is no different. Except that today a machine has its thumb on the skip button. So the craft isn’t keeping AI out. The craft is rethinking what a business email is actually supposed to do. Less text, more point. Reach is output. Impact is outcome. The inbox is no exception.
The mistake lives in the word “real”
Now the point where I have to disagree.
Beneath the nice sentence there is often a quiet assumption. It goes: whatever AI has touched is no longer real. Not a real achievement. Somehow fake. That sounds like protecting authenticity. In truth it’s a dead end.
Because if I hold people responsible for how they use AI tools, and we do, for good reason, then I assume the work was checked. Responsibility and checking are the same thing seen from two sides. To assume from the outset that anything an AI has touched is now a forgery and no longer an achievement isn’t fair to the people who handle these tools well. And it’s expensive.
The industry defending its posters against AI images senses something true here. People value visible effort. In early 2026, a guest article in Tagesspiegel Background, a German policy briefing, pulled together something related: where effort becomes visible, people are more willing to trust a brand and buy from it. Authenticity doesn’t depend on the how alone. It depends on the who. I agree. It just doesn’t follow that every tool destroys authenticity. What follows is that origin and responsibility have to stay visible.
How absurd the other reading gets is shown in a small scene in the German magazine KOM. A self-appointed personal branding expert advises sprinkling your AI-written posts with one or two deliberate typos, so the text feels “more authentic.” Someone in the audience goes one better: he always takes out the dashes so nobody notices the machine. That’s the authenticity we get when we confuse “real” with “untouched by AI.” Staged sloppiness. A typo is not a personality.
Out of philosophy, into the market
Up to here, you could call all of this a matter of taste. Every company is free to decide how it wants to work. And I’m not going to get into the philosophy of any one company. I care about something more sober. A concern about the market.
Because by now it’s fairly clear who in companies uses AI most eagerly and most broadly. It isn’t the people who meet anything digital with suspicion anyway. It’s the digitally fluent high performers.
A Gallup survey from late 2025 shows this clearly. In technology and IT, about three in four employees use AI in their role. In finance and professional services, more than half. In retail it’s a third; in healthcare and manufacturing, a little over four in ten. In its State of AI, McKinsey describes the “high performers” as the organizations that use AI to grow and rebuild processes, rather than to save money. And it’s precisely these that have clear rules for when a result must pass through a human. PwC’s AI Jobs Barometer 2026 measures productivity growth in the top fifth of the most AI-exposed companies that leaves the rest far behind. So among the best, the human layer that checks doesn’t disappear. It gets taken more seriously.
And these people are already in your building. They’re already using AI, approved or not. The numbers on so-called shadow AI are unambiguous. In November 2025, UpGuard reported that about four in five workers use AI tools that haven’t been approved. A Software AG study puts it at half. For executives and senior managers, Cybernews even arrives at over 90%. And blocking it doesn’t win back control: almost half would simply keep using the tools even under an explicit ban, just more quietly. IBM has even calculated that a data breach via shadow AI costs more on average than one through official channels.
Feel free to read that twice.
Der Spiegel is right. Just about something else.
Which brings me back to Der Spiegel. Was Kurbjuweit wrong? No. For a human medium, the decision is right. There, writing is the product. The text is the achievement, the hand is the merchandise. Kurbjuweit even cites a study the Financial Times pointed to: seven in ten readers prefer that no AI was involved in the writing. That’s the love letter, at the scale of an entire newsroom. The reader expects the human hand. And gets it. Rightly so.
But that is exactly the limit. A Spiegel essay is not a quarterly update. What Der Spiegel is protecting is its identity, its Gemüt, its temperament, the thing people buy it for in the first place. An internal status email protects no identity. It carries a piece of information. Whoever takes the Spiegel decision and lays it one to one over their company’s inbox is mistaking the love letter for the quarterly update.
And there’s something else in that editorial that tends to get skipped. Even Der Spiegel doesn’t impose a purity law. “Neither robots nor Reinheitsgebot,” Kurbjuweit writes (in translation; the Reinheitsgebot is Germany’s centuries-old beer purity law). AI stays allowed for translating, for fact-checking, for analyzing large amounts of data. Exceptions are labeled. The house that most resolutely defends the human text does not claim that everything AI touches is fake. It distinguishes by depth of authorship. I didn’t ask for more than that above.
The sentence decides more than it says
I still consider the stance legitimate. I want to say that explicitly, because I don’t want to talk anyone out of wanting their own handwriting. There has to be a space in which a company makes its own decisions. I’ll defend that space.
But the nice sentence holds more than the wish that people keep typing their emails by hand. It holds a digitalization decision. And I’m not sure that in the moment we say the sentence, we’re aware of how far it reaches.
Because whoever demands it is, in the end, mostly demanding manual work. Not just in the inbox. On every text, every concept. And in substance they’re saying: I bought these AI tools, but I don’t actually want anyone to use them. Then working with them is effectively forbidden. Only nobody said so. This is where it connects back to the market. The question isn’t whether a company may decide this way. It may. The question is how long it can keep deciding freely if it meets digital transformation with hard, unexamined reservations. While its own high performers have long been using the tools. And the competition is putting them to work.
Bans can protect us emotionally. That’s no small thing. The wish for the real, for one’s own hand, is human and honorable. But they don’t protect us from reality. Reality is already in the inbox. On both ends.
Where I draw the line
So that no misunderstanding remains. I don’t think AI should be allowed to help write in business “now and then.” I think AI-assisted writing should always be allowed. The line doesn’t run between human and machine. It runs along judgment.
As long as a person can decide what gets said, as long as they hold the position, can judge the result and stand behind it, they may use any tool to say it. Authenticity isn’t decided by the tool, but by whoever has their hands on the wheel. “Allowed” isn’t really the question. Owned is the question.
What then remains as a real problem isn’t the machine. It’s arbitrariness. The moment when nobody decides anymore what is actually supposed to be said. With AI, that moment comes faster. But it isn’t a machine error. It’s an error of judgment.
And the love letter? It isn’t an exception to this rule. It’s a different game. In private, between two people, the gift isn’t the message. The gift is that I did it myself. What counts there is the hand, not a judgment about the content. That’s no ban on AI. It’s a different field with a different value.
I defend both: my conviction that it should be allowed, and a company’s right to see it differently. Only it should be a decision. Not a habit that mistakes itself for a stance.
Maybe that is the actual task. Not keeping the machine out. Not letting it in blindly, either. Knowing when it’s about the relationship and when it’s about the information. When origin counts and when clarity does. When I write the love letter. And when the quarterly update.
One doesn’t betray the other. It’s just a different promise.
What future do we want to build together?