I Read My Old AI-Drafted Emails. Please Do Not Ask Me to Read Them Aloud.

This weekend I did something no sensible man should do on a Sunday. I went into my Sent folder. Not for anything important, only to find one old thread. Instead I ended up reading some of my old emails, the ones I had drafted with AI’s help.

They were borderline cringe. Actually, remove the borderline.

The worst ones were the charming ones

You might expect AI cringe to look stiff and corporate. Mine was the opposite. It was trying very hard to be friendly and human.

Here is one, lightly redacted. I had just started in a new role, and this was my first hello to a colleague I had never met:

Saurabh here – new on the team. Two weeks in now and still exploring cupboards in kitchen.

I am picking up a couple of the digital products initially and trying hard not to add ideas to the pile without anything shipping. You are closer to delivery reality than I am, so I’d value a chat when you have some time.

Read it once and it is fine. Warm, even. Read it again now, and every sentence is wearing a little name badge.

“Still exploring cupboards in kitchen.” That is a joke engineered in a lab to be relatable. It has the shape of humour without the humour. “Trying hard not to add ideas to the pile without anything shipping.” That is humility, performed, the kind of sentence that belongs on LinkedIn with a photo of a sunrise. “You are closer to delivery reality than I am.” Flattery with a purpose, delivered so smoothly you can hear the ball bearings.

And anyone who has worked with me for more than a week knows I do not try hard not to add ideas to the pile. I add ideas to the pile at roughly four per meeting. The email was describing a much better-behaved man.

Once you see the tells, you cannot unsee them

So I did what any architect does with a painful artefact. I mapped it. Line by line, what the email said against what I actually meant. It was therapeutic.

Add up the right-hand column and the whole email was: Hi, I am new. Coffee? Five words did the job of fifty-nine, and the five words were the only ones that sounded like me.

And now look again at the original email. At that little dash sitting right after my own name. Regular readers know this blog has a strict no-em-dash rule. Now you know the origin story. It is not a style preference. It is a trauma response.

Why it reads as cringe, and it is not because it is AI

Here is the architect in me talking. A language model writes the most probable email. Ask it for a friendly new-joiner note and you get the most probable friendly new-joiner note, which is the average of every friendly new-joiner note ever sent. And charm, averaged, is no charm at all. It is a template of charm.

Did my colleague notice? Honestly, I do not know. Maybe, maybe not. And once we actually met, that email probably evaporated. One coffee overwrites fifty-nine words very efficiently. So no great harm was done to them.

The cringe is on my side of the table. That email was a first hello, the one moment when the other person has nothing to go on except what you write. No meetings yet, no history, no hallway chat. It is the only time your words carry the whole of you. And I handed that moment to a template.

One email like this is forgettable. The habit is not. If every first hello, every intro, every “keen to connect” goes out in the same borrowed voice, you are introducing a very polite stranger to everyone you meet, and then spending the first coffee quietly replacing him with yourself.

How I use it now: editor, not author

I did not give up on AI for email. That would be like giving up on the car because you once drove it into a hedge. What changed is the order of operations.

I write the ugly first draft myself. Half sentences allowed. Typos allowed. Then the AI gets a much narrower job:

Here is my draft. Keep my voice.
Cut it by a third and flag anything unclear.
No greetings, no exclamation marks, no jokes I did not write.

That is it. Tighten, do not rewrite. Where it helps, I give it a few of my own real emails as a sample so it knows what “my voice” even means, because left alone it will assume my voice is a LinkedIn post.

And some emails never go near it at all. Apologies. A thank-you that actually means something. First contact with someone I will work with for years. Anything where the relationship is the message. If the reader will care who wrote it, I write it.

The one place AI is genuinely brilliant? The angry email. I write it hot, then ask: how will this land with the person reading it? It answers like a calm friend who has read the room. That single use has saved me more trouble than every polished-by-AI email ever cost me.

Your voice is the one thing you cannot outsource

An email is a small thing. But you send thousands of them, and slowly, without anyone deciding it, people build a picture of you out of those small things. How direct you are. Whether you are funny, or just trying. Whether a message from you is worth reading properly.

Borrow the voice and the picture they build is of someone else. A pleasant someone, polished, and nobody in particular. Your own clunky, slightly too direct, occasionally typo-ridden voice is the only one that builds a picture of you. It is also, not by coincidence, the only one people actually remember.

Cringe is actually good news

So yes. My old emails are cringe. But do one thing before you feel bad about your own: notice that you can only cringe at something you have outgrown. Cringe is just growth looking over its shoulder.

The Sent folder is a kind of mirror. For a while, mine was showing me someone else, polite and charming and nobody in particular. Now it shows me again, rough edges and all.

And honestly? These days when I want a coffee with someone new, I just ask for the coffee. Five words. It turns out nobody ever needed fifty-nine to say yes.

Written for KiwiGPT.co.nz · Generated, Published and Tinkered with AI by a Kiwi