AI is eating copywriting. Here's what it can't chew

In 2023, three researchers looked at what happened to freelance writers on one of the world's biggest online job marketplaces after ChatGPT came out. Within months, freelancers in the jobs most exposed to it had about 2% fewer jobs and were earning about 5% less a month.

That's not the bit that kept me up. This is: being good didn't protect you. If anything, the freelancers with the best track records took a slightly bigger hit.

I've spent fifteen years in communications. I lead a content team. I write for a living, more or less. So I'm going to try and be honest about this, which is harder than it sounds, because the two easy positions are both rubbish. AI will never replace real writers! is cope. Writing is dead, learn to prompt! is a LinkedIn carousel. The truth is more annoying than either.

Let's not pretend. A chunk of copywriting work is going, and some of it has already gone.

Forty product descriptions by Friday. Ten subject lines to A/B test. Six hundred words of SEO filler about "the importance of choosing the right accounting software". The meta descriptions nobody reads. If your job was turning a clear brief into competent words, quickly, I have bad news, and the bad news types faster than you and doesn't need lunch.

You can tell, too. The internet is filling up with copy that has the texture of a hotel lobby. Everything is seamless. Every product empowers you to elevate something. Somebody, somewhere, is always delving. It's competent, it's grammatical, and it's the written equivalent of beige.

Here's the thing nobody tells you when you start out. Most of a writer's job isn't the writing.

It's turning can we do something about leads? into an actual brief. It's finding out what's true. It's getting the right people to look at it, and checking it before it goes out. The words are the last ten percent. The other ninety is chasing, asking, arguing and checking.

AI is brilliant at the ten percent. It's genuinely, irritatingly good at it. What it's terrible at is the ninety. And for years the industry paid for the ten and got the ninety thrown in free, because they came in the same human. That bundle is what's breaking.

I have four rules for my own writing, and they're on my portfolio. Lead with a verified number. Take a costly position. Check every fact against primary sources. Answer every comment. Funny thing: every one of them is something a model is bad at. So here's the longer version, for anyone who writes for a living and is feeling a bit sick about it.

1. Be the one who knows what's true. A model will invent a statistic with the calm confidence of a man giving directions in a city he's never visited. Being the person who goes to the primary source, reads the actual report, and says that number isn't real is not a small skill any more. It's the skill.
2. Take a position that costs something. A language model is, quite literally, built to produce the most likely next word. The most likely next word is also the most forgettable one. Say the inconvenient thing. Admit the weakness in your own argument. Readers can smell the difference between a view and a summary.
3. Go where the model can't. It has read the internet. It has not sat in the canteen. It doesn't know what the engineers actually think of the reorg, or what the people inside a company are worried about this week, because none of that is written down anywhere. Talk to humans. Interview. Listen. Steal good lines from real people (with permission).
4. Own the brief. Turning a vague ask into a clear brief is strategy wearing a cardigan. Whoever writes the brief decides what gets made. Be that person.
5. Be the editor, and the sign-off. Somebody has to put their name on it. A model can't be embarrassed, can't be fired and can't be held to account, which means it can't be responsible. You can. Taste and accountability are the job now.
6. And use it. Properly. The writers who'll struggle most are the ones who refuse to touch it and the ones who touch nothing else. Let it draft, outline, argue with you, find holes. Then do the ninety percent yourself.

Two of them, actually.

First: some of these jobs aren't coming back. Just upskill! is very cheap advice when it comes from someone whose own job isn't on the line. If you were paid by the word to write product descriptions, a list of six tips doesn't pay your rent. I know that. I'm not going to pretend the list fixes it.

Second, and this one's about me. I use AI in content production. Human sign-off, every time, but I use it. I'm part of the reason the pile of "words on demand" work is getting smaller. It would be a bit rich to write this post as if I were standing safely on the shore.

And there's a question under all this that I genuinely can't answer. Juniors used to learn the ninety percent by doing the ten. You wrote forty boring descriptions, someone senior tore them apart, and slowly you learnt taste. If the machine does the forty descriptions, where does the next generation of good writers come from? I don't know. I don't think anyone does yet.

No. But typing is.

AI didn't kill writing. It killed the pretence that the typing was the job. The job The real job was always judgement: what to say, what's true, who it's for, what to leave out, and whether you're willing to put your name on it. For years that was hidden inside the word count. Now it's out in the open, and it's the only part anyone will pay for.

Which, if you're good at the ninety percent, is honestly good news. And if you've only ever done the ten, it's time to go and learn the rest. Today. Not after the next round of layoffs.

The machine writes. You decide. Make sure you're the one deciding.

Is AI replacing copywriters? Partly. The commodity work (quick, competent words to a clear brief) is shrinking fast. The judgement around it isn't.

What's the evidence? One early study of a big freelance marketplace found writing-related freelancers got about 2% fewer jobs and earned about 5% less a month after ChatGPT launched, and being highly rated didn't protect them.

What should a writer learn now? Fact-checking against primary sources, taking real positions, interviewing people, writing briefs, editing, and directing AI well.

Should writers use AI? Yes, for drafts and arguments. Not for the final say.

Will it replace you, Tom? It drafts. I decide. Ask me again next year.

Tom