What artificial intelligence can't write for you

Writing copy has stopped being a time problem. In two minutes you can have a service description, a social post and three versions of an email, all correct and all well written.

The problem has just moved. When everyone uses the same tools to write about the same subjects, the copy starts to sound the same — and your company becomes one more voice on an assembly line that produces correctness without saying anything.

A model writes the average

It's worth understanding what one of these tools actually does, no mystery involved: it learned from an enormous amount of text and produces whatever, statistically, tends to come next. It's good at that, which is why the result always sounds reasonable.

But the average of everything ever written on a subject is, by definition, the cliché of that subject. Ask it for a text about your company and you'll get bespoke solutions, a dedicated team and a commitment to quality — the same words we've already discussed when writing about brands that say what every other brand says.

It isn't a flaw in the tool. It's what it does.

What it can't know

There's a whole category of things no model can invent, however good it is, because they don't exist on the internet: they exist only inside your company.

• Opinions. What you think your sector gets wrong. A model avoids taking positions — it was built to.
• Refusals. The work you turn down, and why.
• Your clients' own words. The exact way they describe the problem when they call. It's almost never the sector's own jargon.
• Your own numbers. Real lead times, what tends to go wrong, what something like this actually costs in Portugal.
• What went wrong. A model writes success stories; it doesn't write the part where you got it wrong and turned back — which is the part that makes people believe you.

Where it's genuinely worth using

None of this is an argument against using it. It's an argument for using it in the right part of the job.

It's excellent at reworking a text that already exists: shortening it, changing the tone, tidying up loose notes, generating five headlines to pick from, translating, turning a technical explanation into language a layperson can follow. And it's a real help for anyone stuck staring at a blank page — a bad first draft is easier to fix than nothing at all.

The rule of thumb that tends to work: use it to shape what you already have, never to discover what you have to say. The substance comes from you; it handles the rest.

What you never hand off

Three things, and for reasons that have nothing to do with style.

A promise. Deadlines, guarantees, what's included. If the tool invents a condition you don't actually offer, it's the company that answers for it.

An apology. When something goes wrong with a client, a perfect, impersonal text does more damage than a clumsy sentence written by someone who was actually there.

A public stance. If the company is going to say it believes in something, someone actually has to believe it. Back to the same point: a model doesn't have positions, it has averages.

One practical consequence

There's a useful irony here. As the internet fills up with average text, text with concrete things inside it becomes worth more — to readers, and, as we've discussed when writing about appearing in AI answers, to the systems deciding which sources to cite as well.

In other words: writing with opinion, with your own numbers, and with what only your own company knows has stopped being a branding luxury and become the only way not to disappear. It's what we do in copywriting.

Could your copy belong to any company?

Send us a piece of your own copy — from the site, a proposal, a post. We'll tell you what's yours and what's just the sector average.

Request a quote