A colleague messaged me this morning, fuming. A client hadn’t liked her article. Their feedback was that it was “too airy-fairy”. So, instead of saying what they wanted, they asked their favourite AI agent for feedback, copy-and-pasted the results, and sent them to her with an injunction to write a new piece entirely.

Here’s the real rub, though: she had written to the original brief. Now she was writing to a different one, to a deadline that was already tight. And in fact, she said, the brief itself had been quite airy-fairy…

Now, I’d love to say this is the exception rather than the rule. More and more, however, I’m finding that clients, writers and AI are in a complicated three-way relationship, and I’ve been watching it play out from the writer’s corner. And a few things keep coming up.

“We’ll just use AI”

Lots of people have decided they no longer need a writer. And if you’re not someone who considers themselves a writer, or who is perhaps writing in their second language, I honestly get it. The tool is fast, free-ish and confident. The output is clean (albeit very American). Hey, presto! You’ve produced something without having to hire a writer or an editor. Win!

But getting an AI to sound human is hard. Getting it to sound like you is much, much harder. Left to its own devices, your AI agent writes like an average of everyone, and average is where your voice goes to die.

Losing it takes surprisingly little. A few unedited drafts, a few pieces that you think are ‘good enough’ because they sound polished, and suddenly your company (or you) sounds like every other company.

Welcome to the kingdom of beige. Please take your seat in the masses, never to be seen again.

“Here’s the brief”

This is a typical brief I get these days.

“[EXECUTIVE] will be speaking at the [TOPIC] Summit on [DATES] in [CITY]. The summit is hosted by [UNIVERSITY] in partnership with [A LONG LIST OF ORGANISATIONS]. They are scheduled to join the opening plenary panel: [PANEL TITLE WITH TWO COLONS AND AN AMPERSAND]. The discussion will focus on: the current landscape; the role of government, business, civil society, philanthropy and communities; practical examples of what is working; questions and calls to action for participants.

“Given the profile of the event, please develop a comprehensive briefing pack, including suggested talking points, potential panel questions and key [COMPANY] proof points.

“Word limit: 600.”

It looks like a brief. It has dates, a venue, a panel title, even a word count. (And please note the requirement of a “comprehensive briefing pack” in 600 words. Sure.)

But it doesn’t say what the executive wants to say. It doesn’t say what the company thinks, or what she’d like the audience to walk away believing. And the topic of the summit is so broad, it’s a bit like saying, “Write me a briefing document for a speech about the future.”

They’ve essentially sent me a diary entry, not a brief.

“Please fix by COB”

Now the opposite scenario. The client does brief properly, and you write exactly what was asked. And then it goes off to be ‘improved’ by AI. You get back a polished, wordy, says-nothing AI ‘edit’ draft that’s been copied and pasted, and the vague feedback to make it more engaging, more authoritative, more impactful. Oh, and let’s not forget, “more punchy” and “make this sentence pop”. And then they send it at 4.58pm and want a new draft “by COB”.

This raises my blood pressure and my levels of irk. Because what people just don’t seem to grasp, is that your AI agent doesn’t truly understand the context you operate in. It doesn’t know that the phrase it wants you to add means something quite different in Johannesburg than it does in Silicon Valley. It doesn’t realise that particular phrasing is going to really annoy the unions or upset every woman in the building. It has never met your CEO, and doesn’t understand how she thinks.

But it arrives as feedback, and you’re the writer, so you have to do as you’re told. And then, at the end of the process, someone asks: “Did you use AI for this?” <insert scream here>

You get the irony, right? No thought in the brief. No thought in the feedback. And the writer is the one being asked whether she did the thinking.

What I actually do all day

The thing that connects all of this is that good writing isn’t really about writing. It’s not only about producing a string of words, sentences and paragraphs. It’s really about thinking.

The words are the last 10% of the job. The other 90% is working out what you think, why it matters, and then, CRUCIALLY, who needs to hear it and what you want them to do next.

AI is very good at the last 10%. It can’t do the other 90% unless someone has done it first. If nobody has, all it can do is produce fluent, confident text with no real substance.

And that’s the trap. The output looks so finished that you skip the part where you work out what you think, who the writing is for, and what they need to get from it.

So, before you brief anyone – human or machine – try answering these questions, in a couple of sentences each:

• What do you actually think about this topic?
• Who do you want to talk to?
• What would you like them to believe, or do, when they finish?
• What are you assuming they will know that they might not?
• What’s the one thing you’d be annoyed to see left out?
• What do you not want said?

If you can answer those six questions, you can start to think about briefing your writer. You can also brief an AI properly, and you’ll know when it’s given you rubbish.

If you can’t answer them, that’s fine. It just means the thinking hasn’t happened yet. Maybe you don’t have time to do it. Maybe you need some support thinking it through – there’s no shame in needing a thinking partner. And that is exactly the bit a writer (a good one, who’s allowed to ask questions, and push back) can help with.

Drop me a line. It’s what I love to do.