Yes, I'm using AI to write...

…and after reading this you’ll understand: IT DOESN’T SAVE TIME AT ALL!

Yesterday The Media Leader published my latest column: “5 reasons why WPP’s next CEO should be AI”.

It’s part-satire, part-diagnosis… but 100% one of the opinion pieces I’m proudest of since leaving TML as editor to be a freelance journalist and consultant in January.

In the spirit of using my new Substack to be honest about the impact of AI on media and advertising, I wanted to share exactly how I use gen-AI tools, in combination with all the other “analogue” stuff that is still highly useful after 15 years of journalism.

Yes, you read that right: I used AI to write this article.

But not in the way, I suspect, you would have thought.

So I’m going to walk you through exactly how I did it. Partly as a record for myself, and partly for anyone else trying to navigate the strange middle ground between human voice and machine tooling.

By the end, you’ll see that this way of working is not some kind of hack, dodge or scam. It’s experimental, and may change a lot, but we’re not going back to ‘the old ways’ any more than we’re going back to using parchment and quills.

The TDLR takeaways are:

• As you’ll see, there are many steps involved and I highly doubt it ends up saving you much time.

• A lot of the ‘value’ from using AI comes from creating shortcuts through common writing obstacles, such as “writer’s block”, building structure, as well as fact checking and proofreading.

• I’ve adapted older-fashioned writing techniques into the AI chatbot interface i.e. imagining I’m pitching to an imaginary editor, rewriting a piece in a particular voice, and other techniques designed to get you to ‘think outside the box’.

And, to be clear, this is not a process I would advocate using as a ghostwriter without express permission from a client. You should NOT be using any IP, or other privileged or confidential information with an open-source LLM like ChatGPT.

---

1. Speaking first, structuring later

A lot of my writing starts like this:

I have an idea I want to explore — in this case, the idea that the role of CEO at a company like WPP has become so structurally hollowed-out that it’s barely distinguishable from a promptable function.

So I opened a dictation app and just talked. I rambled for a while. Then I took the transcript and dropped it into ChatGPT, along with a “briefing note” I’d built in NotebookLM — full of notes on agency structure, books about the industry I’ve taken notes on over the years (I’ve just re-read Madison Avenue Manslaughter, for example), analyst commentary, and recent WPP news/reaction to Mark Read’s resignation as CEO.

I’ve only recently starting using Google’s NotebookLM but it’s quickly becoming an essential part of my content arsenal. If ChatGPT is replacing Google as a better reservoir of human knowledge, NotebookLM is the most effective tool I’ve seen for being your second brain: it only stores what you put into it and gives you a huge amount of scope to slice and dice that knowledge however you want.

This gave me the raw material. But it didn’t save me time. It just gave me a starting point that wasn’t a blank page.

---

2. Letting the machine give me a bad first draft

The first thing ChatGPT gave me was… fine. It was what I expected.

In fairness, my first drafts are usually always terrible. But that’s by design and it’s the reason why I rarely get writer’s block.

You want to get past the horror of the blank page? Just. Start. Writing. Write anything! A bad draft is better than no draft. Get over yourself and write something, then whatever talent you have will compel you to improve on it.

Just like with the AI: getting it to write a terrible first draft can help clarify what you really want to write, as well as how you write it. Much more on that to come.

And yes, it sounded like ChatGPT. The writing was clean but bloodless. Lots of rhetorical binary:

“Mark Read is stepping down. The stock price barely twitched. That’s not a sign of confidence. It’s a sign of indifference.”

You know that style. I see it all over LinkedIn. It’s what happens when you ask AI to mimic “opinion writing” without much instruction. It hits familiar beats, but they feel off. No nuance. No rhythm. No voice.

Still, the draft was useful. It gave me something to edit, which is half the battle.

---

3. Experimenting with tone and format

I started playing with structure.

At one point I prompted ChatGPT to write it like Lawrence O’Donnell — the American news anchor who opens his MSNBC shows with long, fiery monologues. I like his style. It doesn’t always translate to the written page, but it shakes things loose.

Then I did something really difficult. I had a break.

Over the years you often hear other people complain that they don’t like stuff “hanging over them”. While I understand the human need for closure, you kind of have to get over it as a freelancer with multiple clients, multiple projects and multiple deadlines (particuarly when you’re working from home with young kids often marauding around).

In general I would always advise taking a break when doing anything creative. Get the juices flowing, then go away and let your subconscious work on it in the background.

Pro tip: do exercise. Scientific research has proven that physical activity boosts creative thought. Your brain will be working hard on that thing “hanging over you” in the background.

Hemingway went one step further by deliberately stopping himself writing when he knew what he wanted to write next; sometimes in the middle of a sentence.

It was during a run that evening that I remembered some specific feedback I had when I last interviewed Read a couple of years ago: that during our conversation you saw “a different side” than you normally get as when the big company boss speaks.

Eventually I landed on the tactic of turning the piece into a listicle: five reasons WPP’s next CEO should be AI.

Not because I love listicles, but because they’re a good way to group arguments and stop yourself from making one point 1,000 times. In this case, I needed to separate the piece into two: make the point about Read’s “different side” and lead into a wider commentary about how holding company leaders face an increasingly “impossible job”. Listicles are a blunt tool but they get the job done.

BTW, this is something ChatGPT is generally quite good at. If you feed it rich, nuanced source material, it can output a structure that holds together. That’s how the list formed—not from a pre-planned outline, but from a dialogue between me, my notes, and the machine.

---

4. Freewriting the intro (and running with it)

I didn’t like any of the AI-generated intros. Too neat. Too rhetorical.

So I used a technique I come back to often: freewriting.

I gave myself ten minutes to write without stopping, no internal editor, no judgement. I let my fingers go. What came out was a reflection on how hard it is to build a writing and consulting business in the age of AI — and how often people say, “Why would I pay you when ChatGPT is free?”

That unlocked the voice.

It let me contrast the human work of freelancing with the illusion of executive control at the top of a company like WPP. That intro—later rewritten and refined, but emotionally intact—made it into the final version of the article.

And I’m glad it did. It gave the piece heart. And it reminded me that writing with AI doesn’t mean ceding authorship. It means guiding the process — and knowing when to take over.

If you’re new to freewriting, this introductory book is excellent.

---

5. Using AI for structure, then reclaiming the prose

By this point, I had a listicle format, a personal intro, and a solid internal logic.

What I didn’t have was my voice running through all of it.

So I used ChatGPT again — but not to generate. To iterate. I pasted in my own paragraphs, asked it to tighten or expand, then rewrote the bulk of what it gave back. Think of it as editing with a sounding board that never sleeps.

There are many custom GPTs (”Chat with David Beckham”! “Tony Robbins Style Motivator”!) but none I would recommend. We are a very long way off making Westworld a reality.

Your insights will often come from your questions, more than what the AI answers. If you’re asking, “does this section sound too snarky?” then you need to explore why you suspect that.

As an ice-breaker, I vaguely ask, “tell me why this is bad”, just to see what comes back. Again, read the responses and test your reactions to it, rather than treat GPT as gospel.

And definitely watch out for responses like “excellent question!”

Because—and this warning is important—never forget what the AI creators want: engagement and training data. I’m under no illusions that these software creators have every incentive to try every trick in the book to keep you coming back, whether it’s by creating a sycophantic chatbot who tells you how wonderful you are (ChatGPT does a lot of this), or by using sophisticated psychological profiling to manipulate your sensitivities (which, after I asked, ChatGPT assures me it doesn’t do…)

But you can ask it to check facts. That matters as a first step, not a full editing replacement. It flagged some things I knew, and one thing I didn’t (a stat about average salaries that didn’t have a source). That got cut.

And obviously you can use it to spell and grammar check. Make sure you train it to use UK English (training happens with repeated usage and by adding instructions to the “Personalization” section in ChatGPT’s settings).

And, even after having done all of that, you still need to read the piece at least a couple more times. Ideally, put yourself in a different headspace to read it more critically. Little tricks like pasting the article into a different program with a different font, will help. As will reading it out loud.

Even better, send the piece to trusted advisors and ask for feedback. A second brain is great, but a brains trust is even better. Not just for projects, for your whole career!

---

6. So… did AI write it?

Not really. But it definitely played a role.

But I wouldn’t have written it, the way it was, without AI.

AI helped me bypass the blank page.

It helped me structure the idea, test tones, challenge formats, and accelerate the slowest parts of the writing process.

But the argument, the voice, the turns of phrase… they’re mine. The perspective is mine. The risk is mine.

The piece wasn’t about proving AI can do a CEO’s job. And this Substack isn’t about proving AI can’t do my job.

It’s about what happens when we automate judgement and optimise performance until there’s nothing left but choreography.

As for people like me, I think we owe it to clients and readers to be more transparent what tools we’re using.

Because using AI to do your homework is like hair dye; as soon as someone gets close to you, they can always tell!

🔗 Read the original piece

What was all that about?!

So this is me trying—hoping—that honesty and transparency about how I use AI will open up interesting conversations and advice. If nothing else, it’s a departure from all the tiresome “AI will never replace true human creativity” commentary on social media, the home of self-serving straw-man arguments parading as “insight”.

If this resonated with you, you might also like this earlier piece in which I implored marketers (of which I sort-of am one…) to do the hard work of creating core convictions. All the messaging that comes afterward should flow from that.

Tell me why I’m wrong in the comments.

Tell me why I’m right by inquiring about my writing, live events and consulting services. I work at the intersection of media, marketing and strategy — bringing 15 years of experience to help brands and people in media and advertising communicate with impact. omaroakesmedia@gmail.com