Yes, welcome to my “AI in Agencies” TED talk. You’ll be sick to the back teeth reading about how AI is changing everything, everywhere, all at once; I get it. But, since 2022, the use of AI in agencies and with clients has been a bit of a dirty secret. Each side warily protects its process instead of being upfront. Clients feed the brief straight into AI and skip the agency. And why wouldn’t they? The AI doesn’t push back. It just does it, and it’s easier to give feedback to ChatGPT than to client services. And agencies? Low-level copy gets knocked up and out in no time. It doesn’t even need to hit a creative.
But we’re not sharing our reality with each other, which I think is a shame, as we have a lot to learn from each other as we navigate this brave, scary, uncertain, and at times pretty stupid new world.
So, because we’re all grown-ups here, I want to be open with you about how we’re using AI today. And how we see that affecting clients.
Out of the box, AI is trained on the whole world. Every blog post, every press release, every LinkedIn think-piece ever written. Ask it to write something, and it hands you the middle of all that. The average. That's the problem.
You've felt this even if you can't name it. The work slop. The copy that's well written, structured, grammatical, and somehow says nothing. It reads as competent, which is exactly what makes it dangerous. It tricks you. It looks like value because it's polished, but there's no one in it. No taste, no point of view, no business behind it. It's the average of everyone, which means it's no one.
And it's starting to cost. Klaviyo surveyed 8,000 people across 8 countries this year. Even the ones who like AI and use it daily said they hit "AI slop" from brands multiple times a week. Spot AI in a brand's marketing and people are 4 times more likely to trust it less, not more. The polish that fools you doesn't fool them.
People are drowning in work that's technically fine and completely forgettable. Most organisations are responding by making more of it, faster. Playing with AI. Generating more. Going broader.
We've gone the other way.
Everyone at Good is on the same tools now. Claude, installed on every machine, everyone working in the same way, all of it tied to one central source of knowledge. Good knowledge.
That last part is the whole game. We are not asking AI to go out into the world and fetch what it thinks is relevant. We've removed that. The knowledge it draws on is curated by us, end to end.
What's in there? Our services. Every piece we've written for the Journal. Our definitions, the ones we actually argue about: what we mean by brand, what we mean by vision, mission and values, what we mean by product positioning. The distinctions that matter to us and that get blurred everywhere else. We've put the specific, hard-won version of how Good thinks into the machine, so the AI isn't reaching for the average definition of brand it scraped off the internet. It's reaching for ours.
That's the move. You're training it on the best of your business. AI gives you the average of everyone; we pull it back to the sharpest version of us.
If you're a client-side leader, this is the most important part to understand. The value of AI in your organisation isn't the model. Everyone has the same model. The value is what you point it at. Most companies point it at the entire internet and wonder why the output sounds like the entire internet. The ones getting something useful out of it have done the unglamorous work first: deciding what their knowledge actually is, writing it down, and feeding the machine that instead.
The next layer is more ambitious, and we're early in it, so I'll be honest about where we are rather than sell you the finished thing.
We're building our own brand into a machine-readable format. Not a PDF of guidelines that sits in a folder. A structured version that the AI can actually read and act on. Right now, it's foundations, and we’re continually working on this, but it’s worth sharing to see what an AI-readable format looks like.
The design team started this with a simple, awkward question. How do you turn brand guidelines into something an AI can build from? And once you start pulling that thread, you realise it can't just be the colour palette and the logo. That's the easy 10%. The real foundation is everything underneath: tone of voice, the do's and don'ts, the customer groups, the strategy behind the business. All of it goes in.
The point is that the foundation of anything the AI makes, visually or in words, comes from the brand. The machine understands the business and knows how to act on it. So it doesn't just look right, it sounds right, and it's built on the actual thinking, not a vibe.
This matters beyond design. Once your brand exists in a form a machine can read, the AI can do almost anything with it and stay on-brand while it does. That's a different proposition from "we used AI to make a thing." It's "the thing it made was ours from the ground up."
Underneath all of this sit two central bodies of content.
One is about Good. What we believe, our values, our definitions, our ways of working. The other is client information, an easily updatable wiki of who we're working with and what we've done.
Both have to be maintained. This is the part people skip, and it's the part that decides whether any of this works. A knowledge base that nobody tends becomes a graveyard. You've seen it. Every company has the intranet that went stale in 2019, the shared drive nobody trusts, the wiki that's wrong more often than it's right. The reason is always the same: keeping it current is somebody's job, and it's nobody's favourite job, so it doesn't get done.
Here's where AI changes the shape of the problem, and it's subtle. We're building it so the system can prompt the update rather than wait for one. You finish a piece of work, and instead of the knowledge just living in your head or your sent folder, the AI can flag it: this is new, should we put this in? Should the team have this? It turns maintenance from a chore somebody has to remember into a question the system asks you at the moment you've actually got something worth saving.
That shift sounds small. It isn't. It's the difference between a library and a graveyard.
The first payoff has been onboarding. Someone new starts, and instead of three weeks of asking colleagues the same questions, we can say: go to the AI. Ask it about the client, about the work we've done, about how we think. It'll get you most of the way. Anything important, come to us. The knowledge is there, it's current, and it's not trapped in one person's memory.
Let me be clear about the thing everyone's actually worried about, because dodging it would be its own kind of slop.
We're not using this to take people out. Nobody at Good is being replaced by a machine, and that's not coming. What AI clears is the tedious stuff. Call it 25% of the work that has to happen but that nobody's telling their mum about. The daft, repetitive, time-eating tasks that sit between you and the actual thinking.
Clearing that 25% doesn't make us want to produce 25% more. That's the trap, and it's worth being clear-eyed about it. We could churn out 7 or 8 concepts for a client now instead of 3. Quicker through the conceptual stage, more options on the table. And it would be worse. More concepts, less depth in any of them. The average problem again, just with our name on it.
So we spend the time we've saved going deeper. The same 3 concepts, but properly chewed on. More thinking per idea. AI gets us to the starting line faster; the value is what our people do once they're there, which is the part a machine can't do.
Because everything goes out to a human. We are not handing tasks to a machine to finish on its own. Nothing autonomous, nothing unchecked. AI clears the ground; the work still passes through people with taste and judgement. That's the point. The machine raises the floor. The humans set the ceiling.
Worth saying plainly: we don't have all the answers. We're feeling our way through this, like everyone else.
When is it right to use AI-generated images, and when isn't it? What belongs in our AI privacy policy, and what client data should never go near a model? When does using AI need to be declared, and to whom? We've got firm positions on some of these and open questions on others. Anyone who tells you they've got it all settled is either selling something or not paying attention.
So we'll keep working it out, and we'll keep being open about where we've got to.
Back to that dirty secret. Everyone's using AI. Nobody will say so. In a relationship built on trust, that's daft. So we're saying it. We use AI. Here's roughly how. We won't walk you through every prompt, partly because this is an article and not a manual, but mostly because the how matters less than the why. We're telling you because we think it gets you better work. Not quicker work. Better work. AI clears the tedious so our people can go deeper. It never touches the judgement you're paying us for.
You'll see it in the work. Three concepts chewed properly rather than eight churned out. That's what we do with the time AI gives back.
If you're looking for an AI strategy, then start with a brand good enough to hand to a machine.