​

/o

This week, I helped a client set up the context they need to write marketing copy in Claude without coming back to me for every draft.

Which, when you make a living from messaging and copywriting, might sound like a questionable business decision.

I’ve been trying to do some version of this for the past two years. I want clients to do more with the work we’ve done together, whether that means writing a campaign, briefing someone new, or adapting their message for a different audience. Having to ask me what we meant every time they use a messaging framework would be a pretty bad outcome of a messaging project.

So yes, I’m trying to make myself obsolete. At least for the parts of the work where you shouldn’t need me anymore.

What I’m finding is that my understanding of AI makes the research, positioning, and messaging work more useful to clients, not less necessary. We can now put more of that understanding into tools they already use, instead of leaving them with a set of documents and hoping they know what to do next.

That’s the type of abundance I’m interested in. You get more use out of the thinking you’ve already invested in, rather than just more stuff to publish.

Help your team use the strategy after the project ends

In one agency project, we turned the founder’s understanding of the business into messaging his team could use without him repeating the explanation. In another project, for a nonprofit, the team fed our messaging into a custom GPT to help them write consistently across different audiences.

Neither project started with an AI tool. We first had to understand the business, the people it served, and what those people needed to hear. The useful part was making that understanding available beyond the project itself.

Now I’m packaging that idea into what I call Messaging Kits: small, purpose-built messaging brains that help a client write for a specific channel, asset, or launch. Each kit contains the relevant company knowledge, examples, and instructions for using them in a tool like Claude or ChatGPT.

Build Messaging Kits from one company brain

Behind each kit, I keep a much richer body of knowledge for the client: research, customer language, positioning decisions, messaging, approved copy, feedback, and the reasoning behind important choices. That’s what I mean by the company brain. It gives my agents and me somewhere to return when we need to understand why we chose one direction over another, or whether something has changed.

That depth is important when we’re making a strategic decision. When you’re writing an ad, though, you need to know which buyer you’re addressing, which promise should lead, what proof supports it, and what you want them to do next. You shouldn’t have to sort through old workshop notes and rejected headlines to find those answers, or keep explaining to the model which version of the offer to use.

A Messaging Kit makes those choices explicit before you start writing. For this week’s client, we selected the approved messaging, the proof they can use, examples of their voice, and actual human edits that show how to improve the writing. We included instructions for applying those materials and proposed ad examples, clearly separated from approved copy.

The advantage isn’t that a smaller collection of files automatically makes an AI smarter. It’s that you give it the context for this writing job, rather than asking it to choose an audience, message, and proof from everything the company knows. You also have a much clearer basis for judging what it writes. This simplicity is especially important if you’re working with simpler AI chat tools, rather than AI agents.

You might be thinking: doesn’t that leave me with several versions to maintain?

That’s why I’d keep one company brain as the source and build the kits from it, rather than maintain each one as a separate strategy. When you approve a change to an offer, you can ask an agent to compare that change with the kits, identify the affected passages, and prepare the updates. You review those changes and replace the old files in the relevant projects. Claude or ChatGPT won’t automatically synchronize separate uploads for you, but AI can take on much of the checking and rewriting that would otherwise make this a chore.

Give each Messaging Kit a specific writing job

You could have one for a launch, one for a particular channel, or one for a recurring asset like your sales emails. They should draw from the same positioning and company facts, but carry the extra context that makes each job different.

Say you’re launching a product to existing customers. Your kit needs to explain what’s new, who can use it, how it relates to what they already have, and what you can honestly promise. It should also help you address the objections that could stop them from trying it: will they have to migrate their data, pay more, or change how their team works? A kit for reaching people who’ve never heard of you needs more help establishing the problem and explaining why your approach deserves their attention.

Same company. Different starting point for the reader.

You can put that context in a project inside a tool your team already uses. Then they can just ask for the next piece of copy.

Show the AI how to apply your messaging, not just repeat it

This is where understanding positioning, messaging, and copy still matters.

Positioning establishes who you’re the right choice for and why, compared with the alternatives. Messaging turns that into the ideas, promises, and proof you want buyers to understand. Copy expresses those ideas for a particular person, in a particular situation.

When I guide my own agents, I use custom skills, which are reusable instructions for how to approach a particular job. Mine make the agents work through the relevant context before writing. For the client’s Claude setup, we translated that approach into project instructions.

Those instructions tell the model what to treat as the current message, how to use the voice examples, and what to do when a request goes beyond the available proof. If someone asks for a stronger claim than the evidence supports, the model should flag the gap and offer a useful alternative, not just make something up.

We also include human edits because “confident, conversational, and clear” can mean almost anything. Showing a before and after, with an explanation of why the change helps the reader, gives the model something more concrete to work with.

None of this makes every draft good. Someone still has to read it, check it, and decide whether to use it. But you no longer have to rebuild the company’s messaging from scratch every time you open a chat.

That’s my goal when I say I want to “replace” myself.

If your team can apply the strategy without me sitting beside them, we can spend our time on the questions that actually need attention: whether your buyers have changed, whether a new offer fits the story, or whether the data still supports your message or we need to adjust it.

You don’t need to build a huge company brain to try this. Pick one recurring piece of marketing, gather the current messaging and examples it needs, and write down how someone should use them. Then have a teammate try it. Bonus points if it’s someone who’s got no marketing or copy experience.

Whatever they still need help with or wherever their messaging is not hitting the key points it needs to hit, that’s where you then go and fix your messaging kit.

Discovery

Can your marketing agent handle the job again next week?

​Hiten Shah’s “Why Marketing Is a Good Test for AI Agents” gets at something more useful than generating another ad or blog post. So much of marketing happens between tools: connecting what a customer said with what they did, checking whether the website still makes the right promise, and more.

That’s where agents get interesting. But Hiten’s test is whether he’d give an agent the same job again next week. Does it remember what it already reported? Can it stay quiet when nothing important changed? Does it stop when a decision needs you?

I like that because a good first run can hide how much work you still have to do yourself. Read it with one recurring marketing job in mind, and ask whether an agent could take over the follow-through, not just produce the first output.

Join us next Friday: early user discovery without a research team

On October 9, I’m joining Christopher Richards from UX Brite for From Guesswork to Signal: Early User Discovery Without a Research Team.

We’ll talk about finding useful patterns in customer conversations and behavior when you don’t have a research team, a big budget, or much data. Then we’ll work through how to turn what you learn into clearer messaging and product decisions, including where AI can help without mistaking simulated responses for real customer evidence.

If you’re building something new or trying to understand why early users aren’t getting further, join us live and bring your questions.

Resonance