I cloned a game-marketing expert into an AI skill. The trick was priming, not prompting.
TL;DR: I’m a developer, not a marketer, and my game needed marketing help. So I built an AI advisor that thinks like Chris Zukowski, the indie game marketing guy. I didn’t write a clever prompt. I primed three chatbots with a few high-level questions, let them research and ramble, pasted the full output into three files, and had Claude Code pull the best bits into a reusable skill. The technique is priming. Ask big questions, let the model think out loud, don’t even read it all, then compress what comes out into something you can run again. And it’s never finished. You keep priming it with what’s changed and have it rewrite itself, so it’s current the next time you open it.
The problem
I can build the game. Marketing it is the part I’m worst at. Thirty years writing software and I still can’t position a product to save my life.
I’m lucky to have a good friend helping me. He’s spotted bugs and rough edges in the early game I’d have sailed straight past, and he’s put together SEO reports that changed how I think about getting found. But his time is limited and mine moves fast. I can’t park a decision for a week waiting on a favour. I needed something I could reach for at any hour, that I’d never feel bad for leaning on.
So I built a second brain I could argue with at midnight. One that would tell me hard truths about my game without being polite about it.
What I actually did
The whole thing took about an hour, most of it while I was doing something else. The sequence:
- Opened ChatGPT and asked, plain: who is Chris Zukowski? The “How To Market A Game” guy. Let it answer in full.
- Gave it my game. Here’s Inselnova, here’s what it is, what would he say about it? Let it run.
- Asked it to find three more people who think like Chris Z about indie game marketing, and what each would say. Let it run again.
- Ran the same line of questioning through Gemini and Claude.
- Made three markdown files, one per model, and pasted the full output into each. Didn’t tidy a word.
- Handed all three to Claude Code: read the lot, find where they agree, keep the sharpest bits, turn it into a skill I can reuse.
Each question primes the next. By the third one the model isn’t answering cold, it’s reasoning inside a frame I built one step at a time. The three extra experts weren’t really the point either. They were fuel, more context to push the thinking somewhere richer. What came out the other side is a marketing mentor I can invoke any time I want a gut-check. A real skill in the repo now, not a chat I have to dig back up. Blunt, opinionated, and it argues with me. Exactly what I wanted.
Priming beats prompting
Most people open a chatbot, ask one narrow question cold, and get a flat answer back. Then they blame the model.
Priming is the opposite. You spend a few turns loading context and making the model think out loud before you ask for the thing you actually want. A few things that matter:
- You don’t need the perfect prompt. You need a good first question and a few honest follow-ups. The quality comes from the conversation, not one magic incantation.
- You don’t have to read all the output. That sounds lazy. It’s the point. You’re generating raw material, not a final answer. Claude Code reads it so you don’t.
- Running it across a few models widens the net. Each one phrases things differently and surfaces a few points the others miss. You’re not looking for a winner. You’re gathering more good material to compress.
The bit that feels wrong but works
You end up with a pile of text you barely skim. It feels wasteful, like you’ve burned an hour generating words nobody will read.
But the words were never the asset. The skill at the end is. The chatbots did the research. Claude Code did the reading. I did the orchestration and kept the one thing worth keeping. The chat was just the mine. The skill is the metal.
What the mentor actually told me
It didn’t flatter me. The thing it pushed hardest was that I’d been aiming at the wrong audience. I’d been writing for casual gamers, watering the game’s edges down to make it approachable. The mentor said start with the original audience first, the people who already love this kind of game, and come back to the casual crowd later.
I agree with it. The people I talk to on Inselnova already know the genre. They came looking for it. So the copy should lean into that, not hide it.
There’s a trap in that, and I’m watching for it. The players who reply to me are the ones who stuck around, and they’re the genre fans. The ones who churned might be the casual players who never felt at home, and they didn’t answer the survey to tell me so. The loudest signal isn’t always the whole picture. I’m leaning into the original audience anyway, but with my eyes open about who I’m not hearing from.
More than one mentor
The marketing advisor isn’t the only one I’ve built this way. I made another that thinks like Rory Sutherland, for wording and the psychology of how things are framed.
That one taught me the limit of the trick. For wording, it’s great. When I let it push on actual game design, it sometimes led me down the wrong path, suggesting loops that sounded clever and didn’t work in practice. So I use these mentors for what they’re good at. Positioning, phrasing, the way a thing is sold. Not for deciding how the game itself should work. That call stays mine.
Keep the skill alive
A skill like this goes stale. Marketing advice that’s right today is half-wrong in three months, and the game keeps changing under it.
So I don’t treat it as finished. After a real session working with the mentor, the last thing I do is tell the agent: “now update the marketing skill based on our conversation. Remove what’s wrong, update what’s changed, add what’s new, but keep it slim and to the point.” Next time I open it, it’s current. The skill learns from the work instead of drifting away from it.
Then every few weeks I find another games-marketing person, make a document, and run the whole priming process again to fold them in. The skill isn’t a statue. It’s something I keep sharpening. That’s the part people miss about skills. They’re meant to evolve.
And the priming doesn’t stop at creation either. A week on, I’ll open the mentor and start with something like: “review last week’s git logs, the two posts we shipped, and the update to the Inselkampf page.” That primes it on what’s actually happened before I ask it anything. Once it’s caught up on the context, then we get to work. More priming. Always more priming.
What I’d hand over
Stop hunting for the perfect prompt. Prime instead.
Ask the big, almost dumb questions first. Let the model think. Run the same line of questioning across a couple of different models. Then have Claude Code compress the whole pile into something you can use again next week.
And don’t let it set hard. Keep priming it with what’s changed, have it rewrite itself at the end of a session, and add a new voice every few weeks. The skill is the asset, and an asset you keep sharpening beats a clever prompt you typed once.
Inselnova is a free browser strategy game built solo with AI. Play here.
