Bringing AI Into a Creative Project Without Ruining It

Ask AI to help with a creative project and the easy failure mode is asking it to generate the idea itself: give me a concept, a plot, a hook. What comes back is often technically competent and quietly generic, drawing on the most statistically common version of whatever you asked for, because a wide-open prompt has nothing specific to work from and defaults to familiar territory.

The more useful use isn't asking AI to originate the idea. It's bringing it something already half-formed and asking what's actually working in it, and what it's avoiding, a reactive use that keeps the actual creative decision where it belongs, with you, rather than handing the starting point over to a statistical average.

What does AI actually add to a creative project, and what does it flatten?

Bring AI in at the very start, asking it to generate the concept itself, and you'll typically get something competent and forgettable, a premise assembled from whatever's most common in the space you described. It isn't wrong exactly. It's also not particularly yours, because nothing specific to you was in the prompt for it to react to.

The flattening isn't a flaw unique to AI generation. It's what happens whenever a creative starting point comes from the statistical middle of a category instead of from a specific, particular point of view, human-generated brainstorming under time pressure has the same failure mode, producing the same safe, familiar shortlist a room full of people under deadline pressure tends to converge on.

Is AI actually useful for creative work, or just fast?

Unaided, it's genuinely useful once you bring it something specific: a rough draft, a half-built scene, an actual constraint you're working inside. Reacting to something concrete is a fundamentally different task than originating from nothing, and it's the one AI is actually good at.

It's weak at supplying the specific point of view or lived experience that usually makes creative work distinctive rather than merely competent. It can generate a wide field of options fast; it can't tell you which one is actually yours to make, because it has no personal stake, memory, or accumulated history to draw that particular judgment from, however plausible its answer sounds.

What's worth bringing to AI before you start?

These four prompts run in Claude, ChatGPT, or Gemini as written.

Why it works: asking for what's working first, before critique, surfaces what's distinctive about your specific version rather than immediately pulling it toward a more generic, safer shape.

Why it works: this tests a specific element against removal rather than against taste, which catches load-bearing choices you might otherwise cut simply because they currently feel unfashionable or overly bold.

Why it works: a specific constraint reliably produces more distinctive work than an open field of options, and generating several lets you pick the one that actually creates useful friction for this project.

Why it works: multiple divergent directions break a stall better than one confident-sounding suggestion, because the point isn't to accept the first idea, it's to see the shape of the actual decision you're facing.

What's missing even after a good session?

Run all four and you'll likely get unstuck faster and see your own draft more clearly. What you won't get is the specific point of view that makes creative work worth making in the first place, the thing that comes from your own accumulated taste and experience, not from a single session of well-directed prompting, however productive that session felt in the moment. The gap isn't this project; it's that no tool, used however well, substitutes for the slower work of developing your own eye over years of paying attention.

How does Rick Rubin's method actually change how you use AI here?

Rick Rubin's documented practice, described across interviews and his own writing on creativity, treats stripping back as an active discipline: remove an element and check whether the work actually gets worse, rather than assuming more is always safer or that every addition is automatically valuable. He also documents listening, genuinely, without immediately reacting or fixing, as a distinct first step before any editorial judgment gets applied.

Before, generic prompting: asked to improve a rough song demo, a generic AI response suggests additions, a bridge section, more instrumentation, a bigger chorus, treating more elements as the default direction for improvement.

After, Rubin's framework applied: the framework starts by listening first, describing back what the piece is actually doing without proposing changes, then tests specific elements against removal one at a time rather than suggesting additions by default. If a section works fine without the extra instrumentation, the documented verdict is to cut it, since the test is whether removing it damages the piece, not whether adding more would theoretically be interesting. The direction of the question, what can come out, rather than what should go in, changes which choices actually get made.

A generic "how do I improve this" prompt has no built-in reason to bias itself toward subtraction, or toward listening before reacting at all. The framework does, structurally, rather than offering another round of additive suggestions.

How does this framework actually reach Claude, ChatGPT, or Gemini?

Claude installs it as a Skill: Settings → Skills → Add skill, upload the .zip. It auto-invokes once your question matches the skill's description, or gets forced with /rick-rubin-framework at the start of a message. In Claude Code, the package sits in ~/.claude/skills for every project, or a project's own .claude/skills folder for just that one.

ChatGPT has no Skills equivalent, so the plain .md becomes a Custom GPT's Instructions: Create a GPT → Configure → Instructions, pasted in directly. Skip the default Custom Instructions fields, they cap at 1,500 characters; a Custom GPT's Instructions field runs to roughly 8,000, enough for the full framework.

Gemini has no dedicated upload option, so the content goes into the system prompt if your interface has one, or opens the first message instead: "Here is a thinking framework to apply throughout this conversation. Read it carefully, then answer using this framework's approach," with the framework text pasted underneath.

Where should this take you, without a specific tool attached?

Rubin's documented method is one of several in the Creative Visionary category, alongside Walt Disney and Frida Kahlo, each an .md file for Claude, ChatGPT, or Gemini, worth exploring as an ongoing practice rather than a one-time prompt, since a creative project rarely stops needing this kind of attention after just one session.

All the copy-paste prompts from this guide, and the rest of the AI Practice series, live at /prompts too, free to copy without reading the guide first.