Stop prompting. Start directing.

Before we begin, a quick note of context. I’m writing this with ChatGPT in mind because it’s the tool I use most often. From lighter testing across other models, the same principles largely apply. The differences are interesting, but the way you work with them doesn’t change all that much.

Why feedback matters more than prompts

Most people still approach AI as if it were a vending machine. They type in a prompt, wait for a neat answer and feel faintly disappointed when it doesn’t match the imagined masterpiece in their head. Anyone who has ever worked in a creative team knows that’s not how good work happens. The first draft is only the beginning. The value emerges once you start shaping, guiding and questioning. AI behaves in exactly the same way. It is far less like a search engine and far more like a studio assistant who improves rapidly when you explain what you actually meant.

This builds on something I wrote previously about AI expanding the creative process rather than replacing it, and this idea of treating it as a collaborator continues that thread.

Over the past year I’ve ended up using it as a quiet studio presence, a patient assistant who doesn’t mind producing fifteen variations before lunch. What’s most surprising is how coachable it becomes. When you offer clear, specific feedback, it adjusts with the same instinctive willingness you’d expect from someone still learning the craft. It’s the same principle I touched on in my piece about working with AI in my day-to-day creative process.

How to direct an AI like a studio assistant

The simplest way to get better work is to describe what you felt rather than issuing blunt commands. If you say “make it shorter”, it will dutifully trim the words and keep the same underlying problems. If you say “the beginning is strong but it becomes foggy in the middle”, it understands the intention behind the edit and reshapes the work with purpose. It’s the difference between mechanical editing and meaningful improvement.

It also helps to acknowledge what is working. A quick “the tone in that paragraph feels right, keep that” tells the model what to protect. You anchor its instincts in the same way you would with a real studio assistant who needs to know where the energy already sits.

The next step is to diagnose the issue rather than rushing to prescribe the solution. Instead of saying “change the metaphor”, try “the metaphor is clever but pulls attention away from the point”. You’re giving it the room to rethink the idea, not simply obey an instruction. That’s where stronger work tends to emerge.

Explaining why something doesn’t quite land is equally powerful. If you say “too corporate for this audience” or “this reads a little American, let’s make it more British” or “the sentiment is right but the pacing feels too tidy”, the model absorbs the logic. The next version usually arrives closer to what you had in your head, even if you didn’t articulate it perfectly the first time.

Watch for the AI clichés

One habit worth mentioning is how much AI loves neatness. It has a natural instinct for smoothing edges and ending with tidy moral wrap-ups: “it’s a reminder that…” or “ultimately, this shows…”. Once you spot it, you start seeing it everywhere. The good news is that it’s teachable. A simple note such as “avoid the neat conclusion; leave it open” changes the behaviour almost immediately.

These clichés aren’t flaws, they’re habits, and habits can be trained out. This expectation can be built into your workflow. Let the model know upfront that you don’t want the moral, the synthesis or the essay-style conclusion. It will follow the brief. It simply needs to know it exists.

The same realism applies to creativity itself. Asking “give me some taglines for this project” rarely delivers anything remarkable because the model is drawing from existing patterns in language. The spark still has to come from you. The machine’s job is to stretch, refine and pressure-test your thinking, not to replace it.

A simple way to work

A three-stage rhythm tends to work well across most creative tasks. Let the first pass be loose and exploratory. Use the second for clear direction about what landed and what drifted. Use the third to tune the cadence, strengthen the transitions and bring the whole thing into focus. It’s a gentle, reliable way of turning rough sketches into something considered.

Working like this makes something very clear. AI hasn’t reduced the need for human judgement. It has made that judgement more visible. Your sense of tone, pacing, ambiguity, voice and truthfulness becomes the anchor. The model simply amplifies it.

In a way, AI has clarified the creative director’s role rather than threatened it. You provide the taste, the cultural awareness and the instinct for what feels right. The model provides speed, range and the willingness to take notes endlessly without complaint. The meeting point between those two is where the work becomes interesting. Not in the prompt and not in the machine, but in the quiet craft of feedback.