How to Make AI More Creative: 4 Keys to Better Prompting

When I talk about AI creativity, it’s important to start with how most people use Google. When people first start prompting AI tools, they treat them like Google for the most part, sharing short prompts and expecting magical results. For brainstorming and creative results, magic is certainly there sometimes. 

But I find that if you don’t put more energy into your prompting, the creative results from AI often feel more like an interaction with a vending machine. You hope to find something magical, but all the options are things you’ve seen before, some of them a bit old and out of date. 

That’s not how creativity works, and it’s not the best approach for achieving the best creative outputs from large language models.

I’ve been taking some of the frameworks from my creativity book, The Case for Bad Ideas, and testing them as AI prompts for marketers. Through my testing, a few key concepts are clear keys for getting more creative results from AI tools. 

There are 4 keys we’ll review:

1. Ask for pattern breaking and variety
2. Detail how the experts would do it
3. Define an iterative feedback loop 
4. Require a strong opinion

Let’s break down each of these, using a few of my creativity prompts as examples. 

Ps. You can steal 5 of my AI creativity prompts for free over on my creativity newsletter.

1. Detail how the experts would do it

Role prompting is one of the simplest ways to improve results from an LLM. When you tell the model that it is a creative director, a headline copywriter, or a campaign strategist, it snaps into a more relevant tone, format, and decision style. 

But that’s just step one. 

The next level up, which I call process prompting, is telling the model how that expert works. This could be a simple outline of the simple steps that type of expert typically follows, a list of constraints they typically break, or what outcomes they aim for. 

I’ve done this for two of our AI creativity prompts so far with good results. 

For example, with my Messy Ad Writer prompt, I include 15 different ad hook writing templates that copywriters draw from. This is not an exhaustive list, but it’s one of the resources our agency team uses when writing ad copy. 

The results are a big leap up when you spell out a process. You’re essentially starting to build a workflow. When you embed creative process into the prompt, the model stops guessing and starts behaving like a real collaborator. 

The extra bonus here is you now how a process to optimize. Our Messy Ad Writer has 15 ad hook templates or methods, but you can research and add dozens more. You can update the prompt to not overuse certain templates or to match templates to use cases, etc. 

Give your AI prompts a process and continue to adapt each time you use it.

2. Define an iterative feedback loop 

Hemingway famously rewrote the end of his novel A Farewell to Arms 39 times. Great creative thinkers iterate. They obsess over feedback loops. 

This also happens to be one of the best ways to work with AI on creative tasks as well. You can set up an AI tool for success with background info, personas, an expert role, and a process, but it still does not know your taste. 

A great marketing use case here is naming. 

Naming is a brutal creative task. It’s emotional. It’s subjective. And everyone has strong opinions. Our agency has done several brand naming projects. And with each one, we will typically generate a list of at least 1,000 names over multiple feedback rounds with a client, typically lasting months. 

AI is also well-positioned here because it can help you scale ideas quickly. 

My Collision Name Lab prompt isn’t just a name generator. It’s a training system. Because naming is so personal, it optimizes for quick feedback rounds that allow you to review lots of ideas and name winners each time, influencing the next set of ideas you’ll get back. 

Each round fits 20 distinct name options on the screen, so you can quickly review them all. It makes sure every name is numbered, so you only have to type the numbers of your favorites as feedback. You can still give longer form text feedback, but the numbered winner feedback means you can go through hundreds of name ideas in minutes. 

Not every creative or marketing task needs a thousand unique options in the consideration set, but almost every creative task needs iteration and feedback. Build these loops into your prompts more, and the AI tools you use will better understand the creative output you need and optimize for it. 

3. Ask for pattern breaking and variety

Brainstorming in real life works best when you include divergent thinking, pulling from a larger variety of ideas and concepts that break patterns from what’s been done before. The problem here is that LLMs default to being prediction engines. Meaning they are taking educated guesses at what the next word or phrase or pixel should be based on their training. That’s not great if you want results that stand out. 

Think “how can I blow up the way AI wants to think?” How can you get it to turn down dark allies or jump into the deep end of the pool? To take more risks?

I’ve got a lot of these concepts in the book that work well here, like simply asking it for more bad ideas. Here’s an example section of my Bad Idea Campaign Generator prompt.

I’m requiring ideas that are obvious, absurd, or risky. At least one idea has to be physically dangerous. And many of these outputs will feel totally unworkable. But you know what? That’s the point.

Creativity lives in contrast. You don’t get surprising campaign ideas by staying inside the box. You get them by *playing outside* of it, and this prompt gives you a map of the weird and the wrong to explore what might actually be right. 

It’s designed to trigger you into thinking in new directions. Find ways to require the AI tools you’re using to help you find those paths less traveled.

4. Require a strong opinion

Most AI-generated feedback is polite. It hedges. It avoids strong opinions. That’s not helpful when you’re trying to make something that stands out.

A common knowledge block that’s helpful to include in prompts or add to the memory of your AI tools is a customer persona. But, there are often many additional personas that are helpful here. My Collaboration Council prompt is designed to give you multiple sources of opinionated, creative feedback for any project. 

By default, this prompt defines three experts who have strong points of view to share.

• The Tastemaker: Minimalist, brand-first, clarity-obsessed
• The Artist: Emotional, expressive, format-bending
• The Skeptic: No fluff, raw POV, truth-first

When you define these feedback personas, you’re creating a team that has permission to quickly stress test the strategy and scope of your outputs. I’ve used this with ad creative, website projects, and bigger marketing strategies. It’s great for challenging your assumptions and finding gaps in your thinking. 

I encourage you to create prompts that define the opinion or point of view that would most help your outputs. 

Better AI creativity requires a contextual system

We all start using AI tools, but with less context, shorter and direct asks. One shot, one answer. That’s not how creativity works. And it’s not how humans work, either.

These 4 tactics have helped me uplevel creative results from AI tools for me and my clients. As you work to improve AI marketing results, consider how you’re building in process, iteration, pattern breaking, and opinion. Collectively, these will start to give your AI tools something they usually lack: a creative process.

If you want to steal my 5 creativity prompts, you can get them free on my newsletter here.

And if you want to see them in action (how they work, why they work, and how I use them) check out the conversation I had with the CMO of HubSpot about how they work on the Marketing Against the Grain podcast here:

Let’s stop using AI like it’s a slot machine and start training it to be a real creative partner. Let’s go.

The post How to Make AI More Creative: 4 Keys to Better Prompting appeared first on Convince & Convert.