Using Custom GPTs for Game Design

4 minUTE read

How private, tailored AI can enhance your game's development by using your own data for unique, creative insights.


By Justin Eames | CEO

As the games industry increasingly embraces generative AI, there remains some fear and anxiety. A major concern centres around copyright issues, particularly the challenges associated with prompts whose responses have, at some stage, been processed using data that belongs to others.

However, this concern overlooks the huge potential of AI models that are ringfenced to utilise your own data. With the rollout of new AI features from platforms like OpenAI’s ChatGPT, we now have the abilty to create siloed, custom GPTs – versions of ChatGPT that are private and tailored to our specific needs, using our own data.

This development opens up possibilities in game design and team communication, enabling us to be more effective within the bounds of our own creative universe.

Let’s explore.

Why custom GPTs?

You might wonder, “Why bother with a custom GPT when I can just use the standard ChatGPT for game design ideas?” Well, there are two compelling reasons:

Using your own data

The standout feature of a custom GPT is its ability to absorb your own data. Think spreadsheets, PDFs, images, and more. This integration means your GPT isn’t just producing generic responses; it’s drawing from your specific game world.

Ring-fencing responses

Ring-fencing is about shaping the context in which the GPT thinks and responds. It’s tuning its bias to serve a specific purpose. This feature helps ensure the GPT’s responses are not just relevant but also consistent, keeping your team aligned with the game’s vision as your game evolves.

Let’s explore some ways in which we can use a custom GPT in our games studio.

Game design bible as a source of truth

A game design bible is the cornerstone document outlining your game’s world and mechanics. By feeding this to a GPT, your team can ensure their designs and implementations align with this bible. It can be a simple task like listing objects in a scene or as complex as scripting a cutscene. The data that you’ve built to define your game universe may be in written documents, spreadsheets and images. You can feed the GPT all of those and keep updating them over time.

This was a big find – had we have launched with the game set at its current difficulty level, kids would have found the game frustratingly hard. Having this insight continued to be useful throughout the remainder of development because as we all got more used to the game, the easier we found it, and had to resist the urge to make it harder.

Designing style with ring-fenced responses

A picture says a thousand words and a GPT can help designers generate art to illustrate their ideas (but not destined for the final game) in a specific style, like pixel or illustrative. Ringfencing helps the GPT understand and stick to these stylistic choices.

A game character

A pixel art cat

That goes beyond artwork. Feed the GPT style guides and writing tone documents and it will help your team write scripts and dialogue that aligns to the style and tone of the game. It doesn’t have to generate them itself, but it can certainly check what you’ve written and help improve it.

A game character

Marley’s Ghost

Game design archives

For a studio like fish in a bottle, with over 20 years of design documents, a GPT can create a treasure trove. It helps us sift through what’s worked (and what hasn’t) to inspire new game concepts and refine our pitches.

Creating age-appropriate game designs

By feeding the GPT data about children’s cognitive and motor skills and our findings from user testing, we can craft games perfectly suited for specific age groups. This approach ensures our games are not just fun but also developmentally appropriate.

Have a go with our example custom GPT

We’ve created a Custom GPT that you can experiment with. At the time of writing, you’ll need a paid subscription to access it (sorry).

A Christmas Carol Game Design Tool is a custom GPT that’s been fed a small amount of game data for a point-and-click adventure game based on Charles Dickens classic book, A Christmas Carol. It knows about locations, characters and interactive objects as well as style, tone and appropriate puzzles.

A game image

A scene mock-up from A Christmas Carol point-and-click adventure

You can try prompts such as:

“Draw a profile picture of Marley’s Ghost”

“Design a puzzle for Scrooge’s Counting House”


“List the interactive objects in the Cratchit Family Home”

Want to build your own custom GPT?

To create or use a custom GPT using ChatGPT you currently need a paid subscription. Making them is very simple and requires no code, although you will find there are limitations. The GPT doesn’t always respond in the way you’ve set it up to. It also has a habit of inventing new data if it can’t find what it’s looking for in the data you’ve fed it.

The other thing to be aware of is you’ve no absolute assurance that when it’s generating content for you (scripts, images etc) you’re going to be in the all-clear regarding your company’s AI and copyright policy. So that’s one to watch out for.

However, using a custom GPT is a great way to bring teams together around data, and if you are careful about how you use its output, you can benefit with little risk of running into copyright problems.

Are you looking for an experienced games studio?

Game changers logo

Are you a game changer?

Join the free community for people interested in making the world better through playable experiences.

Request to join

More knowledge...

A 1980's style toy robot


Transformers in AI - a simple explanation

Likely to become as revolutionary as the combustion engine. But what are transformers?

An illustration of a person lifting weights with their brain, implying their brain is working out


The Future of Games Based Learning

How game-based learning will transform education and professional development.