How I use ChatGPT to make CORPO

Taking on the work

ChatGPT helps me take on more of CORPO as a solo creator. I use it to work through ideas, then keep checking what those ideas become.

I work on CORPO during breaks in my workday, at night and on weekends. I think about it in my free time. These days, I’d rather spend that time creating than watching Netflix, playing Call of Duty or scrolling. I don’t use social media anymore.

As of writing this article in October 2026, I’ve been working on CORPO for 18 months straight. I’m still learning as I go.

Learning new technologies, software and tools, then creating something with them, is the most engaging part for me. It makes me deeply happy. I want these articles to show the work behind making a game with AI, including what I’m learning as I do it.

I use ChatGPT for research, writing, production planning and work on the game’s website. I also use Codex for development tasks. They’re part of the process alongside Lovable, the art references, the lore and a growing collection of decisions that need to survive the next conversation.

The useful starting point is the material I’ve already made. CORPO has characters with established appearances, a world guide, approved illustrations and specific game rules. If I ask for a new section of the wiki, those are the references. I don’t want a convincing description of some other science-fiction game.

Keeping the decisions

I keep written standards for the parts of this project that are easy to get wrong. There’s guidance for my writing voice, CORPO artwork, dialogue, quest books, battle passes and motion. They give the next task something more useful than “make it premium.” They explain what that means here.

The current game instructions are quite specific. The Book needs to preserve the player’s next mission. A reward reveal needs a readable pause. A character’s face needs to match the approved reference. The same character shouldn’t suddenly talk like every other character. Those instructions can be applied to individual pieces of work and checked against the result.

ChatGPT projects can keep related files and instructions with the work. For development, I also keep guidance alongside the source files. I still make corrections when something changes. An old document doesn’t get the final vote over what I’m asking for now.

Be specific about what is wrong

This website is a good example. I asked for a wiki that felt graphical, light and easy to explore. The first versions still had things I didn’t want: decorative lines, dot-separated labels, generic “story brief” treatments and a logo that wasn’t the current CORPO mark.

Saying “make it better” leaves a lot of room for another version of the same problem. So I pointed to the actual things that needed to change. Browse should be called Wiki. The logo is the Tower, not the old suit-and-tie mark. The interface should borrow from the game. And the Making CORPO articles should explain what I’m doing, in my voice.

That applies to the copy as much as the layout. I don’t want my own website to sound like a software company describing an imaginary customer. I want to explain how I’m making the soundtrack, how the world fits together, and what’s involved in getting this game built.

Building this wiki

For discoverability, I asked for research across search engines, answer engines and generative AI search, along with separate work on game discovery, wikis and blogs. That material became a reusable discoverability skill and a plan for CORPO. The point is to carry useful findings into later work instead of collecting a report that never changes the website.

The wiki is one practical result. It brings together 113 illustrated entries and 47 glossary terms from the source library, with search, categories, related reading and full-image views. That gives people several ways into the world, whether they arrive through the game, a character or a question about something in the lore.

Those entries still need to stay aligned with the current game. Some source material comes from earlier versions. A large library is useful only if I’m willing to correct the names, relationships and descriptions that no longer match.

Checking the result

I want the work checked where somebody will use it. For a page, that means reading it at phone size, opening the menus, following the links and looking for things that collide. For a game change, it means checking the player’s action and the state that comes after it. A clean report can’t substitute for that.

I’m asking ChatGPT to do substantial work on CORPO, and I’m asking it to revise that work when the writing sounds wrong or the design doesn’t belong to the game. That has meant going back over this wiki several times. The Tower logo and the name Wiki might seem like small corrections, but they were wrong, and I wanted them fixed.