Skip to main content

I taught ChatGPT to teach me board games, and now I won’t ever go back

Teaching complex board games seems like the perfect task for ChatGPT. Being able to ask it specific questions is a whole lot easier than flipping through a rules manual or digging through forum posts online.

And after weeks of effort and a healthy dose of prompt engineering, I have created the ultimate board game rules lawyer with ChatGPT. It can “learn” any game you throw at it, base games and expansions, and can answer nuanced questions accurately, with rulebook references so you can check its work.

Recommended Videos

It’s no substitute for knowing the game inside and out from playing it, but for learning a game, or double-checking a tricky rule, I won’t be heading to game night without my very own board game chatbot.

The rocky road to competence

I’ve been trying to get ChatGPT to learn board games for a while now. The problem I always ran into was that as much as ChatGPT might have been trained on untold quantities of data, often that didn’t include the minutia of board game rulebooks.

Although it could speak about some rules, I frequently ran into issues of hallucination, rules it just wasn’t aware of, or misunderstandings in my questions when I didn’t use the very specific wording from the game.

Asking ChatGPT more board game questions.
Ergh! Wrong. Wrong! Image used with permission by copyright holder

In the above example, I asked a very specific question about the popular cooperative game War of the Ring, which just happens to be one of my favorites. Although ChatGPT has got the gist of the rule, it’s completely wrong. The “hunt dice” have nothing to do with how many “Gandalf characters” there are in the fellowship. There can, of course, only ever be one Gandalf. Even if the Maiar are many.

These problems meant that at best, I couldn’t truly trust what ChatGPT was telling me, and at worst, it could be completely misleading.

But the idea of using ChatGPT for board game rules was too enticing for me to give up entirely. I bit the bullet and tried paying the $20-a-month subscription for GPT-4 access through ChatGPT Plus so that I’d have the most cutting-edge chatbot possible to take on the board game rules lawyer persona. I tried using ChatGPT with web browsing capabilities — both official and Chrome extension driven. I even tried using Google Bard as a substitute for the whole thing.

None of them could quite get things right, though. Either they required too specific a question, or couldn’t get into the weeds — which is where board game rules can get tricky and require further clarification.

But then came my solution: ChatGPT plugins.

It’s all in the plugins

Plugins are a bit of a game changer for ChatGPT. They give it all kinds of advanced abilities by bringing to bear the interests and efforts of outside parties. They take what OpenAI made, and make it better.

If you’ve read our coverage on the best ChatGPT plugins to use, you’ll be aware we’ve been playing around with them for a while. But buried in my list of favorite plugins is a quick entry on one that finally brought my board game bot journey to a satisfying conclusion: ChatWithPDF.

ChatWithPDF is a plugin that lets you talk to your PDF documents. Why is that useful in this case? Because you can get just about any board game rulebook in PDF form. Sometimes they’re produced by the manufacturer, but you can also find rulebooks made by fans, rule summaries from the likes of Esoteric Order of Gamers, or you can even make your own if you have the physical rulebook to hand.

Once you’ve made the PDF available to ChatGPT and given it a quick prompt so that it has an “understanding,” of what you want from it, you can ask it any board game rules questions you like; And it works really well.

I’ve tried it on a handful of games, including War of the Ring (and two expansions), Battlestar Galactica (and three expansions), Leaving Earth (and both expansions), Twilight Imperium, and Kingdom Death: Monster. In each case it performed extremely well, answering questions in a nuanced, and clear manner.

Asking ChatGPT a complicated question about Twilight Imperium.
It worked really well after loading in the Twilight Imperium Living Rules PDF. Image used with permission by copyright holder

There were times when the ChatWithPDF plugin seemed to be unavailable, and no matter what I tried, it just would not respond, or throw up error messages about accessing the plugin. That was unfortunate, as there’s no indication from ChatGPT when and why certain Plugins might not be available, and there’s no planning for it. If you show up to the game night with your new game bot and it just isn’t working, there’s nothing much you can do.

Fortunately, most of the time it does work, and it works really well. The only real issue I ran into with ChatWithPDF was that it can only consider one PDF at a time. That meant that when I wanted to ask it questions about a board game expansion, whilst still having the context of the original game — and the ability to reference those rules if I wanted it to — then I needed to combine the PDF documents together.

Learning about War of the Ring dice with ChatGPT.
War of the Ring is a heavy game with a lot of rules that affect one another. BGL-GPT answers complex questions about it with ease. Image used with permission by copyright holder

Fortunately, there are plenty of ways to combine PDF files. You need to tweak the original prompt a little so that ChatGPT understands that these are different games, and you may have questions that relate to either, but ultimately it’s not too tricky.

How to make your own board game rules chatbot

If you want to bring along your own board game rules lawyer to your next game night, here’s how to do it.

You’ll need:

  • A ChatGPT Plus subscription.
  • A laptop, tablet, or smartphone for quick chat access at the table.
  • A single PDF document for the game you’re looking to play, including any expansions.

Step 1: Upload your PDF file to a cloud storage service. I used Google Drive, but you can use any online storage platform that you can share links to.

Step 2: Log in to OpenAI and start a new chat.

Step 3: Select the GPT-4 language model from the top of the screen. Then select Plugins.

Step 4: Select the small box under the language model selection window, then scroll down to the Plugin Store and select that. Use the search tool to find ChatWithPDF and select the Install button.

Note: There are other PDF analyzing plugins, but I’ve not experimented with them as much. Use them alternatively at your own discretion.

ChatWithPDF plugin in ChatGPT.
Image used with permission by copyright holder

Step 5: Close the Plugin Store, then, if you don’t see ChatWithPDF selected, select the small plugin box again, and toggle ChatWithPDF on from the list of plugins.

Step 6: Return to your cloud storage platform and find the PDF you want to use for reference. Make sure it is available to anyone with the link. This will differ depending on your chosen platform. Here’s how to share files on Google Drive.

Step 7: Copy the link to your PDF file, then return to ChatGPT and put in a prompt linking to it. Here’s the one I used that helps ChatGPT cut out some of its rambling and dodges a few common errors, but feel free to write your own. As long as it has the link to the PDF, it should work fine.

“Here is a single PDF comprised of [Number of] board game rule books. They’re for [Board game Name]. Learn them. I will ask you questions about them when I need to check a rule. I want you to be clear and concise. Do not guess or hallucinate answers — if you don’t know the answer, say as such.

When you answer rules-based questions, I want you to provide a reference for which rulebook within the PDF, and what page you found the rule on.

Do not reference yourself as an AI language model, do not apologize, do not add any extraneous information beyond answering the question and providing the rulebook reference.

https://drive.google.com/file/d/[examplefilename]/view?usp=drive_link”

ChatGPT comprehending board game rules.
Image used with permission by copyright holder

Step 8: Hit enter and wait for ChatWithPDF to respond. It can take up to a minute for it to parse large PDF files, but when it’s ready it will tell you that:

“The PDF containing the rule books for [Board Game Name], and its expansions, has been successfully loaded. You can now ask any questions about the rules of these games.”

Step 9: Ask it any questions you like, then wait for a response. It can take around 30 seconds for larger PDFs, but it’s usually less than 10 seconds before it starts to respond.

Improvements to make

My adventures in ChatGPT prompting have been quite varied, but it’s been legitimately useful to nail down a board game rule chatbot. I’ve used it to win a rule clarification race with my brother while he scanned multiple rulebooks, and I used it to teach me the complexities of slingshot maneuvers in the Leaving Earth expansion, Outer Planets.

ChatGPT is great at summarizing. Learning anything new from a PDF rulebook is much easier in bullet point form — and it can even create examples that aren’t half bad.

Learning slingshot moves in Leaving earth board game with ChatGPT.
Image used with permission by copyright holder

It’s still not perfect, though. There are times when it assumes the query is more general, and will teach me about real-world rocketry, rather than the game I’m trying to understand. It works best with very wordy rulebooks, where it can draw on multiple explanations of something to amalgamate its response. I had mixed results using rule summaries, which are fine for a human eye to parse quickly, but don’t necessarily give ChatGPT enough information to work with.

My new chatbot rules lawyer never mislead me, though. Its errors were obvious and not game breaking. I can really, truly use this to help learn games and to help teach others, or find an answer to a question that doesn’t mean Googling for Board Game Geek threads. It’s a real time saver.

That’s where ChatGPT continues to put its best foot forward. I’ll be keeping it at my right hand on board game nights for the foreseeable future.

I just wish it wasn’t locked behind a paywall. Paying $20 a month is a bit much just for a board game rules lawyer, but if you already have a subscription for work or experimenting, this is a great little feature addition. I’d love to see a company put together their own chatbot for public use though.

Where’s the official app, BoardGameGeek? Wizards of the Coast? Got one for DnD yet?

Jon Martindale
Jon Martindale is a freelance evergreen writer and occasional section coordinator, covering how to guides, best-of lists, and…
ChatGPT’s new Canvas feature sure looks a lot like Claude’s Artifacts
ChatGPT's Canvas screen

Hot on the heels of its $6.6 billion funding round, OpenAI on Thursday debuted the beta of a new collaboration interface for ChatGPT, dubbed Canvas.

"We are fundamentally changing how humans can collaborate with ChatGPT since it launched two years ago," Canvas research lead Karina Nguyen wrote in a post on X (formerly Twitter). She describes it as "a new interface for working with ChatGPT on writing and coding projects that go beyond simple chat."

Read more
ChatGPT’s Advanced Voice feature is finally rolling out to Plus and Teams subscribers
The Advanced Voice Mode's UI

OpenAI announced via Twitter on Tuesday that it will begin rolling out its Advanced Voice feature, as well as five new voices for the conversational AI, to subscribers of the Plus and Teams tiers throughout this week. Enterprise and Edu subscribers will gain access starting next week.

https://x.com/OpenAI/status/1838642444365369814

Read more
ChatGPT’s resource demands are getting out of control
a server

It's no secret that the growth of generative AI has demanded ever increasing amounts of water and electricity, but a new study from The Washington Post and researchers from University of California, Riverside shows just how many resources OpenAI's chatbot needs in order to perform even its most basic functions.

In terms of water usage, the amount needed for ChatGPT to write a 100-word email depends on the state and the user's proximity to OpenAI's nearest data center. The less prevalent water is in a given region, and the less expensive electricity is, the more likely the data center is to rely on electrically powered air conditioning units instead. In Texas, for example, the chatbot only consumes an estimated 235 milliliters needed to generate one 100-word email. That same email drafted in Washington, on the other hand, would require 1,408 milliliters (nearly a liter and a half) per email.

Read more