ChatGPT, LLMs, and Automating the Fun Out of Roleplaying

By Leonardo Andrade

My ongoing Daggerheart campaign takes place in my homebrewed setting called Yauros’ Pit: a layered, Ireland-sized vertical tunnel (Lengh = depth, width = diameter), with its entrance surrounded by a ring-shaped island. Upon the ring, there is a city. It is partitioned in districts under the tutelage of different nations that vie for control of the Pit for its mysterious origin, endemic resources, and the relics of a bygone civilization within.

I made up exactly one of these faraway nations, the military dictatorship of Kuryar, and my players came up with the others. I had each player, when creating their character, come up with a home nation, and a reason why they can not go back (at least until their business in the Pit is concluded). We worked together to integrate the nations and get a sense of what the politics are and how the colonial cold-war over the Pit is going by the time the campaign starts. 

Now we skip ahead to the session before last: The adventurers, battered after fighting the temple’s fungal guardian, stunned by the mystical vision induced when the warlock Kareem fitted his arcane focus into the mosaic behind the altar, finally lay down their arms and bodies for a long rest. Unfortunately for them, their respite is interrupted by Kuryar officers with magically collared slave-soldiers in tow. They recognize the party’s guardian, Lady Elizabeth Stauferi, as a ransom-worthy noble due to the family arms on her shield. In trying to persuade the other adventurers to give her up—they won’t, mainly because they all have their reasons to hate the Kuryar—the soldiers let it slip that they had an encounter with another Stauferi, inadvertently giving Liz a long sought after lead about the whereabouts of her House’s missing heir. The reason she can’t go back home.

The fight was not only about survival. It was also about freeing the enslaved people, but, crucially, about getting to question the Kuryar for more information about the heir… and leaving no Kuryar soldier alive afterward to report to their outpost. High stakes and clear ties to characters' motivations and backgrounds elevated this nighttime encounter to something of narrative consequence.

This was possible thanks to all the creativity we put into this before rolling a single die. Fleshing out your character, where they came from, their motivations, their fears, etc makes everything more personal, and serves as a force multiplier for the storytelling that emerges from play. We would not have this if we outsourced the process.

Unfortunately, outsourcing creativity is precisely where it’s already going for many people, including the people in charge of the most popular tabletop RPG ever: Dungeons & Dragons.

Here’s what Hasbro (Wizards of the Coast’s parent company) CEO Chris Cocks has to say on this:

I'm probably more excited though about the playful elements of AI. If you look at a typical D&D player....I play with probably 30 or 40 people regularly. There's not a single person who doesn't use AI somehow for either campaign development or character development or story ideas. That's a clear signal that we need to be embracing it. We need to do it carefully, we need to do it responsibly, we need to make sure we pay creators for their work, and we need to make sure we're clear when something is AI-generated. But the themes around using AI to enable user-generated content, using AI to streamline new player introduction, using AI for emergent storytelling, I think you're going to see that not just our hardcore brands like D&D but also multiple of our brands.

AI for campaign and character development. For story ideas. For emergent storytelling. That’s right. The robots are coming for making stuff up with your friends, and, according to Chris Cocks, the people are all for it.

Are they? Are we?

“My DM can’t stop using AI”, “A question about using AI as lore keeper/note taker”, “Help with AI enthusiast players”,  “My friend and I used AI to bring his D&D character’s backstory to life. He did the voice-over himself!” "I am still seeing players and GMs outsource large swaths of their writing to AI and LLMs”, “Advice about a player using AI”, “Looking for experience - is anyone using an AI notetaker to help with authorial planning, etc.”, “AI note taking - what would you do?”

If Reddit is any indicator—and with the TTRPG, especially D&D, crowd, it probably is—there is a cultural shift underway where people are increasingly relying on LLMs like ChatGPT to "streamline", "kickstart", "simplify", "speed up" things like character backstory creation, worldbuilding, and session prep.

It doesn’t stop there. Of course, a lot of people are using genAI to extrude visual assets like character portraits and background images. There are also many posts of people complaining about this type of approach to the game making its way into their own groups and campaigns, as well as plentiful advertising for gen AI GMs and campaign managers. Some are even excited about getting ChatGPT to run sessions for them: they want to steal Matt Mercer’s soul via Critical Role transcript and have the LLM be their own personal celebrity dungeon master.

There is significantly more hate for commercial products that use GenAI than for private, individual use in a hobbyist capacity, but it is nonetheless a contentious issue. The spirit of the hobby hangs in the balance.

I have been running and playing numerous TTRPGs for seven years now, and writing TTRPGs professionally for five. I love the craft of it, I love how personal it can be. So it dispirited me greatly when two players I invited to a campaign submitted AI-generated backstory materials and character portraits.  I told them, in no uncertain terms, that I wouldn’t have it in my game because it cheapens the experience and taints the process; an AI-generated backstory worms its way into my prep when I want to spotlight their character, and gives them a shakier base to roleplay on top of.

Thankfully, they followed my request and used their own imaginations instead. This was a good resolution to the problem: we can’t expect people to never make mistakes or be as in the know about everything as we are about specific issues we care about, but we can and should require people to be open to new information and change their behavior accordingly. Still, even after the problem was solved in my particular group, it left a bitter aftertaste in my mouth. Why hand over your personal expression to a machine that by the rules of its operation can only produce the frictionlessly banal? That’s automating fun out of the experience while making it more generic.

Some folks may ask, “isn’t relying on LLMs the same as using a pre-written adventure or setting?”

The answer depends on if you think the following two things are the same:

• A lovingly (or at the very least intentionally) crafted book that stems from an author or team’s unique experiences and tastes and process, chasing a particular vision or ideal.
• The entire internet, turned to sludge by indiscriminate data scraping and extruded through a prompt-shaped hole to form a digestible (pre-digested) chunk of on-demand text.

If you do, you need this essay more than most.

Now, then: Why do people let a chatbot do their imagining for them? What is lost when they use LLMs? How can you push back? Strap up, I’m going to get ranty.

Time poverty is a common reason given. It rings true, with most people needing to work more and more for increasingly less reward, but it’s not the whole picture. Most people dick around on the phone a lot more than they are willing to admit. Placing a screentime widget on the home screen for the first time can be absolutely devastating, but also empowering. I speak from experience: cutting my daily phone screen time to under 3 hours—under 2.5 on a really good day—has meant more spontaneous insights about things I’m working on or interested in, more time for personal projects, and feeling less overwhelmed in a daily basis.

Drowning out your thoughts with passive media consumption while taking care of chores—Netflix second-screen slop, podcasts and audiobooks (that don’t count as screen time, necessarily…), YouTube—while extremely common, can eliminate sorely missed opportunities for self-reflection and creativity in a time-poor life. But we are not completely helpless in this matter. Better habits are possible. Making time to be alone with your thoughts and make your own things is as important as demanding better of the media you do consume.

What about the people who just play to socialize? The type of player who doesn’t want to “do homework”?  Playing to socialize is perfectly valid. As far as homework is concerned, however, coming up with a character isn’t exactly a trying term paper.  In every campaign I’ve ever been part of, a couple of paragraphs would do. If you come up with the stuff yourself, you will actually have a connection and a feeling of ownership with it. You will be able to draw from it spontaneously much more than if you have a robot do it for you. You will be a better player in session.

If you are the Game Master and you have a problem with the homework aspect or feel too time-crunched, take a break, switch with someone less time-poor and be a player for a bit—we must not accept the inevitability of the forever GM curse— and even reduce frequency. These are ways to get you energized and inspired again. It is better for you and for the hobby more broadly than resorting to LLMs.

LLMs do not understand anything, have no taste, and only produce what they calculate to be the statistically most likely output based on a prompt. It is necessarily and inescapably mediocre, even if a bit more polished than messy notes. However, I will defend that messy notes with heart and intent not only lead to more exciting gameplay, but they are how you learn to be a better GM and player. Here’s an analogy brought to you by me being a fitness dork on the side: if you wanted to become stronger, a robotic exoskeleton to let “you” rep out crazy weights would not do that for you. Your mental faculties are not that different. Really.

Which leads to what I believe is the more insidious reason behind all this automation going on: insecurity.

In creating TTRPG material, people run into the issue of their taste in fiction (books, TTRPG actual play, etc) outmatching their current creative skills. Not everyone can watch a season of Dimension 20 and instantly DM like Brennan Lee Mulligan or roleplay like Emily Axford.

The gulf between taste and skill is something all creatives grapple with. But now folks can rely on LLM to generate character biographies, settings, combat encounters, and other material closer to what they wish they knew how to make. They'll take perfectly mediocre slop that’s a bit more polished over earnest (and vulnerable) attempts at their own thing. But to do something well, one must do it not that well first. People who try to cheat the process with LLMs will stay stuck in the mud creatively by never working on these skills. 

This fear of shame walks hand in hand with the overwhelming desire for external validation.

ChatGPT and other LLMs are known sycophants that praise users for the most basic ideas—to the point of enabling psychosis or triggering episodes. These models offer cheap validation on demand and give users a misplaced sense of accomplishment when in reality all they did was type sentences of their ideas. But ideas are a dime a dozen; the world is full of good ideas that never amounted to anything. It’s all in the execution. And the creative process is often most effectively learned via trial and error–a process using a chatbot eliminates. Let’s pretend you bring an AI-generated character bio to a session and, surprisingly, it hits. You’ll feel great and claim the laurels since it was “your idea” after all.  And if it sucks, well it was from Claude/Gemini/ChatGPT so who cares?  Nothing is risked and no growth takes place. Because people fear shame.

We made AI so we could do the fun stuff

And it could do the dumb stuff

But now it does the fun stuff

And we become the dumb stuff.

Shame, however, can be wielded as an instrument to push the culture around the hobby towards good things. To some extent, this is already happening. Dunking on AI-slop and its prompters is very popular, and I say the accusatory finger of cringe should be pointed even more liberally. A lot has been said about cringe culture being bad, and I by and large agree, but that’s because it is often directed at earnestness. I think we should direct it at the lack thereof. Especially if it enriches wormy robber barons who concede their product will “sort of lead to the end of the world” but see the creation of “great companies” in the meantime as worth the sacrifice.  

“But what about educating people when they first mess up? You did it with your friends!” 

True, I did. These are people I know personally and whose values I’m familiar with, in a private setting. Giving online randos the benefit of the doubt in the same way when they proudly go out of their way to post their slop is not the same. It will work sometimes, but I’ve seen more people get defensive about it than not. In the relatively well known case of the guy who had a whole book’s worth of AI art that he ditched to pursue a home-grown art style built out of CC images. GIMP and gumption, he specifically said he pivoted away from the slop because he “learned how hated AI art was.”

We can’t completely get rid of genAI, but we can attempt to push this garbage out of our recreational spheres. With perseverance, boycotts, and luck, we might even scare some companies away from sloppifying the things we love. 

The way to save the soul of TTRPGing is not with a vapid “all are welcome”, “no judgement” attitude. It’s with standards—not of preexisting skill, but of heart and care. Putting effort into your materials makes a better experience for everyone, even if it takes time and makes more intellectual demands. What you get is proportional to what you put in. Don’t let the billionaires and their wretched creation poison the beautiful creative collaboration at the heart of TTRPGs. Value yourself and your time with your friends. 

Leo Andrade is a narrative designer and writer from Brazil who has been professionally making stuff up and writing it down since 2021. His TTRPG work is featured in several successfully crowdfunded and webstore-bestselling products like The Kingdom Of Keshanar, Stillfleet, Danse Macabre, Tome of Intangible Treasures, and more. On the videogame side, he is currently working with multiple clients on various upcoming titles in search of the mythical Actual-Shipped-Commercial-Videogame-Credit: found footage horror, action-stealth... fumblecore couch co-op? He always has many irons in the fire. Right now, his translation of Thousand Year Old Vampire to Brazilian Portuguese is crowdfunding on Catarse, too!