![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
Using AI to Write D&D Adventures
Jan. 16th, 2026 10:17 pmI've written recently that I'm getting a D&D adventure started. Sometimes, though, getting started is hard. Like, I have an idea of the theme or setting upon which I want to base the game, but I'm not sure what the story should actually be. Other times I've got the kernel of an idea, and it's elaborating it into a storyline with plot points and multiple encounters that's difficult. I figured generative AI could give me a hand at these challenges.
I used Google Gemini to assist with fleshing out two adventures. In one I described the basic setting and prompted "It should include undead among the monsters" and asked the AI to elaborate the major plot points and encounters of the adventure, and to detail the villain. In the other I described an initial encounter I imagined and asked what it might lead to.
In both cases AI was very helpful. It came up with creative ideas for encounters and summarized them as key points in a storyline. The AI even prompted me to ask it followup questions, like "What might be the villain's motivations?", "What help could a key NPC provide?", and "What are some unique magic items involved in the story?"
While the AI was helpful it also made mistakes. When I described this to a few friends recently, one jumped in with, "It's important to proofread what AI gives you!" That's true but it's not the problem I had. While we've probably all seen fails reposted online where a student copy-pasted an AI answer including the prompts, thus revealing that they were so lazy in using AI they didn't even read what they copied, there are failure modes in AI that go well beyond what can be solved with basic proofreading. These projects demonstrated that using AI requires you have significant domain knowledge to check its output.
The errors I caught were ones where the AI cited D&D rules and had them wrong. For example, it listed the wrong Challenge Ratings (CRs) for about half the monsters it put in the adventures. CRs are simple data lookups from monster stat blocks. It shouldn't be hard for AI to get them right. But they were wrong— and deadly wrong in at least one case. If I didn't know so many CRs by heart I might have taken an encounter with a recommended monster way too tough for the party.
In another instance, the AI assured me that the party of the 4th level characters (a detail I specified) would have key spells like Fireball and Cure Disease to overcome specific challenges. Well, those spells are both too high level for 4th level characters to get. When I challenged the AI on how 4th level characters would get such spells, it initially offered me a spirited— and completely bullshit— defense of its creation. When I challenged it a second time it admitted that it made a mistake.
"Okay, now go back and revise the encounters to correct this mistake," I prompted it. And, to its credit, it did! But the problem remains that I had to have significant domain expertise to fact-check what the AI was giving me.
I used Google Gemini to assist with fleshing out two adventures. In one I described the basic setting and prompted "It should include undead among the monsters" and asked the AI to elaborate the major plot points and encounters of the adventure, and to detail the villain. In the other I described an initial encounter I imagined and asked what it might lead to.
In both cases AI was very helpful. It came up with creative ideas for encounters and summarized them as key points in a storyline. The AI even prompted me to ask it followup questions, like "What might be the villain's motivations?", "What help could a key NPC provide?", and "What are some unique magic items involved in the story?"
While the AI was helpful it also made mistakes. When I described this to a few friends recently, one jumped in with, "It's important to proofread what AI gives you!" That's true but it's not the problem I had. While we've probably all seen fails reposted online where a student copy-pasted an AI answer including the prompts, thus revealing that they were so lazy in using AI they didn't even read what they copied, there are failure modes in AI that go well beyond what can be solved with basic proofreading. These projects demonstrated that using AI requires you have significant domain knowledge to check its output.
The errors I caught were ones where the AI cited D&D rules and had them wrong. For example, it listed the wrong Challenge Ratings (CRs) for about half the monsters it put in the adventures. CRs are simple data lookups from monster stat blocks. It shouldn't be hard for AI to get them right. But they were wrong— and deadly wrong in at least one case. If I didn't know so many CRs by heart I might have taken an encounter with a recommended monster way too tough for the party.
In another instance, the AI assured me that the party of the 4th level characters (a detail I specified) would have key spells like Fireball and Cure Disease to overcome specific challenges. Well, those spells are both too high level for 4th level characters to get. When I challenged the AI on how 4th level characters would get such spells, it initially offered me a spirited— and completely bullshit— defense of its creation. When I challenged it a second time it admitted that it made a mistake.
"Okay, now go back and revise the encounters to correct this mistake," I prompted it. And, to its credit, it did! But the problem remains that I had to have significant domain expertise to fact-check what the AI was giving me.
