Can AI Help You Run a TTRPG Campaign? An Honest Answer

Short answer: AI is genuinely useful for the bookkeeping side of running a campaign, and a poor substitute for the GM. It’s good at remembering what happened, noticing when your notes contradict each other, and drafting a recap. It’s unreliable at the things that make a game good: reading the table, improvising, and judging what’s fun. We build an AI tool for GMs, so we live inside this question, and the honest answer is more useful than the hype in either direction.

Two very different jobs

Conversations about AI and tabletop games go sideways because “help run my campaign” covers two jobs:

Bookkeeping is the work around the game. Who’s dead. Who has the cursed sword. Which quests are open. What the players know. What happened last session. This work has right answers. Either your notes say Old Wren died in session 2 or they don’t.

Game mastering is the game itself. Improvising the innkeeper. Pacing the dungeon so it doesn’t drag. Deciding the villain escapes because the table will love the chase. Noticing that one player has been quiet all night. None of this has a right answer. It depends on these players, tonight.

AI handles the first job well and the second badly, for reasons worth understanding before you trust any tool, including ours.

Where AI is genuinely good

Remembering everything. The model doesn’t get tired at session 30. Given your codex and your notes, it can tell you who the party met in session 4 as reliably as who they met yesterday.

Updating your notes from what happened. After a session, a tool can read your notes and propose the changes: an NPC died, an item changed hands, a secret came out. You approve each one. This is the step most GMs skip, and it’s the reason notes drift. In StoryGuard it’s a button called What changed?

Catching contradictions. When a model is given the established facts alongside new notes, finding where they disagree is a bounded comparison task, which models do well. It catches the dead NPC at the dock, the faction that turned hostile and then sells you rope, the item that’s in two places at once. (We walked through how that works step by step.)

Drafting the recap. Turning messy notes into a readable summary for the group chat is a writing task models are good at. With a list of what’s still secret, a tool can also leave out the twists your players haven’t found.

Why “just paste it into ChatGPT” underdelivers

You can paste your notes into a general chatbot and ask it these questions. It half-works. The problems:

• No source of truth. It compares your notes to its own memory of your notes, which blurs across a long log.
• No sense of order. It struggles to know what was true when, so it gets “dead since session 2” wrong in both directions.
• No idea what’s secret. Ask it for a recap and it will cheerfully spoil the mayor’s allegiance, because it has no way to know your players don’t know.
• It flags the game as errors. An NPC lying to the party looks like a contradiction to a model that doesn’t know NPCs lie.

The fix is structure: a codex with facts, changes recorded by session number, and secrets marked with when they came out. That’s the difference between a chatbot and a purpose-built tool.

Where AI is still bad (honestly)

We’d rather say this than have you find out at the table:

• It doesn’t improvise for you. It can suggest a name for the innkeeper. It can’t read the room and decide this is the moment for a twist.
• It doesn’t know your rules. StoryGuard doesn’t track hit points, spell slots, or action economy, and you shouldn’t trust any general model to adjudicate a rules argument mid-combat. Look it up.
• It only knows what you wrote down. If you improvised a detail and never noted it, no tool can check it.
• It makes mistakes. Fewer with structure, but some. Treat every flag as a question and every proposed change as a draft.
• It can’t tell you if it’s fun. Whether your campaign works is between you and your players.

The realistic split

Whatever tools you use, this division of labor holds up:

1. AI: the bookkeeping. Updating the codex after sessions, checking for contradictions before sessions, drafting recaps.
2. You: the world and the calls. Deciding what’s true in your setting, what’s secret, and what happens next. AI should propose. You decide.
3. The table: everything that matters. The jokes, the plans, the moment the rogue does something nobody expected.

That isn’t a consolation prize for AI’s limits. It’s the right assignment of work. Software forgets nothing and never gets bored of cross-referencing. You know your players. The GMs who get the most out of AI aren’t the ones hoping it will run the game. They’re the ones who stopped spending Sunday afternoon reconciling notes, and spent it on prep instead. If you want to try the bookkeeping half, here’s how to keep track of a campaign with or without a tool.