AI at the RPG Table: Helpful Preparation, Shared Decisions
AI can help with RPG preparation when its job is specific and its output is checked. Names, alternative hooks, short summaries, and variations on a location can be useful. Consequential rulings and player character decisions still need human judgment.
Agree on its use with the group before it becomes part of play. A brief conversation about purpose and information is enough to prevent avoidable surprises.
Ask for options you can evaluate
“Suggest five reasons a ferry captain refuses to cross tonight, each allowing negotiation or investigation” is more useful than “Write my campaign.” The bounded request produces material you can compare against the actual session.
Supply tone, available time, established facts, and the problem to solve. Use fictional summaries instead of private player information. The tool does not need somebody's personal history to suggest a haunted mill.
Select and revise rather than copying the entire output. An option becomes useful when you connect it to people, places, and consequences the group already understands.
Verify mechanical claims
Generated explanations can mix editions or invent plausible rules. Check the actual rulebook or official reference. Repeatedly asking a model to be more certain does not establish accuracy.
Separate brainstorming from mechanics in your notes. A bridge that grows roots is a fictional possibility you can adapt. A statement about a spell's required roll or cost is a rules claim requiring verification.
This distinction keeps checking efficient. You can freely revise invented fiction while treating published procedures with appropriate care.
Maintain your own continuity
Keep a short record of important people, promises, locations, and unresolved events. A generated summary should reorganize that record, not quietly replace it.
Inspect newly introduced facts. An invented sibling, changed motive, or forgotten consequence can alter a character's story. Correct the suggestion before treating it as established.
Your notes should remain understandable without the tool. Export useful material in an accessible form where possible. A campaign is easier to continue when its history is not dependent on recreating an old prompt.
An original preparation example
You need a twenty-minute scene at a flooded checkpoint. The party wants passage; guards need medicine; combat should be avoidable. Ask for three obstacles and clues supporting different solutions.
Select the idea that the commander fears admitting infected travelers. Now prepare what the commander believes, what evidence could change that belief, and an alternate route. Decide what threatening the guards would do to future relationships.
The generated idea has helped explore possibilities. Your revision supplies playable structure. The encounter should still work if you never open the tool during the session.
Keep live use narrow
Live generation can stall a session while everyone waits for text. Use it for a small task such as an incidental name when the group agrees. Avoid making a consequential ruling depend on an unverified answer.
Discuss whether character secrets, campaign notes, or table conversations may be entered. Generated town names and private participant information are different inputs. Respect the group's choices without making acceptance a condition of participation.
A player should retain control of their character's feelings and decisions. A tool suggesting what the character “would obviously do” does not override that control.
Evaluate the session, not the document
Afterward, ask whether the preparation produced clearer choices, useful details, or less wasted effort. A longer generated document is not evidence of any of these improvements.
B3's recommendation is to assign bounded jobs, verify rules, and revise fiction yourself. AI is most helpful when it reduces an identifiable preparation burden while the people at the table keep the decisions that make the game theirs.