How to Use an AI Dungeon Master to Run Solo RPG Adventures That Actually Work
- markstiltner
- Apr 5
- 13 min read
Updated: 8 hours ago
Yes, AI can run a solo RPG. And when you set it up right, it runs a genuinely good one.
An AI Dungeon Master works best when you give it three things: a paid AI model, a structured GM prompt, and a complete adventure module. Give it all three and you get something that feels remarkably close to sitting across the table from a real game master.
What follows are lessons from developing my own solo RPG modules at Dungeon Matters. These modules are made to be copied, pasted, and played in an AI prompt window. I have been using AI as both a player and a Dungeon Master since 2022. Over that time, I have figured out what works, what breaks, and exactly how to fix it.
This guide covers the common mistakes, how to write an AI Dungeon Master prompt, how to convert existing adventures into solo AI modules, how to use procedural generation, and which AI model performs best for the job.
Quick Start: How to Use AI as a Dungeon Master
If you want to get playing right now, here is the short version:
Use a paid AI model. Claude Opus is best. ChatGPT Plus, or Gemini Pro both work well. A free Claude account can work in a pinch.
Paste in a structured GM prompt. This tells the AI how to behave as a Dungeon Master: what to reveal, what to hide, when to call for rolls, and how to run scenes.
A good prompt includes an adventure module. The AI runs a story far better than it invents one. Give it a full module with scenes, stat blocks, clues, and outcomes already written.
Let the AI run scenes and resolve your actions. You play. It narrates, tracks, and referees.
Ready to start? You can download our free AI Dungeon Master prompt and module conversion tools.
What Is an AI Dungeon Master?
An AI Dungeon Master is a large language model, like ChatGPT, Claude, or Gemini, configured with a prompt and adventure content to serve as the game master in a solo RPG session.
It narrates scenes, describes environments, plays NPCs, tracks your character state, resolves dice rolls, and manages the flow of the adventure. It does everything a human DM does at the table, minus ordering pizza. (Although AI can do that too.)
The key difference from a chatbot or AI storytelling tool is structure. A proper AI Dungeon Master does not just tell you a story. It runs a game. It follows rules, hides information you have not earned, calls for rolls when outcomes are uncertain, and keeps the world consistent from scene to scene. AI won't do any of that well by default. It needs instructions and a story to follow.
Why it matters: Without that structure, AI defaults to collaborative fiction. It gives you everything, skips the tension, and hands you the ending. A well-prompted AI Dungeon Master plays tough but fair.
Can AI Be a Dungeon Master?
Yes. AI can absolutely be a Dungeon Master under the right conditions.
Out of the box, if you type "run a D&D game for me" into ChatGPT, it will try. It will dump you into a scenario with no background then build a story for a few scenes. Then it will start drifting. It gives away plot points. It invents encounters that sound connected but go nowhere. Pretty soon you are knee-deep in the slop swamp and the story has stopped making sense.
This can be solved with the right prompt. When you give AI a structured prompt, a clear set of rules, and a complete adventure module, it performs remarkably well. It tracks your hit points, manages NPC behavior, gates clues behind investigation checks, and maintains narrative tension across long sessions.
The short answer: AI can be a Dungeon Master. It just cannot do it alone. You need to use the right tools.
A paid model with a large context window
A GM prompt that defines its role, responsibilities, and constraints
A complete adventure module with scenes, encounters, and outcomes already designed
A lightweight rule system built for solo play
Common Mistakes When Using an AI Dungeon Master
Most players and prompt engineers make the same mistakes when setting up an AI Dungeon Master. Here are the three that matter most and how to fix them.
1. Story Drift
This is the most common failure. You paste a prompt that says something like "Act as a dungeon master and run a D&D adventure for me." A good paid model will start strong. Then, over the course of several scenes, it begins to drift.
It gives away story elements too early. It creates encounters that feel connected on the surface but do not actually lead anywhere. It loses track of what the player knows versus what is hidden. The experience goes from promising to incoherent.
The fix: Use a prompt that instructs and constrains. The AI is much better at following a story and a set of rules that you provide than making up its own. Structure prevents drift.
2. Having AI Create the Story
This was the big breakthrough for me. No amount of prompting constraints fixed the problem when the AI was also responsible for inventing the story. It simply is not good at that — not in the way a game requires.
But once we started writing complete adventure modules and handing them to the AI to run, everything changed. The AI stopped guessing and started executing. Scenes connected. Clues paid off. Endings felt earned.
You do not have to write the module yourself, either. You can use AI to convert an existing adventure, book, movie, or story idea into a solo AI module. We have a prompt for that, and we cover the process later in this guide. In our testing, Claude Opus produces the best module conversions. But for actually running the game, Claude Opus or Sonnet each have their strengths. Opus is best at maintaining the story. Sonnet is faster loading and less verbose.
3. Using Rules That Are Too Complex
D&D 5e is probably the most popular RPG rule system ever made. But the complexity that makes it great for a human-run table can cause real problems for an AI Dungeon Master.
An AI GM is already juggling a lot: your character's position and state, NPC behavior, scene management, hidden information, and prompting constraints. Layer a complex rule system on top of that and things start to slip. When the AI does not have a clear answer, it makes one up. Once it starts hallucinating rules, the game unravels.
What works better: A lightweight rule system designed for solo play. Ideally one where the player makes all the rolls, keeping the action focused on you.
Options that work well for AI solo RPG games:
Dreadful Things Solo RPG System — Dungeon Matters’ own d20-based system, built specifically for AI solo play. Short rules that copy cleanly into a prompt. Low HP and damage values to prevent combat slog. All rolls are player-facing. Streamlined handling of NPC companions, madness, grappling, spells, and other common solo play issues.
Mörk Borg — A streamlined d20 system that works well with AI, though you will want to drastically simplify the written rules before pasting them into a prompt.
Shadowdark — Another solid lightweight d20 option. Same recommendation: summarize and simplify before including in your AI game master prompt.
How to Write an AI Dungeon Master Prompt That Actually Runs a Game
A strong AI Dungeon Master prompt does three things: it gives the model a clear job, a clear boundary, and a clear gameplay loop.
The AI is not asked to "help with a game" in some vague way. It is assigned specific responsibilities: scene runner, referee, narrator, tracker, and assistant. The prompt defines exactly what belongs to the GM, what belongs to the player, and what order to follow when resolving play.
That structure is what makes an AI Dungeon Master feel consistent instead of random.
Role and Rules Hierarchy
The prompt establishes what the AI prioritizes when instructions conflict. In our system, the hierarchy works like this:
Adventure text comes first. The module is the authority.
Core rules come second. When the module is silent, the rules decide.
The oracle comes third. When neither the module nor the rules cover a situation, a simple oracle system resolves the uncertainty.
Information Control and Pacing
This is where most AI Dungeon Master prompts fail. You have to explicitly tell the AI:
Do not reveal hidden information
Do not lead the player toward clues
Do not skip scenes or rush pacing
Do not narrate outcomes before the dice resolve them
Present only what the character can directly perceive
Keep the player in the current location until they choose to move
Convert all behind-the-screen instructions into in-world description
If you want the AI to feel like a real game master, you have to tell it what not to spoil, when to stop, and how to preserve discovery. That is maybe the single most important lesson for anyone writing an AI Dungeon Master prompt.
Edge Cases and Guardrails
The best prompts teach the model how to run the game turn-by-turn.
Track resources internally without revealing totals unless asked
Improvise only within defined limits
Reward smart tactics with advantage instead of auto-success
Use the oracle when rules do not cover a situation
Redirect strange or invalid player actions without breaking character
How to Build an AI Dungeon Master Step-by-Step
Here is the checklist for building an effective AI Dungeon Master prompt.
Define the AI's role in one sentence. Tell it exactly what it is: referee, narrator, tracker, scene runner.
Separate player and GM responsibilities. The model should never roll for the player or take over player choices.
Create a rules hierarchy. Adventure text first, core rules second, oracle third.
Write a gameplay loop. Present scene > ask what the player does > call for a roll only if needed > resolve consequences > update state.
Add explicit roll rules. Include formulas, DC handling, and clear guidance on when rolls are and are not required.
Hide unrevealed information. The AI must never state conclusions the character has not earned.
Block leading behavior. No nudging the player toward clues or announcing outcomes before they happen.
Control scene flow. The AI does not move the player, skip locations, or jump ahead unless the player explicitly chooses it.
Define improvisation boundaries. Specify what the AI can invent and what it cannot, so it stays inside the adventure.
Include internal tracking. Hit points, inventory, clues, objectives, conditions, and world state.
Add a fallback system. An oracle or simple resolution method for outcomes the adventure does not already cover.
Lock in voice and tone. The AI should sound like a live Dungeon Master, not a chatbot explaining rules.
How to Convert Adventures Into Solo AI Modules
You do not have to write an adventure from scratch. You can use AI to convert an existing module, story, movie, or book idea into a solo AI RPG module that your AI Dungeon Master can run.
Start with the right prompt. A vague instruction like "convert this for solo play" produces inconsistent results because the AI has to guess what matters. A strong prompt removes that ambiguity.
Why Conversion Works Better Than Real-Time Improvisation
When the AI converts first and runs second, it is executing a designed system. When it tries to do both at once, it tries solving multiple problems simultaneously and becomes inconsistent. A converted module has every scene, stat block, rule, and outcome already written. The AI does not need to invent anything during play. It just runs the game.
The Conversion Process
The prompt should require the AI to follow a strict sequence:
Read the entire adventure first.
Identify core structure. Plot beats, key locations, NPCs, discovery gates, and multiplayer dependencies.
Redesign for solo play. Convert locations and events into clearly defined scenes. Add explicit choice points. Gate information behind checks. Ensure every scene points to at least one additional scene (except for the ending).
Rebalance mechanics. Scale encounters for one character. Adjust damage and difficulty to maintain tension without instant failure.
Produce a complete, standalone module. Every scene, stat block, and table must be included.
What the Conversion Prompt Must Enforce
Follow a phased process. Analysis, design, conversion, assembly. Each phase is completed before the next begins.
Standardize formatting. Consistent scene headers. Narrative text separated from hidden instructions. Clear, playable structure throughout.
Include progression and endings. A clear adventure hook. At least one advancement point. Multiple endings based on player choices. Outcomes tied to earlier decisions.
Replace the human DM. Add mechanics for tension, escalation, and time pressure. Include optional discoveries that reward investigation. Handle unknown outcomes with an oracle.
Use Procedural Generation Instead of Maps in a Solo AI Crawl
Procedural generation is a method that replaces fixed maps with simple rules and random outcomes. Instead of designing every room ahead of time, you define tables and let the AI generate each space as the player enters it.
This approach works because it plays to the AI's strengths. The AI follows rules instead of tracking a full map. Each space is independent, which prevents continuity errors. Random tables create variety without requiring memory.
How It Works
The core loop is straightforward:
The player enters a new space.
The AI generates the space using predefined tables.
The AI presents the scene and resolves the encounter.
The player acts, rolls if needed, and moves to the next space.
The crawl continues until a defined objective is met.
Example Table Structure
Room or Environment Table — Defines what the space is:
1–2: Narrow corridor or passage
3–4: Standard room or chamber
5: Large or multi-level space
6: Special location tied to the objective
Encounter Table — Defines what happens in the space:
1–2: No encounter
3: Enemy appears
4: Trap or hazard triggers
5: Environmental challenge or obstacle
6: Major threat or elite enemy
Discovery Table — Defines what the player can find:
1–2: Nothing of value
3: Minor loot or supplies
4: Clue or useful information
5: Valuable item or resource
6: Key item tied to progression
These tables can be adjusted to fit any setting. A mansion replaces enemies with social encounters. A hex crawl replaces rooms with terrain types. The system stays the same.
Defining Success Conditions
A procedural crawl needs a clear end state. Without one, the experience becomes endless and loses tension.
Common success structures:
Survive a fixed number of locations. Example: clear 8 rooms before reaching the boss.
Collect required items. Example: find 3 key artifacts from the Discovery Table.
Trigger a final encounter. Example: a specific roll unlocks the boss location.
Combine conditions. Example: survive 6 rooms and recover 2 clues to reveal the final chamber.
Failure conditions should be equally clear. Death, resource depletion, or too many high-risk encounters can all end the crawl.
A Note on Dice Rolling
AI does not actually roll dice. It can simulate a roll by selecting a number within a range, or approximate randomness through internal generation. Both approaches work, but for solo play, it is often better to have the player roll.
Each time a new location is generated, the AI prompts the player to roll on the relevant tables. The player reports results. The AI resolves them. This keeps you actively involved in the system instead of passively watching outcomes happen.
What Is the Best AI Model for an AI Dungeon Master?
The best AI model for running an AI Dungeon Master is Claude Opus. It strikes the right balance between intelligence, speed, and conversational clarity for gameplay.
Not all AI models handle game mastering equally. The difference between a usable experience and a broken one usually comes down to model capability and context window size.
Free Models vs. Paid Models
Free LLM tiers tend to fail for this use case. The main issue is context window size. A solo AI RPG adventure requires the model to track rules, scenes, inventory, and narrative state simultaneously. Smaller context windows cause the model to forget earlier details, leading to contradictions, missed mechanics, and broken progression.
Free models are also more prone to hallucination. They invent rules, skip required steps, and ignore structure. The experience falls apart during longer sessions.
The one exception is a free Claude account. If you use the Sonnet model with a free Claude account, it performs as well as, if not better than, a paid ChatGPT or Gemini account. The one drawback is the limited number of prompts, so during heavy gaming it is possible to run out of LLM calls mid-session. However, in testing, I have been able to complete a full one-shot solo adventure in a single free Claude session.
Use a paid model if you want the best AI DM.
Paid Model Comparison
Claude Opus (versions 4–6) — Best overall for AI Dungeon Master play. Follows structured prompts closely. Delivers natural, immersive narration.
Claude Sonnet — Faster loading and less verbose than Opus, but lacks some of the narrative depth and is not as good at improvising.
ChatGPT 5 — Best option when image generation is part of the experience. Some adventures include instructions to generate character sheets or visual scene art. ChatGPT handles this best in our tests.
Gemini Pro — Performs well overall. Solid choice if it is your preferred platform.
Quick Recommendation
Best gameplay experience: Claude Opus
Minimum requirement: A free Claude account
Download Our Free AI Dungeon Master Prompt and Tools
We spend a lot of time writing, playtesting, and refining our solo AI RPG modules by hand. But we also have AI Dungeon Master tools that you can use to create and run your own solo AI games.
The free download includes:
A complete AI Dungeon Master prompt — ready to paste and play
A conversion prompt to transform any adventure into a solo AI module
Instructions for setup and use
These are the same prompts we use to develop our published adventures. They work with ChatGPT, Claude, Gemini, and other major models.
Frequently Asked Questions
Can AI replace a human Dungeon Master?
Not entirely. AI cannot read the room, adjust tone on instinct, or bring the kind of personal creative energy a great human DM provides. But for solo RPG games, it is the best available option — and when properly prompted with a complete module, the experience is genuinely engaging.
Do I need a paid AI model to use an AI Dungeon Master?
It is recommended. Most free models lack the context window and instruction-following capability needed for sustained gameplay. The one exception is a free Claude account running Sonnet, which does not perform as well as a paid Claude account but is a great option if you are just getting started and want to try solo AI adventuring.
What is the best AI for solo RPGs?
Claude Opus is the best overall for running AI solo RPG adventures. ChatGPT Plus is the best choice when image generation is part of the adventure. Claude Opus is the best for converting existing modules into solo format.
Can AI generate adventures on its own?
AI can generate adventure content, but it performs much better when converting an existing story or module rather than inventing from scratch. A conversion prompt that forces structured analysis before writing produces significantly better results than open-ended generation.
Does this work with D&D 5e?
It can, but results are better with a lighter rule system. AI Dungeon Master 5e setups require heavy simplification of the rules before pasting them into a prompt. Systems like Dreadful Things, Mörk Borg, or Shadowdark are better suited to AI solo play because they use fewer rules and keep all rolls player-facing.
Can ChatGPT run D&D?
ChatGPT can run a solo RPG session when given a structured GM prompt and a complete adventure module. On its own, without structure, it tends to drift into collaborative storytelling rather than actual gameplay. The prompt and module are what make it work.
What AI Dungeon Master tools does Dungeon Matters offer?
We publish complete solo AI RPG modules designed for AI play, a free AI Dungeon Master prompt, and a conversion prompt for transforming existing adventures into solo AI format. All of our tools are available for download.
