How to play solo D&D: a beginner's guide that doesn't need a group

The one problem solo play has to solve

In a normal game the DM knows things you don't. Alone, you can't be surprised by your own decisions — so you need a source of information that isn't you. Every solo technique is a different answer to that single problem.

Approach 1 — Oracles (start here)

An oracle is a yes/no engine with texture. You ask a closed question — "Is the gate guarded?" — decide how likely it is, roll, and read the answer. Good oracles add complications: "yes, but", "no, and". Those two words are where solo play gets its plot.

You can bolt an oracle onto 5e and keep your character, your spells and your monster manual. Roll for the situation, resolve it with the D&D rules you already know. Most people who "can't get into solo play" bounced off because they were asking open questions ("what happens?") instead of closed ones ("does the ambush come from behind?").

Approach 2 — A GM emulator

Emulators are structured procedures — scene setup, an interrupt roll that derails your expectation, random event tables — layered over the oracle. Mythic is the best-known. It gives you the rhythm of a session: you write down what you expect the scene to be, and the procedure decides whether reality agrees.

More bookkeeping than oracles alone, but it's what stops a solo game turning into wish fulfilment.

Approach 3 — A system built for one player

D&D 5e assumes a party and a referee; you'll spend energy patching both. Systems designed for solo play — Ironsworn is the common landing spot, and it's free — build the oracle and the emulator into the rules. You take a vow, you make moves against it, you mark progress, and a single roll shape (strong hit, weak hit, miss) covers everything. A weak hit means you get what you wanted and a cost, which is the mechanic that keeps a solo story moving on its own.

Where AI fits — and where it doesn't

Using a chatbot as your narrator is now the most common first attempt, and the first hour is great. Over a month it fails in four predictable ways I documented in this field report: it fudges the dice, it loses track of your HP and NPCs, it agrees with everything you propose, and so nothing you do has consequences.

The useful split is this: let a model write prose, never let it referee. Dice belong in code, state belongs in a database, rules belong in a rules engine. Here's why the dice part in particular can't be fixed with better instructions.

Your first session, in about twenty minutes

1. Make one character you'd miss. A name, a want, and one thing they're afraid of. Skip the optimisation.
2. Write the goal as a sentence. "Find out who burned the mill." Solo play collapses without a stated objective.
3. Open on a problem, not a tavern. Start in the middle of the trouble.
4. Ask closed questions. "Is anyone still inside?" — likely, roll, obey the answer even when it's inconvenient. Especially then.
5. Write two sentences per beat. The journal is the game. It's also what lets you pick the story back up next week.
6. Stop at a cliffhanger. Always end while you still want the next scene.

If you want it set up for you

Markbound is a solo roleplaying game with an AI narrator that can't cheat — the dice roll in your browser, the rules are real Ironsworn, and the story remembers what you did. Character, oracles, sheet, progress tracks and journal are already wired together, so the twenty minutes above becomes about two. The front page runs two real turns without an account, and the dice there are the same ones a live saga uses.