The DM's pre-session checklist (and how AI cuts prep time in half)

The four-hour prep myth

There's a meme in the D&D community that good DMs spend four to eight hours prepping for every session. Some YouTubers and Reddit posters wear this as a badge of dedication. "I spend ten hours per session because my players deserve it."

Cool. I'm sure your players do deserve it. They probably also deserve a DM who doesn't burn out and quit running the game by month four, which is what happens to most of those DMs.

I run a session a week, sometimes two. I prep in about ninety minutes. The campaigns aren't shallow. The world has hundreds of NPCs, eight active factions, dozens of intersecting plotlines. The prep doesn't take long because I've gotten ruthless about what I actually need to prep, and I use AI for the parts that are administrative grunt work rather than creative.

Here's the checklist and the workflow.

Step zero: knowing what you don't need to prep

Before the checklist, the biggest prep-time saving is realizing that most things you'd think to prep, you don't actually need to prep.

You don't need to prep what the BBEG would do in every contingency. You'll figure it out at the table.

You don't need to prep stat blocks for NPCs the party probably won't fight. Default to "I'll improvise the rolls" or use a generic CR 1/2 commoner if violence breaks out unexpectedly.

You don't need to prep dialogue trees. You'll talk like a person and so will your NPCs.

You don't need to prep a beautiful map for every scene. Theater of the mind plus a quick sketch on paper if combat breaks out covers most situations.

You don't need to prep contingencies for player choices that haven't been telegraphed. If the party hasn't mentioned going to the temple in three sessions, don't prep the temple. They aren't going there.

You don't need to write a recap. AI can do this in 90 seconds (more on this below).

Cutting these out alone reduces prep time by half for most DMs. The next ninety-minute checklist is what's left.

The 90-minute checklist

Here's the actual structure, broken into stages.

Stage 1 (15 min): figure out what's likely to happen

Look at last session's notes. What did the party say they were going to do next? What threads are open? What's pulling at them?

In your head or on paper, sketch:

• One sentence describing this session's most likely arc.
• Two to three scenes that will probably happen.
• One "if they go off-script" path you have a sense of.

That's it. You're not committing the session to a script. You're just orienting yourself.

Example from a recent session:

Most likely arc: party investigates the dock warehouse the smuggler tipped them off about. Scenes: arrival at the warehouse, sneaking past or fighting guards, finding the back room with the contraband and the kidnapped scribe. Off-script: if they decide to go to the city watch first instead, the watch captain is corrupt and tries to redirect them.

Three minutes to write that down. It clarifies what I actually need to prep next.

Stage 2 (15 min): identify what's known and what's missing

Open your campaign notes. For the scenes you sketched in stage 1, do you have what you need?

The warehouse: do I have a description? Yes, I noted it three sessions ago. Do I have the guards stat-blocked? No. Do I have the back room described? No. Is the kidnapped scribe a person yet? No, they were just a name a contact mentioned.

So I need: guard stat blocks (probably four guards, CR 1/2), back room description, scribe NPC.

The corrupt watch captain: do I have a name and personality? No, this is new to me too. Is there a faction angle? Yes, watch is loosely allied with the Iron Crown.

So I need: watch captain NPC with name, voice, motivation.

Now I have a list of four things to prep. That's manageable. If your list at this stage is twelve things long, your scope is too big. Cut.

Stage 3 (30-40 min): batch prep with AI assistance

This is where AI saves the most time. Here's the workflow.

Open your AI tool of choice. ChatGPT, Claude, or Dungeon Diary's AI generation if you want it to be campaign-aware (more on this in a second).

Knock out the items from stage 2 in batch.

For the guards: "Give me a quick stat block for four warehouse guards working for a smuggler, CR 1/2 each, in D&D 5e. Two should be slightly different from the other two." Ten seconds. Tweak the result, save it.

For the back room: "Describe a smuggler's back room in a dock warehouse: contraband stacked, evidence of a struggle, a chair where someone's been kept tied up. Tone: gritty, lit by a single oil lamp." Ten seconds. Edit lightly, save.

For the scribe: "I need a kidnapped scribe NPC for D&D. Name, race, one personality detail, what they know that's valuable, what they're afraid of." Twenty seconds. Edit a bit. Done.

For the watch captain: this one's a recurring NPC, I'll be using him for several sessions, so I write him myself. Ten minutes. The model couldn't get the voice right anyway.

Here's the thing about AI in prep. Use it for the high-volume, low-creativity stuff. Stat blocks, generic descriptions, throwaway NPCs. Use yourself for the recurring NPCs and the major plot beats. The split saves real time without flattening the game.

Stage 4 (10 min): check pacing and stakes

Read back through your prep. Two questions:

What's the pace? If the session is all combat, that's a long night. If it's all conversation, players will get restless. Aim for at least one combat and at least one social or exploration scene. Not always, but usually.

What are the stakes? In each scene, what does success look like and what does failure look like? "Find the scribe" is a goal, but if there's no consequence to not finding them, it doesn't matter. "Find the scribe before the smugglers move them at dawn" has stakes.

Adjust your scene sketches if the pacing or stakes are off. This usually takes five to ten minutes.

Stage 5 (5 min): write the cold open

The first sixty seconds of a session set the tone. I almost always write the cold open in advance because it's high-leverage and short.

Cold open is the first thing the party hears or sees when the session begins. Two to four sentences.

Example:

The morning fog hasn't lifted off the docks. You can hear gulls and the creak of rope but you can also hear, faintly, someone weeping. It's coming from the warehouse three berths down. The door's cracked open, just barely.

That's the open. Now I'm not stuttering when the session begins.

Stage 6 (10 min): prep the recap

For the recap, I paste my notes from last session into AI and ask for a 100-word "Previously on..." paragraph. The model writes it in 90 seconds. I edit for accuracy and tone. Saves the painful "uh, where did we leave off..." opening.

If you don't want to use AI for this, write it yourself in the same ten minutes. Either way, having the recap ready is a small thing that smooths the start of the session noticeably.

What's not on the checklist (intentionally)

I want to call out things that are NOT on the checklist, because they're things DMs often think they should prep but shouldn't.

Pre-written dialogue. I never write what NPCs say in advance. I write what they want, what they know, and what they're hiding. The actual words come at the table.

Skill check DCs. I don't prep these. I set them as the situation calls for it. DC 12 for easy, 15 for moderate, 18 for hard, 20+ for very hard. I adjust based on circumstance.

Maps for every scene. I sketch maps on the fly during play if needed. Pre-made maps are nice but they're not required. Theater of the mind covers most situations.

Loot tables for every encounter. I have a few generic loot rolls in my head and I improvise. The party's "ooh what'd we get?" reaction is the same whether I planned it or made it up two seconds ago.

Detailed encounter tactics. I think about the enemies' goals (defend, attack, capture, retreat) but I don't script their turn-by-turn actions. They'll act according to what makes sense in the moment.

Long-term plot details. Yes, I have an arc plan. No, I don't prep three sessions ahead. Players will choose paths I didn't expect. Prep the next session, not the next month.

Campaign-aware AI versus generic AI

I want to come back to this because it matters for the prep workflow.

When you ask ChatGPT for a smuggler's back room, you get a generic smuggler's back room. Fine for one-shot use. But if your campaign has a specific smuggling ring (the Crooked Anchor, run by Madame Velasque, with ties to the Veilkeepers), and you want the back room to feel like THAT smuggling ring, raw ChatGPT won't give you that without you stuffing context into the prompt every time.

That's where campaign-aware AI helps. Tools like Dungeon Diary keep your campaign data (factions, NPCs, settlements, lore) in vector embeddings, so when you ask the AI for a new NPC or scene, it pulls relevant existing context automatically. The smuggler's back room generated this way will reference your existing factions, your established characters, your campaign's tone.

This isn't magical. It's just retrieval-augmented generation. But for D&D prep specifically, it's the difference between "AI gave me a generic scene" and "AI gave me a scene that fits my world."

You can replicate a weaker version of this by maintaining a campaign bible doc and pasting relevant sections into each prompt. Works fine if you're disciplined about it. Most DMs aren't.

The shortcut version

If you only have thirty minutes to prep:

1. Five minutes: read last session's notes, decide one likely arc.
2. Ten minutes: identify three things you need (NPC, scene, encounter), batch-generate with AI.
3. Five minutes: write the cold open.
4. Five minutes: glance at the active factions to make sure you remember which ones are present in this scene.
5. Five minutes: pace check, stakes check, breathe.

You'll be a little less prepared than the full ninety-minute version, but you'll run a session that's 80 percent as good as fully prepped. Sometimes that's the difference between running and canceling, and any session is better than no session.

What goes wrong without prep

To be fair to the "more prep is better" crowd, here's what happens when you don't prep at all.

You forget which faction the merchant works for. You give the wrong NPC the same name as a different NPC. You contradict an established lore detail. You blank when the party asks "where do we go from here?"

These are real costs. The fix isn't four hours of prep. It's a working campaign manager (whatever tool you use, but a purpose-built one helps) and the ninety-minute checklist above. The structure is what saves you, not the time.

If you're in a campaign where you genuinely need four hours of prep to cover what's in motion, your campaign is too complex. Simplify the active threads. Park some plotlines as background. Resume them when the active set is resolved. Most "I need more prep time" problems are actually "I'm trying to juggle too many balls" problems.

A two-week trial

If your prep currently takes three or more hours per session and you want to cut it down, try this for two weeks.

Week one: stick to the ninety-minute checklist exactly. Set a timer. When ninety minutes are up, stop prepping. Run the session. Notice what you missed and what you didn't actually need.

Week two: adjust the checklist based on what you learned. If you skipped something important, add it. If you spent time on something the players didn't notice, drop it.

By week three, you'll have a checklist tuned to your specific campaign and table. It probably won't be exactly mine. That's fine. The point isn't my checklist. It's having a checklist at all.

The DMs I know who burn out are the ones who treat every session like a final exam. The ones who keep going long-term treat prep like cooking dinner. You have a routine. You don't reinvent the cuisine each time. The food is good, you sit down, you eat together, you do it again next week.

That's the goal. Sustainable prep. Ninety minutes a week, ten years of campaigns.