AI at My Table: How AI Actually Helps with D&D Game Prep

In 1979, Gary Gygax published the Dungeon Masters Guide with page after page of random tables. Random dungeon dressing. Random NPC traits. Random harlot encounters, which tells you something about the era. The tables weren't elegant. They weren't narrative. What they were, quietly and importantly, was an admission: running a game requires more content than one person can reasonably produce alone.

That admission has been the unspoken foundation of GM prep tools ever since. Donjon's random generators. The d100 lists on Reddit. The Notion templates, the OneNote binders, the spreadsheets that grow like kudzu across a campaign's lifespan. Each one answers the same question Gygax was answering forty-five years ago: how do you fill the gaps between the moments you actually care about?

AI is the latest answer. And it's genuinely different from the ones before it. Not because it's smarter than a random table (that bar is low) but because it understands something tables structurally cannot: context. A random table doesn't know you're running a nautical campaign. It doesn't know your party just lost their cleric. It doesn't know that the last three NPCs all had tragic backstories and the table desperately needs someone who's just having a good day.

AI knows. Or at least, it can be told. And that changes the relationship between you and your prep in ways worth thinking through carefully.

The Work That Was Never Yours

There's a romantic idea about GM prep that nobody says out loud but everyone absorbs: that doing it all yourself is part of the craft. That hand-writing stat blocks and manually calculating encounter XP and inventing thirty NPC names from scratch is the authentic way. That using tools is cutting corners.

This idea is, respectfully, nonsense.

Consider what goes into a single session of prep. You need NPCs with names, motivations, and mechanical stats. You need encounters balanced against your party's level, composition, and current resources. You need locations described in enough sensory detail that the table can picture them. You need plot hooks that connect to threads you've already established. You need contingencies for when your players do the thing you didn't expect, which is every session.

Some of that work is genuinely creative. The villain's speech. The twist that reframes the whole arc. The moment you've been building toward for six sessions. That work belongs to you because only you know your table, your players, your story.

The rest of it? Stat block arithmetic. SRD compliance checking. Generating the fifth shopkeeper this month. That's clerical work wearing a creative costume. It sits next to the real creative work and absorbs the same hours, and after a while you can't tell where the storytelling ends and the bookkeeping begins. AI is useful precisely because it draws that line for you.

What AI Actually Does Well

Let's be specific. Vague claims about AI "helping with prep" don't help anyone.

The mechanical heavy lifting

Stat blocks. Encounter math. AC, HP, ability scores, saving throws, action economy. The kind of work where there are right answers and wrong answers, where a mistake wastes table time, where the creative contribution is nearly zero. AI handles this with a consistency and speed that makes manual calculation look like a hobby choice rather than a practical one.

A purpose-built generator takes this further. CritForge's encounter tools, for instance, produce stat blocks that follow 5e SRD conventions, factor in CR-appropriate action economy, and include tactical notes about how the creature actually fights. Not just numbers on a page. Numbers with behavior.

NPC population

The town your party just wandered into needs twenty people in it. You prepped three. AI fills the other seventeen with characters who have names, occupations, personality traits, and at least one reason to exist beyond "shopkeeper number four." The coherence matters here. A random name generator gives you a name. AI gives you a person whose name, job, and demeanor all suggest the same life history. The nervous herbalist who keeps glancing at the guard tower isn't nervous because of a random trait roll. She's nervous because the narrative context implies something worth being nervous about.

Compliance as infrastructure

If you're creating content you'll ever share, publish, or store in a tool, SRD 5.2 compliance isn't optional. It's a legal boundary. And it's the kind of boundary that's invisible until you cross it. You won't notice the trademarked creature name in your stat block until someone else does. AI systems built for tabletop content can catch these before they reach the page, replacing non-SRD terms with open alternatives automatically. It's unglamorous work. It's also work that, done manually, requires memorizing a blocklist of seventy-plus settings and sixty-plus locations. That's not a creative task. That's a lookup table, and lookup tables are exactly what machines should do.

Structural templates

The Three-Clue Rule, branching plot outcomes, doom clocks, encounter pacing across a session. These are well-documented frameworks with known best practices. When you're exhausted the night before a session and you know your mystery plot needs three independent paths to each conclusion but you can only think of one, AI can generate the structural framework. You bring the texture. The apprentice's nervous stammer. The specific smell of the catacomb entrance. The detail that makes your table lean forward. But the architecture? The architecture is a solved problem, and solved problems are exactly where AI earns its keep.

What AI Cannot Do

This section matters more than the one above it, because the limits define the tool. And the limits are real.

Read your table

AI doesn't know that your player Sarah gets quiet when violence toward animals comes up. It doesn't know that Marcus plays a stoic fighter but lights up whenever an NPC shows him genuine kindness. It doesn't know that the whole table groaned last time you introduced another "mysterious stranger at the tavern" and that what they really want is a mystery where they're the ones being investigated.

This kind of knowledge lives in you. It's accumulated through sessions, through the micro-expressions you catch across the table, through the post-game conversations where someone casually mentions the best part of the night and you file it away for later. No model has access to this information, and even if you typed it all out, the nuance would be lost in translation. The difference between "Sarah doesn't like animal violence" and understanding the specific, particular way Sarah's engagement shifts when you describe a wounded dog is the difference between data and empathy.

Improvise in the moment

The party does the unexpected thing. They always do the unexpected thing. The rogue picks a lock you didn't plan for. The bard seduces the villain's lieutenant. The wizard asks a question about the local economy that implies they've been thinking about this world more carefully than you have.

In that moment, you need to respond with something that feels consistent with everything that's come before, that honors the player's investment, that moves the story forward without breaking it. AI can't do this. Not because the technology isn't fast enough, but because the response needs to account for the entire history of the campaign as experienced by the specific people at the table. It's an act of synthesis that requires being there.

Know your story's emotional arc

You know that session twelve needs to feel different from session three. You know the pacing should tighten as the arc approaches its climax. You know that the quiet conversation in session nine, the one where the party sat around a campfire and talked about what they were fighting for, needs to echo through the final confrontation.

AI generates content. It doesn't generate trajectory. The emotional shape of a campaign is something you build across months of play, adjusting as the players reveal what they care about, and no tool can hold the thread the way you do. This isn't a limitation of AI specifically. It's a recognition that storytelling across time is a fundamentally human act.

The Shift from Creator to Curator

Here's the part that took me a while to articulate, because it sounds reductive at first: AI changes the GM's role from content creator to content curator.

That sounds like a demotion. It's not.

A museum curator doesn't make the art. But they decide what goes in the gallery, how the pieces relate to each other, what story the collection tells when you walk through it in order. They bring judgment, taste, and a point of view that transforms a pile of objects into an experience. The art matters, but the curation is what makes it meaningful.

When you use AI for prep, you stop spending three hours producing raw content and start spending one hour selecting, modifying, and arranging content that already exists. The blacksmith's backstory gets trimmed. The encounter gets a terrain feature that connects to last session's cliffhanger. The NPC's motivation gets adjusted because you know your party will find the original one too sympathetic for a character who's supposed to be an antagonist.

That editorial work? That's the creative work. It always was. We just couldn't see it because we were too busy doing arithmetic.

The Confidence to Improvise

This is the part that surprised me. I expected AI-assisted prep to make me more prepared. I didn't expect it to make me better at the moments I hadn't prepared for.

But the math is simple. When your NPC roster has fifteen fleshed-out characters instead of three, you have twelve more personalities to draw from when the party goes off-script. When your encounter notes include tactical behaviors and terrain interactions, you understand how combat flows well enough to adjust on the fly. When you've already read through a generated mystery plot with branching outcomes, your brain has rehearsed the narrative logic enough to extend it in real time.

Preparation and improvisation aren't opposites. They're the same muscle working at different speeds. The more structural context you've internalized before the session, the faster and more confidently you can respond when the structure breaks.

The more you prepare, the less you need to follow your prep. Not because the prep was wasted, but because it became the foundation for something more fluid than any script.

What Changes in Practice

The shift isn't theoretical. It's felt most clearly in what you stop doing.

You stop hand-calculating encounter XP. You stop generating NPC names one at a time, each in a slightly different mental context, so the tavern keeper from Monday's prep session feels like she's from a different world than the guard captain you wrote on Thursday. You stop cross-referencing the SRD blocklist every time you write a stat block.

What you keep doing is the work that never felt like a chore. Deciding that the confrontation in the warehouse should happen before the party learns about the betrayal, not after. Choosing which of three generated NPCs gets the personality quirk that mirrors your player's backstory. Noticing that the batch of encounters you just produced has too much combat and not enough negotiation, and swapping one out for a social scene that raises the emotional stakes.

If you want a structured breakdown of how to organize your prep time around this, our session prep toolkit walks through it phase by phase. The short version: the story thinking comes first, the generation fills the middle, and the assembly is where your judgment makes everything cohere. Total time runs about an hour. Most of it spent on decisions, not production.

Is AI Going to Replace Game Masters?

No. And the question itself reveals a misunderstanding about what a GM actually does.

A Game Master is not a content production engine. A Game Master is the person who sits across from four other humans and builds a shared experience in real time, adjusting for emotion, humor, tension, fatigue, the fact that someone brought really good snacks and everyone is in a better mood than last week. That's not a content problem. That's a human problem. And human problems don't get automated.

What AI replaces is the part of prep that was never the point. The hours spent on mechanical tasks that masqueraded as creative work because they sat next to it. The naming. The numbering. The stat block formatting. The SRD cross-referencing. Nobody became a Game Master because they loved calculating encounter XP. They became a Game Master because they loved the moment at the table when the story became real.

AI gives you more of those moments. Not by being creative for you, but by clearing the path so your creativity has somewhere to go.

What Gygax Was Really Saying

Those random tables in the 1979 Dungeon Masters Guide weren't a crutch. They were an acknowledgment that the GM's real job was never to produce content. It was to run a game. To read the table, to pace the story, to know when to push and when to pull back, to make four people feel like they're living inside something that matters.

The tables handled the rest. They handled it badly, by modern standards, with all the narrative coherence of a slot machine. But the principle was sound: give the GM a starting point, and trust them to make it sing.

Forty-five years later, the starting points are better. The coherence is built in. The mechanical precision is handled. And the GM's job is the same as it ever was.

You sit down at the table. Your players look at you. The session begins. Everything you prepped, every stat block and NPC name and encounter skeleton, dissolves into the background where it belongs. What's left is the story. The one that belongs to the five of you and nobody else. The one that no tool will ever write, because the best parts haven't happened yet.