Intro
I have been on a pretty big Game of Thrones kick lately. I just finished A Knight of the Seven Kingdoms — which, by the way, I heavily recommend to anyone who likes fantasy, full stop — and it sent me straight back into a rewatch of the original series.
I came to Game of Thrones late. I didn’t start until season three, and I hadn’t read the books, so I marathoned seasons one and two to catch up — arriving just in time for the Red Wedding. Imagine that being one of your first real experiences with the series when the only fantasy you’d loved up to that point was The Lord of the Rings.
So I’m rewatching those early seasons, full of genuinely masterful subversions of expectations. And then I get to the end of the show and see… well. Not masterful subversions of expectations.
It got me thinking about us. Because we’re storytellers too, and we want to surprise our players. We want those moments where the table is slack-jawed, staring at us while we describe something horrific or beautiful that nobody saw coming.
Here’s the thing: there is a right way to subvert expectations as a DM, and there is a very wrong way. I have done both, because apparently as a Dungeon Master — scratch that, as a person — I have to learn every lesson the hard way. So hopefully I can save you some trouble.
First, a Clarification: This Isn’t About Railroading
Before we get into it, I want to be clear about something, because I know where this conversation usually goes.
I am not telling you to railroad your players. There have been a million videos about railroading versus open world, and that’s not what we’re doing here. If you’ve watched my stuff, you know I don’t think railroading is inherently bad — there’s a lot of stigma around the word, when really we’re describing a clearly defined narrative that players follow along a fairly straight path. There’s nothing wrong with that, as long as the players still have agency.
I’m also not saying you should manipulate the story to force specific outcomes. Your players’ agency is sacred. What we’re talking about is how you present the world and the NPCs in it — narrative framing — not how you control what the players do.
Remember, these are real people at your table. They aren’t characters in a book you’re writing. They’re your friends. Their choices should matter.
The One Rule
Everything below comes down to a single line, so let’s put it up front:
You can subvert expectations about the world and how NPCs behave. You cannot subvert expectations about payoffs your players have earned through gameplay.
Hold onto that. Now let’s look at the three places this comes up most.
Scenario 1: Don’t Steal the Villain Kill
Let’s start with one I see a lot of DMs make, and one I have absolutely made myself.
Picture the end of The Two Towers into the beginning of The Return of the King. Your party has been chasing an evil wizard for months. Maybe he’s tied to a character’s backstory. Maybe he’s been the source of every problem in the campaign. Either way, he has been a genuine thorn in their side, and he is unambiguously the villain.
The party storms his tower — weapons drawn, spells prepared, ready for the confrontation they’ve been building toward. And then an NPC advisor walks in and shanks him. Maybe the wizard said something cruel one time too many. Maybe it’s "I was working against him the whole time." Credits roll.
In The Lord of the Rings, that works. In a D&D campaign, your players are going to feel cheated — and they should be.
Why it fails:
- The players have ownership over that confrontation.
- They’ve invested hours of gameplay, character development, and emotional energy into reaching it.
- You’re not subverting a trope. You’re subverting the actual payoff they’ve been working toward.
It’s like spending sixty hours on Elden Ring and having someone wander in while you’re away and beat the final boss for you. Without asking. That kind of sucks, right? That’s what your players feel.
The right way: an example from the Crimson Court
My party got word that a powerful wizard was under siege by cultists, so they rushed to help. They expected a desperate rescue mission.
When they arrived, the wizard was completely indifferent. "Why are you here?" Zephyrus asked, entirely unbothered. When they explained they’d come to save him, he laughed. "Save me? I have teleportation circles in this tower to a dozen of my estates. If I wanted to leave, I could leave at any time. These cultists think they have me trapped."
See the difference? The expectation was subverted — no epic siege, no desperate rescue — but the players didn’t lose anything that was theirs, except the assumption that they had a big fight on their hands. They still chose how to approach the situation. They still had agency in what came next.
The subversion was about the NPC’s situation, not about taking away a player’s moment.
Scenario 2: The Backstory Villain, and the Line You Can’t Cross
This one shows up at nearly every table. During session zero, your rogue tells you their mentor betrayed them and left them for dead. Your cleric’s temple was burned down by a specific warlord.
Those backstory elements become threads in your campaign. Players get invested. They want the confrontation. They want to look their enemy in the eye and get resolution.
So here’s what you don’t do:
- Don’t have that villain die off screen.
- Don’t have another NPC take revenge on the player’s behalf.
- Don’t have another player character kill the villain from someone’s backstory.
- And for the love of whatever deities you observe, don’t reveal that the villain wasn’t actually that bad without giving the player a meaningful choice in how to respond.
You might be noticing a through line.
A good subversion vs. a bad one
Ned Stark dying at the end of season one is a great subversion. It overturns our expectations about the kind of story we’re watching. We expect the honorable hero to win because of his honor, or at minimum to make it north to the Wall. Instead, the honorable hero dies because of that honor, and because the world is cruel. It teaches us something about the world we’re in: good men die here.
Arya killing the Night King in season eight is a bad one. It took the payoff away from characters who had been built toward that exact purpose across multiple seasons. The audience felt robbed of the confrontation they’d been promised — Azor Ahai, the prince that was promised, all of it.
And keep this in mind: D&D is a fantasy story, and one of the through lines of almost all fantasy is that in the end, the good guy wins. In something like A Song of Ice and Fire, that win might be Pyrrhic — it might not feel like a victory at all — but they still come out the other side. We usually know the good guy is going to win. The story is about the process of getting there. So if you spend a campaign setting things up and then snatch it away at the last minute, of course that doesn’t feel good. There’s no fulfillment in it.
So how do you subvert a backstory?
You absolutely can — you just do it in ways that give the player more agency, not less.
Say your fighter’s mentor betrayed them. The wrong subversion is showing up to find someone else already killed them. The right subversions:
- The party tracks the mentor down and finds them dying of some unrelated illness.
- The mentor is genuinely remorseful and wants to make amends.
- The mentor reveals the fighter’s parents were the real villains, and the betrayal was justified.
All three subvert the simple revenge story. All three hand the player a decision. Do they forgive? Do they take revenge on a dying person anyway? Do they sit with uncomfortable new information about their own past? That’s the player’s call. They still have ownership.
Your job is to provide unexpected complications, not unexpected solutions. The players are the ones who provide the solutions.
A great example in media is "The Southern Raiders" from Book Three of Avatar: The Last Airbender. Katara goes with Zuko to find the man who killed her mother. We expect a monster. We get a pathetic old man who begs and weeps for his life at the first sign of a challenge. The expectation is subverted completely — and the choice of what to do about it stays exactly where it belongs.
Scenario 3: Quest Objectives — Subvert the How, Not the Why
This is where I think a lot of DMs get nervous about subverting anything at all.
Your players hear a village is being attacked by gnolls. They rush to help. Classic D&D.
Bad subversion #1: They arrive and every villager is already dead. Quest over. Nothing they could have done, no clues to find.
Bad subversion #2: They arrive and another adventuring party already killed the gnolls. Thanks anyway!
In isolation, both are bad for the same reason: they remove player agency retroactively. The players chose to help, and you’ve told them the choice didn’t matter. (Also — why did you present the quest at all?)
The Crimson Court example from earlier is the model. The quest was "rescue the wizard from the siege." The subversion was that the wizard didn’t need rescuing. But critically, it didn’t end the quest. It complicated it. Suddenly my players had new questions. Why is this wizard letting himself be besieged? What is he learning about these cultists? Do we still fight them? Should we be worried that he isn’t worried?
The subversion opened up more choices, not fewer.
Fixing the two bad versions
If the village is already dead, let them collect clues. This probably isn’t the only village these gnolls will hit. Are they working for someone? Maybe a few of the dead ones carry brands. Maybe they’re carrying good steel instead of the bronze you’d expect — somebody has been funding upgrades. Now there’s a question worth chasing. Who’s been buying weapons in town? Which blacksmith has been selling?
If another party already handled it, ask how they heard about it first. Were they just passing through? Or is something else going on — maybe embezzlement inside the adventurers’ guild, where they capture gnolls, release them near a village, send in their own people to "save" the day, and collect the reward. And then do it again. Maybe the adventurers smirk on their way out.
The point: when you tell your players to go do a thing and then say "never mind, it’s solved" and give them nothing else — yeah, that’s going to suck. Don’t do that. Even if you think it’s funny.
You want them thinking "oh, this is more complicated than I thought." That’s the secret sauce. That’s what they put in the Krabby Patty.
The Ownership List
Here’s the whole thing in one place.
Players own:
- Confrontations with their backstory villains
- The resolution of quests they’ve actively pursued
- Combat encounters they’ve prepared for
- Revenge they’ve been seeking
- Relationships they’ve been building
Players don’t own:
- How NPCs behave
- Whether situations are what they initially seem
- How complicated the problems they encounter turn out to be
- The motivations of antagonists
- The state of the world when they arrive somewhere
Don’t subvert expectations by taking things away. Subvert by giving them more complicated choices to make.
Closing Thoughts
We all want shocking moments at our tables. We all want our players telling stories about the time we completely surprised them — we want them to feel something more than "I’m going to roll my dice and hope I hit."
The difference between a good surprise and a bad one is simple: good surprises give players more interesting choices. Bad surprises take choices away.
I’ve been a teacher, so I know you have to say a thing nine times before half the room hears it once. So: the Ned Stark moment works because it teaches you about the world. The Arya and the Night King moment doesn’t, because it robs characters of an earned payoff. In D&D this matters even more, because your players aren’t watching — they’re in it.
Don’t subvert their participation. Subvert their assumptions about how the world works, and then let them respond however they want.
If there’s something you’re running into at your own table, drop it in the comments on the video. I think it’s genuinely useful to hear what other DMs are dealing with, because we learn from each other’s mistakes — and trust me, I’ve made enough to know that learning from other people’s is always the better deal.