“Remind me,” your player says, leaning in with a grin. “What did Garrick swear he’d do if we ever brought his daughter home?”

Four faces turn to you, waiting. You have no idea who Garrick is.

You invented him three sessions ago—a blacksmith, a grudge, a limp—and the table loved him so much they wrote his daughter into their backstories. He’s the best NPC you never planned. And right now, in front of everyone, you’re about to retcon him into a stranger.

That’s improv at the table. It’s the heart of running D&D—you can prep a dungeon, but you can’t prep the moment a player asks your throwaway shopkeeper what he thinks about the war. So you riff. You invent a name, a grudge, a limp. The table leans in. It’s great.

And then, three sessions later, it quietly falls apart—because the best thing you made, you made on instinct, and nothing was there to catch it.

Here are the improv mistakes that do the most damage—and the one that does the quiet damage no one notices until it’s too late.

Mistake #1: Not writing anything down

This is the big one. The silent campaign-killer—and it’s exactly how you ended up blanking on Garrick.

In the moment, you are certain you’ll remember. The name is obviously Garrick. The daughter ran off with a smuggler. He hates the baron. It’s vivid. It’s burned in.

It is not burned in.

So next session, when the player walks back into the forge, you improvise again—and now the daughter is a son, the grudge is against the church instead of the baron, and Garrick has picked up a completely different accent.

The players notice. They always notice. And now you’re doing an awkward retcon in real time, breaking the fiction to patch a hole you made yourself.

The character stops meaning anything. Not because the idea was bad—because the why got lost. An NPC is only as real as the throughline you can remember.

Mistake #2: Railroading the improv

The opposite failure. You improvise an NPC, but secretly you’ve already decided what they’ll do. The players push, negotiate, get clever—and it doesn’t matter, because the NPC was always going to hand over the quest / refuse the bribe / die in the ambush.

That’s not improv. That’s a cutscene with extra steps. Players can feel the rails under the wheels even when they can’t see them, and it teaches them the worst lesson: their choices don’t move anything.

Mistake #3: Trying to fix mush with “no”

Improv comedy gives us “yes, and”—accept what’s offered, build on it. Most DMs absorb this as gospel. And taken too literally, it curdles.

If the answer to every player request is yes, and here’s more, the world goes soft. No friction, no scarcity, no stakes. The party asks the harbormaster for the fastest ship and gets it. Asks the duke for an army and gets it. Nothing is earned because nothing was ever in doubt.

So DMs reach for the obvious antidote: start saying no. Block the request. Soften it to “no, but” and call it nuance.

That’s the trap. Blocking is the original sin of improv. “No” negates what the player just put on the table—it stalls the momentum and quietly teaches them that offering anything is pointless. And “no, but” is just a block wearing a coat: the player still hears the thing you wanted didn’t happen. You haven’t cured the mush. You’ve traded a world with no stakes for a scene with no air.

Here’s the part everyone misses. The mush was never caused by “yes, and.” It was caused by a toothless and—an “and” that only ever adds gifts. A good “and” adds a complication. You keep saying yes to the player’s reality—their competence, their want, their contribution—and you let the world push back through cost, consequence, and its own agenda.

Yes, the ship exists. And it’s short two crew after a brawl, so passage will cost you sweat, not coin. Nobody got told no. The friction lives entirely in the “and.”

A worked example: “yes, and” with teeth

Here’s the situation. The party needs to leave the city tonight, and there’s exactly one ship sailing before dawn: the Gull’s Mercy. As DM, I want them on that ship—the next chapter happens at sea. But if I just hand it over, the scene is mush. So I never say no. I keep saying yes, and—and I let the “and” do all the work.

Player (Vth, the rogue): We need passage on the Gull’s Mercy. Tonight. Whatever it costs.

DM (as Harbormaster Oluf): He doesn’t look up from his ledger. “Aye, she sails before dawn—and she’s chartered full, grain to the last berth. And Reyes came in two hands short after a brawl at the Drowned Crow last night. She won’t sail undermanned, and she won’t wait.” He finally glances up. “So. Depends what those hands of yours are good for.”

I never told them no. Passage exists—yes. The “and” just keeps stacking the world on top: full ship, and a brawl opened two berths, and those berths are work, not coin. Every clause accepts the scene and tightens it. The players didn’t hit a wall; they got handed a lever.

Player (Vth): We can crew. We’re stronger than we look, and the half-orc could haul cargo all night.

DM (as Oluf): “I believe it.” He taps the ledger. “And Reyes won’t sign strangers on my word alone—last lot I vouched for walked off with the cargo. So you’ll be wanting to give me a reason. There’s a crate going aboard that nobody signed for. Tell me where it ends up, and your names go down as vouched.”

Still no “no.” I yes-and their competence—I believe it—and then the “and” raises the cost and, this is the important part, smuggles in the plot hook (the unsigned crate) as something they’re choosing, not something I’m imposing. No rails. They reached for it.

Player (Vth): Done. We watch your mystery crate, you get us aboard.

DM (as Oluf): He stamps the ledger and slides a chit across the desk. “Yes. And tell Reyes that Oluf says you’re worth your salt—she’ll take it from me. And—since you’re already curious—the man who dropped that crate paid in old crowns. Pre-siege coin. Nobody’s spent those in twenty years. Make of that what you will.”

Once they’ve bought in—paid a cost, made a choice—the “and” pays them back with interest: the vouch, and a thread that deepens the mystery they just signed up for. The scene accelerates instead of going slack.

Same NPC, same goal, three turns, and not a single “no.” The friction never came from refusing the players—it came from the world having its own weight, delivered through the “and.” Nobody got railroaded, nobody got blocked, and the party walked onto exactly the ship I needed them on—thinking it was their idea. Because it was.

The part where it all comes back to memory

Here’s the catch that ties every one of these mistakes together.

That scene only matters if Oluf shows up again. If the unsigned crate pays off. If, ten sessions from now, a player says “old crowns—wait, like the harbormaster mentioned?” and you can actually run with it.

Improv that no one remembers isn’t worldbuilding. It’s a really good campfire story you’ll never tell the same way twice.

You can do everything else right—friction that never blocks, a hook they reached for, zero railroading—and still lose it all to Mistake #1. The complication you built, the thread they took, the name of the NPC who handed it over: gone by next session unless something catches it.

Let Loracle remember the why

This is the gap Loracle fills.

You run the game. You riff, you improvise, you let the “and” carry the friction so the scene never goes soft or slams shut. Loracle records the session, transcribes it, and pulls out the entities—every NPC, the promises they made, the threads they dangled.

So when a player asks Garrick about his daughter, you search and you know: her name, where she went, why he’s angry. No retcon. No “I think it started with a G?”

The improv stays yours. The remembering stops being your job.

Riff fearlessly. We’ll keep the throughline.