Let Players Generate Your NPCs

14 Sep, 2026

I struggle sometimes to improvise fast enough at my table. My players meet a new NPC that I have to create on the fly, and in an instant I have to describe their look and their outlook, their ways and their miens. Even when supported by the best spark tables available, I worry that who I create is cardboard-thin.

So I was pleased to recently read some of the best GM advice I've seen from Joe B. It's an attitude, not a technique:

Believe that what you know is enough, and the vast gulf of what you don't know will cease to scare you. With the barrier of fear demolished, the unknown will become accessible to you.

In short, love yourself.

More concretely, Scipio202 recently posted a list of exercises GMs can take to extend themselves:

[P]ick one or two areas to focus on, and each day you try to grow them a little bit past where you are. You do it with intention and reflection - and a healthy amount of giving yourself patience and grace!

In short, build your skills.

BUT INSTEAD, what if we had a procedure to lighten the burden of improvisation?

The Procedure

When encountering a new NPC, one of the players will generate them off a spark table, describe them and their mien. Maybe even offer a line of dialogue.

The burden of creating a new person to inhabit the world -- clothe them in flesh, flesh them in cloth and breathe life into their frame -- this now falls on the players.

The referee will separately generate factors about the NPC that are hidden from player view. This information will be incorporated into the referee's ongoing portrayal of the NPC - from here on, the referee plays the NPC as per usual.

Separate tables exist for the players and the referee to roll on. The player-facing table features easily visible information - the NPC's appearance, their current activity, their mood. Enough for a new character to be introduced. The referee-facing table features information that is not immediately apparent about the new character, and can be used to flesh them out.

Hidden Knowledge and the Joy of Discovery

The goal is to reduce the throughput on the referee while preserving the joy of discovery for players. Players who generate a new NPC will know everything about their appearance, but there are still deeper facts that they will have to work to uncover.

So what belongs on each table? If you think of an NPC as an iceberg, players should immediately be able to see everything above the waterline. (This applies to other game entities too - a trap, a room, a vista, a mysterious object, etc.) Let's use the classic Landmark, Hidden, Secret blogpost as a framework to unpack this iceberg image.

• Landmark information is "automatic and free". The part of the iceberg that juts from the water.
• Hidden information "isn't automatic" and "often isn't free". The iceberg just beneath the waterline. In the right conditions, this might be seen in hazy outline.
• Secret information "is the opposite of automatic, and it's always expensive." The part of the iceberg obscured in the cold briny deep. You'll have to dive to know its contours.

So what goes on each table? The solution seems simple:

• Landmark information is safe to include on the player-facing table.
• Hidden and Secret information is on the referee-facing table. This is only granted to players who go looking for it.

Here's the problem, though: there is no logical connection between the factors the players know about the new NPC, and the hidden facets of their character. How are players to gain a deeper understanding of the NPC?

Hidden information might lie below the waterline of our information iceberg, but some part of each Hidden fact should be visible above the waterline. Otherwise each Landmark fact stands alone, with no logical connection to the deeper facets of the NPC's situation.

For example: take an old veteran travelling incognito. She trudges the road in her dusty travel clothes. She has lined features and a scar down her cheek. She carries a simple but well-maintained sword, marches steadily despite her slight limp, is efficient and fastidious when she makes and breaks camp.

All of these facts are Landmark. Some of the facts point gently to her having once served in an army (a Hidden fact), and astute players may be able to take a hunch and find new evidence to corroborate that theory. There is a logical consistency between the Landmark information as presented, and the Hidden information that players can work to bring to light.

A Handle for Hidden Information

I think about Traps in Chris McDowall's Electric Bastionland: "At least one part of the trap is always immediately visible, typically the trigger or the hazard." Let's call this feature the Handle - something for players to grasp and act upon, to drag the obscured information into the light.

Crucially, for hidden information to work, the Handle should be visible - there must be some hint that something is there that can be investigated further.¹

Let's present this idea another way: For each Hidden fact, there should be some Landmark fact that is a Handle.

If we take for granted that every entity worth exploring in a game has unknown factors, there is a corollary: Some Landmark fact about each entity should point to Hidden information.²

Proceduralising player-facing handles

This presents a problem for our new procedure.

In a standard procedure there's no problem. When the referee generates a new NPC, they know what information is Hidden, and can present a Landmark fact that might point towards it.

But in our new procedure, Players have access to all Landmark facts. To preserve the Joy of Discovery, we can't reveal which are handles. So we're going to have to tie the Player-facing and Referee-facing tables together.

First thought: When a new NPC is encountered, Players deliver 3 Landmark facts, and the referee picks one of them to be the handle of a hidden fact about the new NPC. The trick is that we write our tables so every Landmark fact is a potential Handle for Hidden information.

An example procedure will have to wait, this post has spiralled enough already. Here are some posts I'll be referring to if I write one up:

• Nael at Billhook's hotcake-fresh Landmark Hidden Secret NPC generator
• Forlorn Encystment's offhand comment on Discord, "I really like Traits, Ideals, and Flaws as LHS for social encounters." (Hoping and praying that this soon becomes a blogpost in its own right.)
• Amanda P's Characters as Landmarks with Secrets

Issues

Odd thoughts, mostly critiques.

• Q: If players are surrogate-birthing a character that they will never play, will they bring preconceptions to their interaction?

  • A: Yes, but this is probably good actually.

• Q: Does this pierce the illusion of discovery?

  • A: Maybe? Some players have a lower tolerance for this sort of tomfoolery. In the case where the alternative is for the referee to generate an NPC on the fly, perhaps it doesn't feel that different for players if you hand them the generator themselves. Playtesting would tell.

• Q: The referee's job is now reconciling their own tables with what the players generated... is that harder than creating the NPC in the first place?

  • A: lol maybe

• Q: What do we do about plot-critical NPCs?

  • A: Some NPCs are plot-critical due to their position (like faction leaders). They are drawn up by the GM offscreen. They're not a random encounter, so the procedure doesn't apply. The issue is, it's clear to players that these particular NPCs are pre-prepped; they might as well have a glowing exclamation mark floating above their heads.
    • Maybe it doesn't matter?
    • Perhaps you could still let players still generate the NPC's Landmark features, and have the referee hide the plot-critical functions... dunno, sounds like a bit of work.

• Q: What about a sandbox that incorporates modules?

  • A: Modules come with fleshed-out NPCs. Now some NPCs are described in total by the referee, and it's clear to players that these particular NPCs are pre-prepped, and connected with the module; they might as well have a glowing exclamation mark floating above their heads. Perhaps you could still let players still generate the NPC's Landmark features, and have the referee hide the plot-critical functions... dunno, sounds like a bit of work.

• Q: Could this procedure be used for other systems? To generate biomes, settlements, individual buildings and rooms?

  • Provocative thought!

• Q: Is the juice worth the squeeze?

  • A: lol maybe

Reflection

The whole procedure feels like a gimmick - like my Journalling XP and my weird Mapping Procedure. (Though never forget that gimmicks can work! Josh of Rise Up Comus just gives his players the map when they enter the dungeon; the rule is text in His Majesty the Worm.)

There's a throughline in these systems of taking apart a standard OSR procedure and splitting it down the middle -- giving a player role to 2 players, for instance, or giving half a GM role to the players. In doing this I try to maintain the underlying principle or feeling of OSR play while reducing throughput on the GM. I'm not quite sure why my brain keeps going here. Maybe I should finally read Errant and see how they did it.

It's been a valuable exercise for me though. Thinking in depth about LHS has been fruitful, and I got a chance to chat to Anne about the post and its provenance.

I now think that a Handle does not really come from a simple Landmark fact, but an assemblage of Landmark facts. From our example: the Hidden information (that the NPC is a veteran) really comes from piecing together the disparate Landmark facts - the limp, the sword, the campfire procedures, rather than hinging on a single fact. Hopefully though I've exorcised this particular line of design thought for now, we'll see if it returns.

Reflection-digression

This whole thought process came from listening to Nightcrawler is My Boyfriend, in which Aaron King and Caleb Zane Huett read old X-Men comics and create a game that captures the series' spirit. It's a big vector in my game design thinking at the moment. I'm 3 episodes in, and the hosts have already made the bones of something that feels supers-y.

A big part of the game is providing frameworks for fact creation, typically done by players as a group activity. This is cool - the game procedures form the structure of the story, a skeleton. Then the gameplay is about putting flesh on those bones. Because the skeleton is already there, you know that whatever you create will share the form of the skeleton. (In this case, an X-Men story.)

Unfortunately I don't love Fact-Creation games. I come to the gaming table to solve my issues with a sword, not a pen. But I got to thinking: how far can you push Fact Creation on players while maintaining the JOY OF DISCOVERY? I still wonder.