Companion Characters in Game Narrative Design | Skrib

>

Every other medium gives its supporting cast a protagonist the writer controls. The novelist decides what the hero says in the doorway. The screenwriter decides whether he apologises.

Games don’t work that way. A companion character is the only character in fiction who has to stay in character while their scene partner is improvising.

The player may be brutal for six hours and gentle for the next two. They may skip the conversation that was supposed to establish the relationship, or trigger the emotional payoff before the setup, or never take this companion out of camp at all. The companion has to remain a coherent person through every one of those, without the writer having authored the sequence.

That constraint changes almost everything about how you build one. This sits under our character architecture guide, which covers the interior structure this applies to.

What is a companion character?

A companion is a non-player character who accompanies the player over a sustained portion of a game, reacts to their choices, and typically carries narrative content that unlocks through proximity or approval rather than through the main plot. The distinction from a general NPC is duration and reactivity: companions are present for tens of hours and respond to what the player does.

Companion, sidekick, NPC

The terms blur in practice. The useful line is reactivity: a companion’s content is gated on player behaviour, which is what creates the design problems the rest of this article addresses. A sidekick who accompanies the player through a linear campaign without responding to their choices is closer to a screen character and can be written like one.

Why the companion carries the arc

The player character is frequently flat by design - a cipher, a silent protagonist, or a character whose personality the player supplies. That’s a deliberate choice, because projection is part of what games trade on.

But flat protagonists don’t arc, so the emotional change has to live somewhere else. In games, it lives in the companion.

This is the structural reason companions matter disproportionately. They aren’t accessories to the story; they’re where the story’s arc is kept. A player who remembers a game emotionally is usually remembering a companion, because that’s the only character who visibly changed.

Character arcs explained covers why flat arcs relocate change to the supporting cast rather than eliminating it - the mechanism is the same one procedural television has used for decades, and games inherited it for different reasons.

The design implication: if your player character is a cipher and your companions are also flat, nothing in the game changes, and the story reads as a sequence of events. At least one companion needs a real arc.

Build the architecture to survive any protagonist

The standard character build applies - a formative wound, a false rule drawn from it, a want, a contradiction. The pillar covers construction; what changes here is the robustness requirement.

In linear media, a character’s false rule needs to generate a reaction to the events you wrote. In games, it needs to generate a reaction to every category of thing the player might do - including things you’d rather they didn’t, and things they’ll do in an order you didn’t anticipate.

The practical test is a coverage pass. Take your companion’s rule and check it against the categories of player behaviour your game permits:

• Cruelty to someone weaker

• Mercy that costs the party

• Dishonesty for gain

• Refusing a reward

• Ignoring them entirely for twenty hours

If the rule doesn’t produce a distinct, predictable reaction to each, the companion will go silent or generic somewhere - and generic is where the character dies, because the player experiences it as the illusion breaking.

Kim Kitsuragi is the reference case for a companion who never breaks. Whatever the player does, however erratically, the reaction is legible as coming from one specific person with a consistent professional and moral framework. That coherence is architectural, not written line by line.

States, not sequences

Here’s the structural consequence that most affects a writing plan.

You cannot write a companion arc as a sequence of scenes, because the player controls the order and may skip most of it. What you write instead is a set of states, each with entry conditions, each internally complete.

A state defines: - What the companion currently believes about the player - What they currently believe about themselves - their position on the false rule - What content is available while they’re in it - What moves them out of it, in either direction

Each state has to be a legitimate resting place. A player may sit in state two for thirty hours and finish the game there. If state two only makes sense as a waypoint, that player’s experience is broken and they’ll never know why.

This is the single biggest adjustment for writers arriving from film or prose. A screenplay’s second act is unplayable in isolation. A companion’s second state has to work as an ending.

How those states connect to dialogue structure is a systems problem - Writing Branching Dialogue Trees Without Losing Narrative Control covers construction from that side.

The presence problem

A film character is on screen for perhaps forty minutes. A companion is present for forty hours.

Characters designed for film-scale exposure exhaust the player at game scale. A distinctive verbal habit that’s charming in hour two is grating by hour nine. A defining trait shown four times reads as depth; shown two hundred times it reads as a loop.

Two mitigations, both structural rather than cosmetic.

Depth reserves. Write more interiority than you plan to surface, and release it slowly against duration rather than against plot. A companion should still be capable of surprising the player in hour thirty, which requires having withheld something.

Reaction variety over content volume. More dialogue is not the answer; more kinds of response is. A companion with five reactions to one situation feels thinner than one with one reaction each to five distinguishable situations.

Companions also generate large volumes of repeated ambient lines, which is a distinct production and writing discipline - Barks and Ambient Dialogue covers it.

Reaction as characterisation

In linear media, characters are established through scenes. In games, companions are established through hundreds of small reactions encountered in unpredictable order.

That shifts where the craft effort goes.

Reactions must be individually complete. A player may encounter a companion’s third-most-important reaction first. Each one has to characterise on its own, without depending on the player having seen anything prior.

Silence characterises. A companion who says nothing when the player does something cruel has said something. Designing which situations produce no reaction is as much a characterisation decision as writing the lines.

Reactions to the player are the primary channel. More than backstory, more than personal quest content, the player learns who a companion is from how that companion responds to them.

Where multiple companions travel together, the party becomes an ensemble, and keeping several of them audibly separate is its own discipline - ensemble casts covers the differentiation problem.

Approval systems and the resource problem

Many games gate companion content behind an approval or affinity value. It’s a proven structure and it carries an honest cost worth naming.

Approval converts a character into a resource. Once a player understands the system, some proportion of them will optimise it - choosing dialogue to raise a number rather than to express anything. At that point the relationship the system was built to model is being simulated by a player gaming it.

Two things reduce the damage:

Make approval opaque or partial. Precise numeric feedback invites optimisation. Ambiguity keeps players choosing from character rather than from arithmetic.

Gate content on behaviour, not accumulation. A companion who opens up because the player did a specific thing they cared about reads as a person. One who opens up at 40 points reads as a lock.

The deeper point: approval systems work best when disapproval is also content. A companion who only rewards is a vending machine. One who can be disappointed in the player, visibly and durably, is a character.

Where companions fail

Generic fallback. The companion has no reaction to a category of player behaviour and defaults to neutral. The most common failure, and the most damaging, because it’s the moment the character stops existing.

The unearned confidant. A companion who discloses their deepest history early, unprompted, because the content had to go somewhere. Disclosure needs demand behind it - the same principle covered in backstory dumps vs. earned reveals.

Loop exposure. A trait repeated past its tolerance. Fixable in design, nearly unfixable in VO once recorded.

Arc gated behind optional content. The companion’s change is real and most players never see it, so the game reads as arcless to the majority.

Sycophancy. A companion who approves of everything. Without the capacity to be disappointed, they cannot characterise the player’s choices, which is most of their job.

Reactivity without memory. The companion responds to the last thing and forgets the pattern. Players notice cumulative amnesia faster than they notice missing content.

Keeping state maps beside the writing

Companion writing has a shape that fits a linear document badly. States, entry conditions, reaction coverage and approval gates are a graph; a draft is a line. Most of the failures above are traceable to a writer holding that graph in memory while working in a text file.

Skrib is built for that adjacency. The Board is a canvas that suits a graph better than a document does - a companion’s states can sit as connected cards with their entry conditions, alongside a reaction-coverage grid and reference material, while the Draft stays open beside it. Dialogue and bark tables come in and out as .csv and .xlsx, which is where most of this work actually lives, and the Library keeps the world bible and every previous draft one search away.

Frequently asked questions

What makes a good video game companion character? Coherence under unpredictable player behaviour. A companion needs an internal rule that generates a distinct reaction to every category of thing the player can do - including cruelty, mercy, dishonesty and being ignored. Where that coverage has gaps, the character falls back to generic and the illusion breaks.

How is writing a companion different from writing a sidekick in a film? The writer doesn’t control the protagonist. A film sidekick reacts to authored behaviour in an authored order; a companion reacts to an improvising player who may skip content or encounter it out of sequence. That’s why companion arcs are built as states rather than scene sequences.

Should companions have their own character arcs? Usually yes, and often they must. Player characters are frequently flat by design so players can project onto them, which means the emotional change in the story has nowhere else to live. If both the protagonist and the companions are flat, nothing in the game changes.

How do you stop a companion becoming annoying over a long game? Write more interiority than you surface and release it against duration rather than plot, and favour more kinds of reaction over more lines. A trait that reads as depth across four appearances reads as a loop across two hundred.

Do approval systems make companions better? They’re effective and they carry a cost: once players understand the system, some will optimise the number rather than choose in character. Opaque or partial feedback helps, as does gating content on specific meaningful behaviour rather than accumulated points - and making disapproval into real content rather than an absence of reward.