One of the most consequential design choices in Baldur's Gate 3 sits quietly in the background: a reputation system that stops you doing the very thing the game otherwise encourages. Larian Studios permits extraordinary player freedom across environmental puzzles, quest solutions, and moral choice. Yet when it comes to companions, the studio enforces boundaries.
"Sometimes the answer is to just let the player break the game, but when it comes to party members, usually not," says writer Kevin VanOrd. This principle shapes everything from individual NPC arcs to the broader mechanics by which relationships develop and deepen.
The system itself is familiar: party members in Baldur's Gate 3 communicate their opinions through pop-ups like in Dragon Age or Knights of the Old Republic 2. When you speak in favour of charity or cruelty, the companions react visibly. Approval accumulates or declines. But beneath that readable surface lies a more sophisticated constraint.
If a character trusts you enough, they'll go along with whatever you say," VanOrd explains, addressing why reputation exists. The inverse is true: companions with low approval will resist or refuse certain actions. This prevents a particular failure mode: the player coercing a character into unwanted behaviour through social pressure.
The practical writing problem is acute. "We can't have every character reacting to every situation, otherwise we would be working on the game forever. So this is a way of showing how they feel without writing new dialogue." The approval system serves as a compression tool. Rather than script unique dialogue for every companion in every scenario, the game communicates emotional state through a visible meter and branching conversation trees.
Where the system becomes philosophically interesting is in romance. "But if you need to convince them, that's when it becomes a dice roll. Every time you see the 'Lae'zel approves' or 'Lae'zel disapproves' pop up, you learn a little something about that character," VanOrd notes. Once approval reaches sufficient levels, romantic progression shifts from deterministic to probabilistic. This appears deliberate: it preserves emotional authenticity at the cost of player predictability.
The reasoning is sound. The system meshes with the Dungeons & Dragons inspiration behind Baldur's Gate 3, keeping players on their toes with dice rolls where necessary but also allowing fans to build meaningful relationships with companions. Pure approval-meter romance would reduce intimacy to accountancy: gift the right items, say the right lines, unlock the romantic scene. By introducing chance at higher approval thresholds, Larian preserves the sense that acceptance remains contingent and earned.
What goes unmentioned is the tension this creates. The system constrains player agency in service of companion agency. You cannot talk an ideologically opposed companion into supporting your goals indefinitely. You cannot make Astarion endorse pure charity through brute-force persuasion, just as you cannot earn Wyll's approval by causing needless suffering. This is design discipline, not mere limitation.
The approval framework also reflects developmental realism. Large-scale RPGs with branching dialogue face exponential complexity: each new dialogue tree multiplies the permutations writers must account for. By standardising how companions signal their responses, Larian avoids a spiral in which each new scene demands entirely fresh companion reactions. The dice-roll mechanic serves practical necessity while preserving the games foundational design philosophy.
Whether players experience this as a satisfying compromise or a constraint depends on perspective. For some, it enables the feeling of genuine relationship negotiation. For others, it may feel like an imposed ceiling on what romance can become. The system respects neither pure determinism nor complete freedom. Instead, it occupies the difficult middle ground where meaningful player choice meets believable NPC autonomy. That balance, Vanord suggests, requires both design discipline and the occasional acceptance of controlled randomness.