Plenty of games offer choices. What separates Baldur's Gate 3 from most of them is not the number of decisions it puts in front of you — it is that the game is built to keep going after you get things wrong.
That sounds like a small distinction. It is the whole design.
Failure is content, not a mistake to undo
In most RPGs, a failed check is a wall. You reload, pick a different dialogue option, and the story proceeds along the path the writers actually cared about. The failure state exists to be avoided.
Baldur's Gate 3 mostly refuses that. Fail a Persuasion roll and the conversation does not stop; it curdles. Fail to save someone in the opening hours and they stay dead for a hundred more, with other characters referring to it. Get caught stealing and the shopkeeper remembers, adjusts prices, and eventually stops dealing with you altogether.
The practical effect is that the game stops feeling like a test you are being graded on and starts feeling like a situation you are inside of. You are not looking for the correct answer. You are finding out what happens.
This is expensive to build. Every failure branch is content that most players will never see, which is precisely why so few studios commit to it. Larian's willingness to spend that money is the actual achievement — not the branching itself, but the volume of writing dedicated to outcomes the design does not reward you for reaching.
The dice make the choice public
Rolling visible dice for conversations was the loudest early criticism, and it turned out to be the game's best structural idea.
A hidden skill check is a quiet negotiation between you and the systems. A visible d20 is a commitment. You see the number you need, you take the swing, and the result belongs to you rather than to a spreadsheet you cannot inspect. It converts dialogue from menu navigation into something with the texture of gambling.
It also, quietly, teaches the ruleset. After twenty hours you have internalised what a +7 to Deception actually means, which is more D&D literacy than most tie-in games ever manage to transmit.
The honest counterpoint: inspiration re-rolls blunt this. Once you have a stack of them, a failed roll becomes a suggestion rather than a verdict, and the tension the visible dice created leaks away. It is a reasonable accessibility valve, but it does soften the very thing that made the system distinctive.
Companions who are not waiting for you
The party writing is where the depth is most obvious, and it comes from a single decision: your companions want things independently of you.
Astarion's arc is not a reward you unlock by being nice to him. Lae'zel will argue with you about her own culture and can be wrong in ways she does not resolve. Wyll's obligations conflict with the party's interests, and the game does not arbitrate. They approve and disapprove of choices in ways that are legible but not gamified into a simple meter you can farm.
Crucially, they can leave. A companion walking out over something you did is one of the strongest signals a narrative game can send, because it costs you a party member — a mechanical loss, not just a narrative one. The game is willing to make you worse at playing it because of who you decided to be.
Where it frays
Act 3 is the strongest argument against pretending this is flawless.
The density that makes the earlier acts feel handcrafted becomes a liability in a city. There are too many threads live at once, and some resolve with noticeably less care than they were set up with — plotlines that clearly expected more room than they got. The reactivity is still there, but it thins.
There is also a category of choice that looks consequential and is not. Some decisions produce a line of acknowledgement and then rejoin the same path, and once you notice the seam you start seeing it elsewhere. That is not unique to this game; it is just more visible because the surrounding standard is so high.
And the launch state was rough in the third act specifically — the part of the game most dependent on many systems agreeing with each other at once.
What it actually changed
The temptation is to treat Baldur's Gate 3 as a new baseline every RPG must now meet. That framing is not useful, because most of what makes it work is budget and time spent on content players will never see, and that is not a design lesson anyone can simply adopt.
The transferable lesson is cheaper than full branching: let the game continue after failure. Acknowledge what the player did, including the parts they did badly, and do not quietly route them back to the intended path. That is a writing and scoping discipline more than a technical one, and it is available to studios that will never have Larian's resources.
Choice is not really the point. Consequence is. Baldur's Gate 3's contribution is demonstrating, at scale, that players will happily live with bad outcomes as long as the game admits they happened.