Branching Skies

Stories that remember you.

← All posts

Editorial card reading Three Games Wear One Name, from the post on what an AI RPG actually is, Branching Skies.

What AI RPG actually means in 2026

AI RPG in 2026 means a roleplaying game where an AI fills the role of Game Master. The category splits into three shapes: open-ended generation platforms, chat companions, and curated campaigns. Each serves a different audience. Here's how to tell them apart.

If you've searched "AI RPG" in 2026, you've probably ended up in three different kinds of product without realizing it. They all call themselves AI RPGs. They share approximately nothing in actual experience. The taxonomy below sorts them out. I built one of them. I've spent the last two years using the other two.

1. Open-ended generation platforms

What they are. You type a one-line prompt. The AI generates a world, NPCs, opening scene, and inciting incident from that prompt. Everything you type back generates the next scene.

Who they're for. Creator-class enthusiasts and world-building hobbyists: the "I want to build my own setting and explore it" audience.

What they're good at. Infinite possibility. Wild West RPG with vampire ghosts? Generated. Hard sci-fi mystery on a generation ship? Generated.

What they fail at. Canon. There isn't one. The world is being generated as you play, which means it has no underlying truth. Ask about a city's history twice and you'll get different answers. NPCs reset between sessions. Mythologies drift. The hallucination problem is structural because hallucination is the product.

Platform: mostly desktop and web. Pricing: paid subscription tiers.

2. Chat companions and AI roleplay

What they are. An AI character you talk to. The conversation is the product. Most are oriented around romance, friendship, or casual chat with the AI persona.

Who they're for. People who want a friend or a partner in chat form. The use case is closer to a digital pen pal than a game.

What they're good at. Emotional engagement. The category has gotten good at making the AI feel responsive and warm. Some products have meaningful long-term memory for individual character traits.

What they fail at. They're not really games: there is no quest structure, campaign arc, or mechanical resolution of choices. If you want a "I rolled a 12 to convince the guard" experience, this isn't it.

Platform: mobile-friendly. Pricing: freemium with subscription tiers.

3. Curated AI campaigns

What they are. A hand-built world with an AI narrator running a structured campaign through it. The world's lore, NPCs, factions, and history were authored before the AI ever spoke. The AI is a Game Master inside the authored world.

Who they're for. D&D players. TTRPG enthusiasts. Narrative-game players. People who want the depth and continuity of tabletop in a format that fits a phone and a 20-minute play session.

What they're good at. Memory and world coherence. Because the world is authored, the AI draws on it instead of inventing it. NPCs stay consistent. Lore doesn't drift. Choices have lasting consequences because the world keeps track of them.

What they fail at. Flexibility. You're playing in our world, not your own. If you want to design your own setting and explore it, this category isn't for you.

Platform: Branching Skies is mobile-native, designed for short play sessions. Pricing: free to play with optional paid tiers. No paywall.

Which one is right for you

If you want maximum control over building your own world, the first category. If you want a friend to talk to, the second. If you want a campaign with structure and memory in a hand-built world, the third.

Branching Skies is the third. Pre-launch right now, with the first closed beta next on our build plan.

The taxonomy matters because most of the complaints about "AI RPGs" in 2026 are actually complaints about the first category being marketed like the third. The memory failures, the lore inconsistency, the generic NPCs are all symptoms of open-ended generation being sold as a campaign experience when it's structurally not. If you're evaluating the third category, the seven-point memory checklist and the consequences test are the fastest ways to separate the real ones from the marketing.