Search "AI board game maker" and you'll get two dozen results that are the same thing wearing different logos: a text box wired to a general-purpose model, handing back a themed paragraph, a list of plausible-sounding rules, and some art. Free, no signup, and useless past the first ten minutes.
The reason is that a board game is a system, not a document. Mechanics have to interact without collapsing, components need real counts, the numbers need a balance curve, and the rulebook has to survive a stranger who has never heard you explain it. Generating text that describes a board game is easy. Generating one that works is much harder, and only a couple of the tools below are really trying.
ShuffleKit
Best for: AI aimed at the boring half of a real design tool, not at the design itself.
ShuffleKit designs the whole box, not just the deck: cards, boards, mats, tokens and tiles, dice, meeples, score pads, screens and tuck boxes, each designed in the same editor, playtested on a shared online table, then exported to print or ordered through The Game Crafter.
The AI is narrower than that. It generates artwork, drafts and edits card data in bulk, suggests a card layout and wires it to your columns, keeps design documents beside the project, and will argue about mechanics with you. What it won't do is generate a board, lay out a tile set, or hand you a finished game.
That scoping is deliberate. Bulk data entry, artwork and layout wiring eat hours without being the fun part; the design itself is the fun part. It's free to start, with paid plans from $8 a month and Print & Ship on the $20 tier, and you keep full commercial rights to what you make, AI artwork included.
GameGrammar
Best for: a structured first draft of an actual game, not a description of one.
From Dynamind Research, and the only tool here built for the structural problem rather than the prose one. You give it a theme and some constraints, and a pipeline of specialised agents hands off to each other: a MechanicsArchitect designs the core systems, a ThemeWeaver integrates the theme, a ComponentDesigner specifies the physical pieces, a DetailsArchitect writes the turn structure, then a BalanceCritic and a FunFactorJudge rate fairness and engagement. It works from 35 mechanisms across seven categories, with suggestions grounded in the top 2,000 games on BoardGameGeek.
What comes out, in under a minute, is a structured draft: mechanics, a component list, scoring, turn phases, and a critique carrying balance notes and a fun rating out of ten. Exports run to Tabletop Simulator, JSON, Markdown and PDF. It's free during beta, with 500 credits and 2 projects on the free tier plus a 7-day Pro trial. Pro is $19 a month and Studio $49, which is where the conversational co-designer and the AI playtesting sit.
Treat it as a fast way to get a starting point and a checklist of suspicious numbers. Two caveats. Those balance scores are a model's opinion of your rules, not evidence from play. And a generated first draft tends to land on the safe, familiar mechanisms rather than the odd idea that makes a game worth playing, so it's a tool to think with rather than one to ship.
ChatGPT and Claude
Best for: the rulebook, which is the component most likely to sink you.
These are the workhorses, and they land on the use designers accept most readily. Editing and proofreading a rulebook is the least controversial application of AI in tabletop, and both are very good at it. Paste your rules in and ask what's ambiguous, what a new player would misread, or which edge case you haven't covered. That matters more for a board game than a card game, because the rulebook carries the whole system and you're the one person alive who can't read it fresh. As a thinking partner they earn their keep too, less by having good ideas than by making you articulate yours.
The hard limits are the same for both. They don't know your component list, can't lay out a board, can't apply a design across 80 tiles, can't playtest, and can't hand you a print file. And treat the maths with real suspicion, because both are unreliable at probability and confident about it either way. Use them to write the formula, then run the numbers in a spreadsheet yourself.
Gemini and Midjourney
Best for: component art that still looks like one game across forty pieces.
Art is where the general tools do their best board game work, and the requirement differs from a card game's. A board game isn't one image. It's a board, forty cards and a dozen tile types that all have to look like they came from the same world, so consistency beats peak quality.
Gemini is the strongest for iterating, because it's unusually good at changing one element of an image while leaving the rest intact. Once you have a token you like, "same style, same lighting, swap the icon" actually works. Midjourney still sets the bar for quality and has reusable style references, which pin a look and reapply it across every component in the box.
Read the rights, because they differ. Google won't claim ownership and allows commercial use. Midjourney has no free tier at all since it pulled the trial in 2023, so it's $10 a month at minimum, every paid plan carries commercial rights, and companies over $1 million a year in revenue have to be on Pro or Mega. Your images also sit in the public gallery unless you're on Pro or Mega, which is where stealth mode lives. Worth knowing that AI art in a finished, sold game still draws a strong reaction in this hobby, so plenty of designers keep it to prototypes and commission the final set.
What AI still can't do
Notice what none of it replaces: actual players. GameGrammar's Studio tier does now ship AI playtesting, which is further than this field was a year ago, but that's a model reasoning about your rules rather than four people discovering that the trading phase drags. Plenty of tools will put your prototype on a table, ShuffleKit included, and that still means finding humans and running the session.
That's the honest state of it in 2026. AI is good at the draft, the edit, the placeholder and the formula. A model that runs your rules a thousand times can catch a broken strategy, but it cannot tell you the table went quiet in round three.
For the wider tool landscape, see the best board game makers, or our guide on how to design a card game.
Want AI that does the drafting and the bulk data inside a tool that also designs your components, tests them and prints them? That's what we built ShuffleKit to do, and I'd love for you to try it and tell me what you think. Start free.
