Make a Match-3 Game With AI: Prompt, Edits and Limits
Make a Match-3 Game With AI: Prompt, Edits and Limits
Describe the board, the piece types and the swap rule, and Zugo builds a browser match-3 you can actually play: adjacent swaps, lines of three clearing, pieces falling from above, a running score. A build costs 6 credits and an edit costs 3, so cascades, bonuses and level goals are cheap to add afterwards.
Match-3 survives the trip through a text prompt better than most genres, because its central rule is formal: three in a row disappears. Below is what belongs in the prompt, which edits change how the game feels, and the point where writing sentences stops being enough.
What does one prompt actually give you?
A grid of pieces, a swap between two neighbors, match detection, clearing, and refill from the top. That loop is the genre, and it arrives with the first build. You get something playable rather than a screenshot of a board.
On top of the loop, a single prompt comfortably carries a few more rules: five or six piece types, points that scale with match length, a move limit or a timer, and a target score that ends the round. All of those are numbers, and numbers compress into text without loss.
Decide early whether the player is racing a clock or spending moves, because the two produce different games. Moves make the board thoughtful and let people plan. A timer turns the same board into a scramble. Ask for one of them in the first build and add the other later if you genuinely miss it.
What does not fit into one prompt: a hundred hand-tuned levels, a meta layer with unlockable characters, or a lives-and-energy economy. The wider picture of what generation handles well is in can AI build a game.
What should the prompt name?
Six items. The first three define the game, the last three define its pace.
| What to name | Why it matters | Example wording |
|---|---|---|
| Board | Sets the difficulty floor | "an 8 by 8 grid" |
| Pieces | Controls how often matches happen | "five piece types in distinct colors" |
| Rule | The core of the genre | "swap two adjacent pieces, any line of three or more clears" |
| Cascades | Where the fun lives | "pieces that fall into a new line clear as well" |
| Limit | Creates pressure | "thirty moves per round" |
| Goal | Turns a score into a win | "reach 1,000 points before the moves run out" |
A prompt built from that table: "A browser match-3 puzzle. An 8 by 8 board with five piece types. A swap between two adjacent pieces is only allowed if it forms a line of three or more. Cleared pieces vanish, the ones above fall down, new pieces drop in from the top. If a fall creates another line, it clears too and scores more. Thirty moves per round, target 1,000 points, win and lose screens with a restart button."
The cascade line is the one people forget. Without it the game works and feels flat, because nothing you do ever sets off a chain. Ask for it as its own sentence.
Which edits make a match-3 feel good?
Five, and none of them change the rules. Each is one edit at 3 credits.
Block the pointless swap. If the board lets you trade two pieces where no line forms, players lose a move by accident and blame the game. Ask for the swap to snap back with a short animation when it fails.
Add an idle hint. After a few seconds of no input, highlight one legal move. This single change does more for first-time players than any new mechanic, because it removes the fear of a dead board.
Speed up the drop. First builds almost always animate falling pieces too slowly, and a three-step cascade turns into a wait. Ask for the fall and clear animations to run roughly twice as fast.
Show the points where they were earned. A number floating up from the match reads better than a counter ticking in the corner. Finally, ask for an automatic reshuffle when no legal move exists, otherwise the game quietly stalls and looks broken.
Why is your board boring?
Boredom in match-3 usually traces to one of five causes, and each has a one-line fix rather than a rebuild.
| Symptom | Cause | What to ask for |
|---|---|---|
| Matches form by themselves | Too few piece types | "use six piece types instead of four" |
| Moves never run out | No pressure | "cut the limit to twenty-five moves" |
| Nothing feels big | No special pieces | "a line of four creates a piece that clears its whole row" |
| The round never ends | No goal | "add a target score and a win screen" |
| Every round is the same | No progression | "each round needs a third more points than the last" |
The first row surprises people. It seems like more matches should mean more fun, but a board that solves itself asks nothing of the player and becomes something you watch. Adding a piece type is the cleanest difficulty dial in the genre.
Add special pieces one at a time. Three of them at once turns a tuned board into noise, and you lose the ability to say which addition broke the pacing.
What does a match-3 cost in credits?
A build is 6 credits and an edit is 3. Hi-Fi mode doubles both: 12 to build, 6 to edit. Free gives 5 credits once, which is slightly under a single build, so a finished game realistically starts on a paid plan.
Match-3 needs more edits than an arcade game because animation timing and cascade feel are only judged by playing. A build plus a dozen or so refinements is normal. Pro is $25 a month for 200 credits, which is 33 builds or 66 edits if you spend them only here. Business is $99 a month for 800 credits.
Templates shorten the start. Zugo ships 25 of them and 5 are games, so a working board with a swap already in place can be reskinned faster than a blank prompt can be written. A simple build takes about a minute, and complex projects take several, so plan by number of edits rather than by clock.
How do you make levels feel different?
Change the goal, not the board. Ten rounds that all say "score points" read as one round played ten times, while three rounds with three different objectives read as three games.
The goals that work in this genre are short enough to fit one sentence each. Reach a score within a move limit. Collect twenty pieces of one specific color. Bring an object down to the bottom row by clearing beneath it. Clear marked tiles that only disappear when a match happens on top of them.
Phrase it as an edit: "add a second round where the goal is to collect twenty blue pieces in twenty-five moves." That is 3 credits and it buys more variety than five rounds of a rising score target.
Obstacles are the other lever. Blocks that cannot be swapped but break when a match happens next to them change the geometry of every decision more than a new color ever will. Ask for one obstacle type at a time.
Where does the genre hit a wall?
Four honest boundaries.
- 2D and browser only. No 3D effects, no native app bundle. That is the shape of the product, not a temporary gap.
- A campaign is not one prompt. One round with a goal generates fine. A curve of forty rounds that gets harder in an interesting way is authored work, done edit by edit.
- Cross-device progress needs a backend. Saving in the browser works out of the box. A shared leaderboard means connecting Supabase and treating it as its own task.
- Art is what makes it yours. Piece shapes, background and the sound of a combo carry the identity, and a prompt does not replace them.
Before delivery each build is launched in a sandbox, and one that fails to open is reported as a failed build instead of being handed to you. That removes the blank-screen risk. It says nothing about whether the board is fun, which is a question only playing answers.
What should you do after the first build?
Play three full rounds and note the exact moment you wanted the animation to hurry up. That is the most common complaint about a first match-3 build, and it is one line to fix.
Check the board on a phone. Eight by eight leaves pieces you can still hit with a thumb, ten by ten usually does not. Shrinking the board is a single edit and solves it completely, while enlarging pieces on a wide screen only half solves it.
Then hand the phone to someone who plays this genre regularly. In one round they will find the missing hint, the unclear combo feedback, or the piece colors that blur together, all things you stopped seeing hours ago.
Publishing puts the game on an address like your-game.zugo.run, which opens on phones and desktops, and a custom domain can be attached. The comparison of free hosting options is in publish a game online for free, and a quieter genre to try next is covered in make a memory game with AI. Describe your board at zugo.dev and play a round tonight.