Can AI Really Make a Video Game? We Put It to the Test

Online, two camps are shouting past each other. On one side, dazzling demos: "I made a game in one minute with ChatGPT." On the other, irritated developers pointing out that a video game is not three hundred lines of JavaScript.

Both are right, and that's exactly what makes the question interesting. So we took it seriously: we asked AI models to build games, we played the results, and we wrote down what held up and what fell apart.

The Real Question: Can AI Make a Complete Video Game, or Just a Chunk of Code?

First we need to agree on words, because "video game" covers two very different things.

The studio game: out of reach

Dozens of hours of content, whole teams, proprietary engines, six-figure budgets. Nobody serious claims an AI produces that today, and it isn't what this article is about.

The mini-game: the real testing ground

A few minutes of fun and one solid mechanic. That's the format an AI can realistically aim for, and the one we focused every test on.

So our question was the useful version of the question: can an AI produce, on its own, a complete and genuinely playable mini-game?

Our Test Protocol for Making an AI Create a Game

We deliberately kept a method anyone can reproduce at home.

Five genres tested: an obstacle runner, a brick breaker, a quiz, a 2048-style puzzle, a small platformer.
Three AI models: ChatGPT, Claude and Gemini, in their consumer versions.
One absolute rule: not a single line of code written or fixed by hand. Everything through plain-language requests.
One imposed format: a single self-contained HTML file, no external resources, so the game opens in any browser.
One success criterion: the game must be playable start to finish by someone who read no instructions, on desktop and on a phone.

Worth stating plainly: these results are qualitative, and they shift between models and from month to month. Run the test yourself and you'll see.

What AI Genuinely Gets Right When It Makes a Video Game

Simple-rule games: clear success

Runner, brick breaker, Snake, fixed shooter: in these cases the first draft is almost always playable. The game loop works, the score increments, the round ends. The result isn't a broken prototype, it's a game.

That makes sense: these mechanics are so well documented that models reproduce them reliably. If you're starting out, start here. A Snake or a brick breaker is the best possible first attempt.

Quizzes and puzzles: excellent

This is where AI is most impressive, because it generates both the code and the content. A twenty-question quiz on a specific topic, with scoring and explanations, arrives complete on the first try.

Polyzzle Shapes, a puzzle game published on Hoora with its statistics

The result is nothing like a prototype: this geometric puzzle game published on Hoora has passed 42,000 views and over a thousand likes. Puzzles are exactly the terrain where AI excels, because the logic is clean-cut and the content gets generated along with it.

Screens and polish: very good

Title screen, end screen, best score kept for the session, clean layout: ask and you get them. This is what makes a game look like a real game, and AI does it very well.

What AI Gets Wrong When You Ask It to Make a Video Game

Four flaws show up systematically, whichever model you test:

Game feel: this is where it breaks. The main weakness. AI almost always miscalibrates: games that are brutal in the first second, mushy jumps, excessive speeds. The code is correct, the fun isn't there. Good news: these are the easiest flaws to fix in plain language. "Slow the start down and ramp the difficulty up gradually" is often enough. We break these settings down in our guide to creating a platformer with ai, the genre that demands the most on this front.
Mobile: forgotten by default. Across the five tests, no game was usable by touch unless we asked explicitly. AI writes for a keyboard. You have to demand touch controls and a vertical full-screen layout, every time.
Long projects: losing the thread. Beyond a certain size (roughly three or four interlocking mechanics and several hundred lines), fixes start breaking what already worked. You fix the jump, the enemies stop spawning. That's the hardest limit of the whole approach.
Graphics, sound, multiplayer. Visuals stay geometric: shapes drawn in code, not hand-drawn sprites. Sound works if it's generated in code, not if you want real music. And online multiplayer is out of reach: it needs a server, which means actual development.

The Verdict: Yes for a Mini-Game, No for a Studio Game

After the test, the answer is clear and not remotely binary.

Yes, for a complete and publishable mini-game

An AI can create a playable mini-game from start to finish on its own. Budget fifteen minutes to two hours depending on the genre, with two or three rounds of fixes. All five genres ended up playable: quizzes and puzzles effortlessly, the platformer only after several iterations on the jump.

No, for an ambitious project

It replaces neither a studio, nor an engine, nor a developer as soon as the project grows. And that's fine, it simply isn't the same category of product.

What the test really taught us

The conclusion is elsewhere, and slightly unexpected. The hard part is no longer building the game: that's solved, and it's become nearly trivial. The hard part is finding people who play it.

Reproduce Our Test: The Prompt We Used to Make an AI Create a Video Game

Copy it, replace the brackets, and judge for yourself:

Create a complete web game in a SINGLE self-contained HTML file.
Mandatory technical constraints:
- all HTML, CSS and JavaScript in the same file
- no external resources (no images, no fonts, no libraries)
- playable with a keyboard on desktop AND by touch on mobile, full screen
The game: [your idea in two sentences]
Rules: [how to win, how to lose]
Difficulty: easy for the first 15 seconds, then progressive
Display: score and best score at the top, end screen with a "Play again" button
Style: [colors, mood]
Give me the complete file, ready to open in a browser.

The three fixes that improved our results the most

"It's too hard from the start, slow it down and make the difficulty ramp gradual."
"Add touch controls and adapt the layout to phone screens."
"Add an end screen with the score, the best score and a Play again button."

Publishing the Video Game AI Made for You

Since the real obstacle is no longer building, you may as well tackle the one that's left: finding players.

Three steps to put your game online

Go to the Hoora website and click the purple "Import" button in the left-hand menu.

The Import button in the Hoora website side menu

Drop in the .html file the AI gave you, and you get an instant playable preview, the best way to check whether your game holds up before showing it to anyone. Title, description, publish. It's free, and your players need nothing installed and no account.

What the statistics tell you

Your game then enters a feed people browse by swiping, with a creator profile, followers and public stats: views, likes, shares, comments, saves.

A game published on Hoora with its community statistics

That's also the best possible test of this article's question: does your AI-generated game entertain real players? The numbers will answer better than any opinion. The app also offers built-in AI creation, if you want to try all of this without touching a file at all.

Frequently Asked Questions About AI Making a Video Game

Can AI create a video game entirely on its own? Yes for a mini-game, as long as you accept two or three rounds of fixes. A fully successful game from a single request is still luck.
Which AI should I use to make a game? All three big ones hold up on mini-games. See our comparison of chatgpt vs claude vs gemini for the details.
Do I need a paid plan for good results? Free tiers are enough for a mini-game. Paid tiers mainly help hold longer files together across successive fixes.
Does the AI-made game belong to me? You remain the creator of what you publish. AI providers' terms around generated code keep evolving, so it's worth a look at their terms if you're aiming for commercial use.
Does this replace a developer? No. It replaces the barrier to entry. A developer stays essential as soon as the project grows.
Can players tell the difference from a "handmade" game? On a well-tuned mini-game, no. What they notice is bad difficulty balance or broken mobile controls, so work on those two.
Do I need to code to run this test? No, all of it is plain language. Our guide to making a video game without coding starts from zero.