A lot of AI-generated games feel good for about thirty seconds. You move. You shoot. You merge. You dodge. You place a tile. The first action works, so you think the prototype has life. Then two minutes pass and the whole thing starts to feel oddly hollow.
Most of the time, the problem is simple. The game only has one real verb.
I do not mean one button. I mean one meaningful kind of decision. The player keeps doing the same thing, in the same mental mode, for the entire session. That is where a lot of AI game builders still give themselves away. They can generate an action. They struggle to generate interplay.
One Verb Can Carry a Toy. It Usually Cannot Carry a Game.
There is nothing wrong with a strong single action. Flappy Bird is mostly one verb. Canabalt is mostly one verb. Tiny mobile games can live or die on the feel of one repeated move.
But even those games are not as one-note as people remember. Flappy Bird is really about tapping and spacing. Vampire Survivors looks automatic, yet the real play is movement plus route planning plus upgrade timing. Into the Breach works because moving, attacking, blocking, and sacrificing position are constantly rubbing against each other.
The friction between verbs is where strategy starts. Without that friction, you do not have tension. You have repetition with particles.
Why AI Builders Keep Shipping One-Verb Prototypes
This is a very natural failure mode for generation systems. If you ask for a top-down shooter, the tool knows what the center of that fantasy looks like. It can give you movement, aiming, enemies, projectiles, health pickups, and a win state. That is enough to make something playable. It is not enough to make something interesting for long.
A second verb is harder because it has to interrupt the first one in a useful way. It has to create tradeoffs. Reloading can do that. Building cover can do that. Throwing a decoy can do that. Switching between stealth and combat can do that. Those are not just extra features. They change the shape of attention.
AI tools usually add content before they add contrast. So instead of a second verb, you get more enemies, more bullets, a bigger map, or a faster timer. Scale goes up. Depth does not.
The Fastest Way to Diagnose the Problem
Watch someone play your prototype for three minutes. Ignore whether they win. Ignore whether they smile. Just write down the verbs they actually use to solve problems.
If your notes read like this, you have a one-verb game:
- Move toward enemy
- Shoot enemy
- Move toward next enemy
- Shoot enemy
- Collect drop because it is nearby
If every decision collapses into the same action, your game is asking for execution only. That can work for an arcade score chase. It falls apart fast in most other formats.
| Question | Healthy sign | Warning sign |
|---|---|---|
| Do players switch mental modes? | They alternate between aggression, positioning, and setup | They stay in one rhythm the whole time |
| Does the second verb create a cost? | Using it means giving something else up | It is just a bonus button |
| Would a harder level change behavior? | Players use different tools under pressure | They do the same thing, just faster |
| Can you describe the loop without repeating one word? | The sentence has at least two distinct actions | It is basically one action over and over |
Two warning signs is usually enough. I stop tuning numbers at that point. The structure is the problem.
The Second Verb Should Complicate the First One
This is the part many creators miss. The answer is not to bolt on crafting, dialogue, or a skill tree because the game feels thin. The answer is to add one more verb that makes the first verb less automatic.
If your main verb is shooting, the second verb might be scanning. You pause to expose weak points. Now accuracy is not enough. If your main verb is placing towers, the second verb might be rerouting enemies with temporary barriers. Now location choices stay alive after placement. If your main verb is platforming, the second verb might be rewinding one object in the room. Now timing and planning finally have something to talk to.
The best pairings create a small argument in the player's head. Attack now or prepare first? Spend the item or save it? Move forward or hold ground? That argument is game design. AI builders rarely invent it on their own.
Examples That Actually Work
Portal is a clean example. Entering a portal is not interesting by itself. Placing portals in relation to momentum, angle, and room layout is what makes the game sing. Two verbs, one system, endless consequences.
Slay the Spire lives on attack versus defend, but really it is attack, defend, scale, stall, and manipulate draw order. Every turn asks what kind of turn this should be. That is why a handful of cards can produce so much thought.
Even Minecraft, which many people reduce to mining and building, gets its texture from the tension between gathering, shaping, and surviving. You are never just doing one thing. The verbs keep colliding.
That is the standard I use when I look at AI game tools. Not whether they can generate mechanics, but whether those mechanics force each other to matter.
How to Prompt for Better Verb Design
If you ask an AI builder to make a survival game, it will probably give you gathering, hunger, and enemies because those are common genre markers. If you want depth, prompt for a conflict between verbs instead of a pile of features.
I would write something closer to this:
- Main verb: ranged attack against fast enemies
- Second verb: deploy temporary light sources that slow enemies but reveal your position
- Tradeoff: light makes aiming easier, but increases spawn pressure
- Failure mode to avoid: do not solve every problem by shooting more
That is much more useful than saying, "make a dark survival shooter with roguelike elements." The generator needs tension in the prompt, not just theme.
What I Would Fix First in Most AI Prototypes
If I had an hour with a bland AI-generated game, I would not start with shaders, UI polish, or more content. I would ask what the player can do besides the obvious thing, and whether that second action changes the obvious thing in any meaningful way.
If the answer is no, that is the whole week.
Add one verb. Make it costly. Make it powerful enough to matter. Make the player choose between using it now or later. Then test whether the game finally produces stories instead of just inputs.
That is the shift I am waiting to see from AI game builders. We already have tools that can generate a toy. The next leap is tools that understand interplay. Until then, if your prototype feels flat, do not ask for more stuff. Ask for a second verb.