A lot of AI-generated games understand damage. They can spawn a bullet, swing a sword, drop a spike trap, and remove half your health. What they often do not understand is the moment right before the hit.
That moment matters more than people think.
If the player cannot read danger before it lands, the game starts to feel cheap even when the numbers are technically balanced. You are not losing because you made a bad choice. You are losing because the game failed to speak clearly enough, fast enough.
Good Combat Lets You See Trouble Coming
I do not mean every attack should be obvious. I mean every important threat should have a shape the player can learn.
In Hades, enemies flash, turn, pause, and commit. In Dead Cells, attack windups are part of the language of the game. In Punch-Out!!, the whole drama comes from reading a tell and acting on it a fraction of a second later. These games are not easy because they telegraph well. They are readable.
A telegraph can be animation, sound, spacing, color, rhythm, or all of them at once. A sniper laser. A boss raising both arms. A floor tile that glows just before it detonates. A brief silence before a charge attack. The exact method matters less than the promise: if you pay attention, you had a chance.
Without that promise, difficulty turns sour very quickly.
Why AI Builders Keep Missing It
This blind spot is predictable. Generators are good at events. Spawn enemy. Move enemy. Fire projectile. Apply damage. Those are discrete actions with clean logic.
Telegraphing is harder because it lives in timing and presentation. You need to decide how early the warning appears, how obvious it is, whether it differs from other motions, and what the player is expected to do with that information. Too subtle and nobody reads it. Too long and the threat loses teeth.
Most AI tools would rather finish the attack than shape the anticipation. So the prototype gets harm before it gets readability.
The Fastest Smell Test
Pick one enemy or hazard in your game. Get hit by it three times on purpose. Then ask whether each hit taught you something visible.
| Question | Healthy sign | Warning sign |
|---|---|---|
| Can the player identify the attack before impact? | The move has a readable windup or cue | Damage arrives before the move is understood |
| Does the telegraph suggest a response? | You instinctively dodge, block, jump, or reposition | You only react after taking the hit |
| Do repeated hits feel informative? | Each failure makes the next read easier | Each failure feels random or blurry |
| Can spectators explain what killed you? | Someone watching can point to the warning you missed | Death looks abrupt even from the outside |
If you hit two warning signs, I would stop tuning health numbers. Readability is the actual problem.
Telegraphs Are Not Mercy, They Are Tension
Some creators hear this and worry that clearer attacks will make combat easier. Usually the opposite happens. Clear danger lets you make enemies more aggressive because players can understand what they are dealing with.
Dark Souls is full of punishing attacks, but the game earns a lot of that punishment through readable commitment. You see the knight draw back. You panic. You roll too early. That failure feels fair because the game offered information and you misused it.
Unreadable damage is not hardcore. It is just noisy.
This is why many AI prototypes feel harsher than their designers intended. The issue is not raw damage. The issue is that the player never gets the half-second of clarity that turns panic into skill.
The Best Telegraphs Fit the Fantasy
A telegraph should sound like it belongs in the world of the game.
In a horror game, the warning might be audio first, footsteps behind a door, a radio burst, a flicker in the corridor. In a colorful platformer, the cue might be exaggerated squash and stretch before an enemy lunges. In a tactics game, telegraphing can live in turn order, tile highlights, or facing direction. In a spaceship shooter, it might be charge glow, formation movement, or a distinct projectile lane.
The point is not to paste a red exclamation mark over every threat. The point is to make danger legible in the native language of the game.
How I Would Spec It
If you prompt an AI game builder with “make combat harder,” you will usually get more speed, more enemies, or more damage. I would rather prompt for readable aggression.
- Main threat: melee enemies should hit hard at close range
- Telegraph: every heavy attack needs a distinct 0.4 to 0.7 second windup with a unique pose or flash
- Player response: the cue should imply one clean counter, dodge, block, parry, or move away
- Variation rule: light attacks can be fast, but elite attacks must be easier to read than basic ones
- Failure mode to avoid: no attack should feel instantaneous unless the whole game is built around rhythm-speed reactions
That brief is much better than “make enemies smarter.” It gives the tool a fairness constraint instead of pure hostility.
I have seen this issue in quick builds from Chatforce, in browser-first tools like Rosebud, and in manual Godot prototypes where AI assistants handled part of the combat scripting. Different workflow, same missing layer.
What I Would Fix First This Week
If your combat feels cheap, do not start by lowering enemy damage by 20 percent. That can make the game softer without making it clearer.
Pick the most annoying enemy in your prototype and give it one unmistakable tell. A shoulder pullback. A charge sound. A shadow on the ground before impact. Then test whether players start saying “I saw it, but I messed up” instead of “what even hit me?”
That sentence tells you almost everything. The first one means your game has a chance. The second one means the combat language is still broken.
AI game builders are already decent at generating threats. The bigger challenge now is making those threats readable before they connect. If players cannot learn the warning, they cannot enjoy the danger. They can only endure it, and most of them will not endure it for long.