What Roblox Developers Actually Ask AI For — 2,945 Real Prompts Analysed (2026)

TL;DR

We counted topic mentions across 2,945 real user prompts sent to Picoo, a Roblox Studio AI plugin. The ranking is not what the tutorial ecosystem would predict. UI and menus lead at 11.2%, terrain and maps come second, and animation is third at 6.9% — despite being the one thing AI code generation fundamentally cannot do. Meanwhile tycoon (1.2%) and obby (1.0%), two of the most-covered genres on YouTube, sit near the bottom.

How this was measured

Every prompt sent to the plugin is a developer describing something they want built, in their own words. That makes the corpus a reasonable proxy for what people are actually trying to make — not what they say they want in a survey, and not what tutorial creators choose to cover.

Method, stated plainly so you can judge it:

• Corpus: 2,945 user-role messages sent to Picoo through July 2026.
• **Counting:** case-insensitive regex with word boundaries, one count per message per topic. \yanimations?\y matches "animation" and "animations" but not "animated" inside another word — an early version without boundaries counted "experience" as a hit for "xp" and inflated that topic by 6x.
• Overlap is expected. A prompt saying "combat system with a health bar and hit particles" counts toward three topics. These are not mutually exclusive buckets and the percentages do not sum to 100.
• What this is not: a measure of the whole Roblox platform. It is a measure of what people bring to an AI tool, which skews toward systems work and away from art and level design.

The ranking

Four things worth noticing

1. UI is the biggest category, by a wide margin

At 11.2%, menus and interfaces outrank every gameplay system. That is a little surprising until you think about what UI work actually is: repetitive, fiddly, and almost entirely uninteresting to write by hand. Nobody enjoys positioning a fourth panel or wiring the eleventh button. It is exactly the kind of work people hand off first.

It also means UI quality is where an AI tool is most often judged, and where the platform-specific traps bite hardest. The two that come up constantly: **ScreenGui.ResetOnSpawn defaults to true**, so a menu is destroyed and rebuilt on every respawn and its connections silently point at a dead instance; and **a custom loading screen only works from ReplicatedFirst**, because that container is delivered before anything else — the same code in StarterPlayerScripts runs after the default screen has already gone.

2. Animation is the third-biggest ask, and code cannot deliver it

This is the sharpest finding in the data. Animation sits at 6.9% — ahead of combat, ahead of inventory, ahead of saving.

And code cannot produce a keyframe animation. Roblox animations are assets, authored in the Animation Editor and published to the platform with an ID. No amount of Luau creates one. What Luau can do is **play an existing animation** (Animator:LoadAnimation with a real, owned asset ID) and **drive procedural motion** by manipulating Motor6D joints directly — which is genuinely capable, and how most weapon swings and idle sways are built.

The failure mode here is specific and worth naming: a general-purpose LLM asked for a Roblox animation will confidently emit an rbxassetid:// that does not exist, or belongs to somebody else. The script loads, the animation does not play, and there is no error explaining why. If you take one thing from this table, take this: when a tool offers to "generate an animation" for Roblox, ask what asset ID it is using.

3. The atmosphere trio is bigger than combat

Lighting (4.8%), sound (4.4%) and particles (3.7%) together account for 12.9% of prompts — more than combat, damage and NPCs combined. Developers spend a lot of their AI budget on how a game feels rather than what it does.

Each of these has one dominant gotcha, and they are all "the setting you are changing is being ignored":

• **Lighting:** if an Atmosphere object exists under Lighting, the old FogStart/FogEnd properties do nothing at all. Most modern templates ship with one, so people drag fog sliders and watch nothing happen.
• **Sound:** a Sound parented to a BasePart or Attachment is 3D and positional; the same Sound anywhere else is global and plays at full volume everywhere. Parenting is the entire behaviour.
• **Particles:** destroying an emitter deletes its live particles, so the effect visibly cuts off. Bursts want Emit(n) with cleanup delayed past the emitter's lifetime.

4. The famous genres are at the bottom

Tycoon at 1.2% and obby at 1.0% are the two most heavily tutorialised genres on YouTube, and they are the least-requested things in this corpus.

The honest reading is not "nobody makes tycoons". It is that those genres are so well covered by copy-paste tutorials and free models that people do not need an AI for them. What people bring to an AI is the work with no obvious tutorial: their own UI, their own map, their own systems glued together in a combination nobody has written up.

What this means if you are building

If you make tools or content for Roblox developers: the demand is concentrated in UI, world-building and presentation, not in the genre templates that dominate tutorial content. There is a visible gap between what is taught and what is asked for.

**If you are choosing an AI tool:** test it on UI and on animation, because those are the two highest-volume categories and they fail in opposite ways. UI failures are structural (the tool does not know ResetOnSpawn), animation failures are dishonest (the tool invents an asset ID). A tool that handles both well is handling the hard 18%.

If you are learning: the ranking is a decent curriculum. UI, terrain, shops and NPCs are what working developers spend their time on. Combat is further down the list than its share of tutorials suggests.

FAQ

Is this representative of all Roblox developers?

No, and it should not be treated that way. It represents people who chose to use an AI plugin — a group that skews toward systems and scripting work, and away from artists and level designers who never open a code panel. Read it as "what developers bring to AI", not "what Roblox development is".

Why does the table not add up to 100%?

Because a single prompt can mention several topics. "A shop UI where you buy swords" counts toward UI, shop, and combat. These are overlapping tags, not exclusive buckets.

Can AI actually make Roblox animations?

Not keyframe animations. Those are assets authored in the Animation Editor and published with an ID; code cannot create one. Code can play an existing animation by ID, and it can drive procedural motion through Motor6D manipulation — which covers a lot of what people actually want, including weapon swings and idle sway. Be suspicious of any tool that emits an rbxassetid:// for an animation it claims to have generated.

What is the most common thing AI gets wrong in Roblox?

In our experience it is the client/server boundary — code that changes something on the client and assumes everyone sees it. Appearance applied on the client, currency added on the client, and health written on the client all look correct to the person testing alone and do nothing for anyone else.

Where does this data come from?

Prompts sent to Picoo, a Roblox Studio AI plugin, through July 2026. Counted in aggregate by topic; no prompt content is reproduced here.

Build any of these

Every topic in the table above has a step-by-step guide with real Luau and the design decisions behind it — UI and loading screens, animation, lighting and atmosphere, particles, damage, and the full index.