Using ChatGPT to write SketchUp Ruby scripts

How to ask ChatGPT for a SketchUp script

ChatGPT has read a lot of SketchUp Ruby, but it can't see your model. Everything it knows about your file comes from your prompt. Put these in it:

• Your SketchUp version and platform. "SketchUp Pro 2025 on Windows." Some methods only exist in newer versions.
• What the script should act on. "The current selection", "every component named Chair" or "the whole model". Say whether it should reach inside groups and components.
• Units. SketchUp stores every length in inches internally, whatever your template shows. If you think in millimeters, say so and ask it to use .mm (for example 2400.mm).
• One undo step. Ask it to wrap the change in model.start_operation("Name", true) and model.commit_operation, and to call model.abort_operation if anything fails.
• A summary at the end. "Print how many objects you changed." A count tells you at once whether it did what you meant.

A prompt that works well:

"Write a SketchUp Ruby script for SketchUp 2025. For the current selection, including faces inside selected groups and components, set every face's material to a material named Walnut (create it if missing). Wrap it in one start_operation with abort on error. Print the number of faces changed. Only use methods from the official SketchUp Ruby API."

The script that comes back should look roughly like this:

model = Sketchup.active_model
model.start_operation("Paint selection", true)
begin
  mat = model.materials["Walnut"] || model.materials.add("Walnut")
  count = 0
  paint = lambda do |ents|
    ents.to_a.each do |e|
      if e.is_a?(Sketchup::Face)
        e.material = mat
        count += 1
      elsif e.is_a?(Sketchup::Group) || e.is_a?(Sketchup::ComponentInstance)
        paint.call(e.definition.entities)
      end
    end
  end
  paint.call(model.selection)
  model.commit_operation
  puts "Painted #{count} faces"
rescue => err
  model.abort_operation
  raise err
end

Note that painting inside a component changes its definition, so every copy of that component changes too. That is usually what you want, but ask ChatGPT to call make_unique first if it isn't.

How to run a ChatGPT script in SketchUp

The short version:

1. Save your model, or work on a copy.
2. Open Extensions › Developer › Ruby Console (SketchUp 2021.1 and newer; older versions have it under Window › Ruby Console). The path is the same on Mac and Windows.
3. Paste the script into the input box and press Enter.
4. Check the output line and look at the model. If it's wrong, use Edit › Undo (Cmd+Z on Mac, Ctrl+Z on Windows).

The Ruby Console only exists in desktop SketchUp. SketchUp Free (web), SketchUp Go and the iPad app can't run scripts. For loading .rb files, multi-line scripts and more on undo, see how to run a Ruby script in SketchUp.

Common mistakes ChatGPT makes with the SketchUp API

Most broken scripts fail in one of a few familiar ways. Knowing them makes the error messages much easier to read.

• Made-up methods. It invents names that sound right, like face.set_color or model.find_by_name. The console says undefined method. Ask it to check the method against ruby.sketchup.com or use a different approach.
• Ignoring groups and components. model.entities is only the top level. Geometry inside a group or component lives in its definition.entities, so a script that never goes inside finds nothing, or only half the model.
• Inches vs millimeters. A script meant to move something 100 mm that writes 100 moves it 100 inches. Lengths need .mm, .cm or .m, and output needs .to_mm if you want millimeters.
• No undo operation. Without start_operation, each change can become its own undo step. A script that touches 200 objects can need 200 undos, and one that fails halfway leaves half the change behind.
• Changing a collection while looping over it. Erasing or adding entities inside entities.each skips items or behaves oddly. The API docs say to iterate over a copy: entities.to_a.each.
• Editing every copy by accident. Changing a component's definition changes all its instances. Editing a group's entities makes that group unique, but editing through group.definition does not.
• Using the wrong "current" context. model.entities is the top level even when you're inside a group. For "where I'm working now", it should use model.active_entities.

How to fix a script that fails

When the console shows an error, copy the whole message, including the line number, and paste it back to ChatGPT with one sentence about what you saw. "It ran without errors but nothing changed" is useful too.

A few tips make each round count:

• Ask for a read-only version first: "Don't change anything. Print the names and counts of what you would change." That tells you whether it is finding the right objects.
• If it keeps guessing methods, paste the relevant page from ruby.sketchup.com into the chat.
• If the script ran but was wrong, undo it before you run the next version, or the second run starts from a half-changed model.
• Describe your model's structure when it matters: "The chairs are components nested inside a group called Dining."

When this approach stops working well

Pasting code works for short, clear jobs: rename these, recolor those, count something. It gets harder as the job grows, for one reason: ChatGPT never sees your model or the result of its code. It is writing for a model it imagines.

You'll notice it when:

• The fix for one error causes another and you're on round five.
• The job depends on what the geometry looks like, like "line the shelves up with the window", which code alone can't judge.
• The script works on a test object but not on your real model, because your nesting or names differ from what it assumed.

At that point either break the job into smaller scripts, or use a setup where the AI can run code in your model and look at the result itself. The comparison of pasting ChatGPT code against an agent that sees the model covers the trade-offs. If a script turns out to be something you'll reuse, you can turn it into an extension.

Questions