How to Write Effective CAD Prompts
The quality of your AI-generated CAD model is almost entirely determined by how well you describe it. This guide covers the techniques that consistently produce accurate, printable geometry.
The Golden Rule: Be Specific About Dimensions
Vague prompts produce vague models. The AI will guess at sizes, and those guesses are often wrong for your use case. Always specify dimensions in millimeters for every major feature.
- ❌ "A bracket with some holes"
- ✅ "A 60×40×5mm rectangular bracket with two 4mm holes, centered 10mm from each short edge"
Start With the Primary Shape
Name the dominant geometry first. The AI builds models incrementally: it needs to know what to start with before it can add features. Use terms like: box, cylinder, plate, tube, L-bracket, T-bracket, ring, cone, wedge, sphere.
Describe Features in Build Order
Think like a machinist: describe what you start with, then what you add or subtract:
"Start with a 80×60×10mm rectangular plate. Cut a 50×30mm rectangular pocket 5mm deep into the top face, centered. Add four 3.5mm holes in the corners, 8mm from each edge."
Use Negative Constraints
If you know what you don't want, say it explicitly. The AI responds well to constraints like "no fillets", "flat bottom only", "keep it simple", or "minimize boolean operations".
Pick the Engine That Matches Your Part
Models are built by a precision modeling engine that runs locally in your browser and handles everything from boxes and brackets to smooth curves, exact holes, and deep boolean stacks. If a first attempt looks wrong, tightening the prompt — adding explicit millimeter dimensions and naming the distinctive features — is the fastest fix.
Iterating With Feedback
After generation, the Give feedback button lets you type changes. Be specific about what's wrong:
- ❌ "Make it better"
- ✅ "The hole is 4mm but should be 6mm. Also extend the length from 60mm to 80mm."
Each revision is processed with context from your current model: the AI sees the existing code and makes targeted edits, rather than regenerating from scratch.
Common Prompt Patterns That Work Well
- Mounting brackets: "An L-shaped bracket, 80mm tall arm × 60mm base, 4mm thick, with two 4mm holes in each arm."
- Enclosures: "A 100×60×40mm box with 2mm walls, a flat removable lid, and a 20×12mm rectangular cutout on the front face for a USB port."
- Cable clips: "A 20mm cable clip with a 6mm channel, designed to mount with a single M3 screw through the flat back."
- Rings/gaskets: "A flat ring with 40mm outer diameter, 30mm inner diameter, and 3mm thickness."
What Doesn't Work Well
AI CAD has real limitations. Avoid prompts that ask for: organic/freeform shapes, complex multi-part assemblies, very small features under 1mm, or extremely high geometric detail. For these, a traditional CAD tool is more appropriate.
For more, see our guide on Text-to-CAD vs Traditional CAD.