# ChatGPT for CAD: what a general chatbot can and cannot do > ChatGPT cannot open your part file or run an operation in SOLIDWORKS. It can write macros and scripts that might. What it is good at, where it breaks, and what CAD-native means. URL: https://cadabrai.com/chatgpt-for-cad/ Updated: 2026-08-14 --- # ChatGPT for CAD. What a general chatbot can do for CAD work, and what it cannot. ## The short answer. **ChatGPT cannot open your part file, cannot see your feature tree, and cannot execute an operation in SOLIDWORKS.** It can write code that might, and you then run and check that code yourself. A general assistant has a text window; a CAD copilot has the open document, the API, and the rebuild result. ## Where it earns its place. All of these are text tasks with a human checking the output. ### Drafting a macro A VBA or C# starting point for the SOLIDWORKS API. Expect to fix the API surface it invents. ### OpenSCAD and CadQuery Code-first CAD is text, so a language model can write it, and the output runs. ### Tolerance and stack-up maths Worked arithmetic with the reasoning shown. ### Explaining an error Turning an unfamiliar SOLIDWORKS message into an explanation, without knowing your specific file. ### Checklists and specs A DFM checklist, a drawing review list, a first draft of a spec section. ### Standards orientation Explaining what a GD&T symbol means. ## Where it breaks. **No live document context.** It cannot know which face you selected, what your feature is named, or that the boss was added two operations ago. Everything it writes is written blind. **Invented API surface.** Assistants confidently produce SOLIDWORKS API calls that do not exist or that changed release-to-release. It looks correct in the chat and fails when you run it. **No verification.** It cannot rebuild your model and tell you the rebuild succeeded. A macro that runs and quietly does the wrong thing looks identical to one that worked. **No standards.** Your title blocks, hole standards, tolerance defaults, and approved components are not in the conversation, so nothing it produces matches what your team ships. ## General assistant against a CAD-native copilot. | | A general chatbot | CADABRA | | --- | --- | --- | | Sees the open document | No | Yes | | Executes CAD operations | No, writes code you run | Yes, through the CAD API | | Knows your standards | No | Configured once, applied always | | Confirms the rebuild | No | Yes, and shows what changed | | Produces drawings and BOMs | No | Yes, on your title block | | Good for | Macros, scripts, maths, explanations | Changing and documenting real models | ## Use both, for different things. Keep using one for the text half of the job: the script, the checklist, the calculation, the explanation. For the half that involves your actual model, use something with access to it. That is what [an in-CAD copilot](https://cadabrai.com/ai-for-solidworks/) is, and it is the only difference that matters here. ## ChatGPT and CAD, answered. ### Can ChatGPT create a CAD model? Not directly. ChatGPT cannot open your part file, cannot see your feature tree, and cannot execute an operation in SOLIDWORKS. It can write code that might: a VBA macro, an OpenSCAD or CadQuery script, or a formula. You then run that code yourself and check what it did. A CAD-native copilot performs the operation in the open document instead, and can verify that the model rebuilt. ### What is ChatGPT useful for in mechanical design? Drafting a SOLIDWORKS macro to start from, writing OpenSCAD or CadQuery for a parametric shape, working through a tolerance stack, explaining an unfamiliar error message, and drafting a DFM or drawing-review checklist. All of these are text tasks with a human checking the output. None of them require access to your model. ### Why can’t ChatGPT read my CAD file? A SLDPRT or a STEP file describes boundary-representation geometry: surfaces, edges, and topology produced by a geometric kernel. It is not text, and reading it meaningfully requires a kernel to evaluate. Uploading one to a chat window does not give the model a kernel, so at best it reads metadata. ### Why not just write a macro or an API script? Macros work for a fixed, repeated sequence: they break when the feature tree changes, they need a developer to maintain, and each new variation is a new script. CADABRA resolves intent against the actual model at run time, so one prompt covers cases that would otherwise need a separate macro each. ### Can CADABRA fix SOLIDWORKS errors? CADABRA diagnoses rebuild errors, mate failures, dangling sketch relations, and missing references in the context of your open model, and explains what to change. Because it can see the feature tree and the failing feature, the answer is specific to your file. ### How accurate is it, and who checks the work? You do. CADABRA performs real operations on your model and shows exactly what it changed, so every edit is reviewable in the feature tree and revertible. ## See what document access changes. The same instruction, run against your real part. ## Keep reading - [Natural Language CADThe pipeline from a sentence to a rebuilt model, and how to write for it.](https://cadabrai.com/natural-language-cad/) - [Text to CADPrompts that resolve into native SOLIDWORKS features, not an imported mesh.](https://cadabrai.com/text-to-cad/) - [AI for SOLIDWORKSAn add-in that performs operations, not a chatbot beside the window.](https://cadabrai.com/ai-for-solidworks/) - [SOLIDWORKS AutomationIntent resolved at run time, instead of a macro per variation.](https://cadabrai.com/solidworks-automation/)