Can you use ChatGPT or Claude to write a Revit tagging script?

How to get a first version from an AI assistant

People search for things like "Claude Revit", "ChatGPT Revit Dynamo", and "Revit MCP". The method is the same for all of them.

1. Decide where the code will run. A Dynamo Python node, a pyRevit script, a Revit macro, or a C# add-in each give the assistant a different starting point.
2. Describe the task in plain English. Say which categories to tag, which views, and which tag type to use.
3. Ask for code that opens a transaction, collects the elements in the active view, and creates one tag per element through the Revit API.
4. Paste the code into the tool, run it on a copy of a small test view, and give the assistant the error messages.
5. Repeat until the tags appear.

This works for a quick prototype. It is also where the trouble starts.

Where it breaks

It works on the view you tested. The assistant writes for the case you describe. A new project brings linked models, section views, rotated views, fabrication parts, and crowded sheets, and the script breaks or produces tags on top of each other. You cannot tell what it will do on the next project until you run it.

The Revit API is hard, and the training data is thin. A lot of the Revit API is undocumented, and the knowledge sits in third-party blogs. An assistant that has not seen these cases invents calls that do not exist or uses the wrong one for the Revit version, and Revit changes its API between versions.

You fix one bug and break another. Each fix changes code that nobody on your team designed or owns. A case that worked before stops working. The script grows into half-finished logic with poor structure and silent failures, where an error goes unnoticed and the run carries on as if nothing happened. This is the complexity wall that vibe-coded plugins hit.

It can damage your work. People run AI-written automations on live production drawings. The script deletes or edits something it should not have, and because nobody built an undo, you cannot get it back. Revit has no native autosave, so if the script crashes Revit you can lose two or three hours of work. Crashes from plugins are the number one reason Revit crashes.

Your data goes to the AI provider. On consumer subscriptions such as the $20 plans, the provider can use your chats for training unless you opt out. Anthropic's Free, Pro, and Max chats are used for training by default, and ChatGPT and Gemini have their own training and review settings. Business plans and the API are treated differently. If you paste project data or run this on projects without your firm's approval, you put your firm at risk and you may lose your job.

It feels fast, then it is not

You are a BIM worker with four hours of work from your manager. You spend two hours getting an AI to automate it, and it does not work. You scramble through the last two hours by hand. People end up spending more time on the automation than the work would have taken, and every new edge case means more prompting. The trade-off between fixing the bug and doing the work yourself eats the time you saved.

Small tasks will not need a script

For tiny click-saver tasks, Autodesk is building the answer into Revit. Autodesk Assistant takes natural-language commands and is in tech preview in Revit 2027. Writing and maintaining your own plugin for those is no longer worth it.

The harder tasks are the ones general chatbots cannot do reliably: tag placement, BIM modeling, and clash resolution. You can prompt an LLM in English and it will do things, but it is very hard to make them work reliably across projects.

What it costs

If you want a prototype for a one-off task, an AI assistant is a fine way to start. For production MEPF sheets, RayCAD is an affordable plugin that already handles placement, standards, undo, crash safety, and every Revit version from 2021 to 2027. See build vs buy for the full comparison, and how RayCAD protects your work.

What building an AI-written Revit tagging script yourself really costs

• Time. Whoever builds and maintains the script is doing software development instead of the billable engineering and drafting work they were hired for.
• The AI loop.With Claude or ChatGPT it becomes an endless loop of “fix this”, “now fix that”. Each fix can break a case that worked before, and nobody on the team owns the code.
• Cost. Hours spent building and maintaining scripts, times your hourly rate, against $200 per month for RayCAD for the whole company.

You can build an AI-written Revit tagging script, and this page shows how. For production MEPF sheets, RayCAD is the cheaper and safer choice. Try it free for 15 days.

Try RayCAD