Architecture prompts for ChatGPT, Claude, and design work

01

Start with the architecture work

An architecture prompt should begin with a real deliverable: an estimate, drawing review, visual study, model operation, linework asset, proposal, presentation, or project record. It then tells the AI what evidence it may use, what must stay fixed, what it should produce, and how a person will review the result.

This makes the prompt useful to architects, designers, contractors, engineers, developers, marketers, and students without pretending that every role needs the same workflow. The task, project phase, required source files, and responsible reviewer remain explicit.

• — Named deliverable
• — Required source material
• — Software and project context
• — Review owner

02

Match the Prompt to the software and model

For Revit, AutoCAD, Rhino, Grasshopper, SketchUp, 3ds Max, and the Adobe suite, name the application, version, units, file types, and intended handoff. A model cannot infer an office standard, plug-in dependency, or drawing convention that was never supplied.

For image tools, including Gemini image-generation experiences such as Nano Banana, define the source geometry, camera, composition, materials, and elements that may or may not change. For text and code models such as ChatGPT, Claude, and Gemini, define the expected structure, fields, citations, scripts, or files.

• — Application and version
• — Accepted input files
• — Allowed changes
• — Output format and destination

03

Understand Prompts, skills, and MCPs

A Prompt is the reusable instruction and review method. A skill is a packaged capability that can include instructions, scripts, references, and assets so an agent can perform a repeatable kind of work. An MCP connection gives an AI application controlled access to external tools, data, or workflows.

Use a Prompt when the task can be completed from supplied context. Use a skill when the same method should be invoked consistently with bundled resources. Use an MCP when the AI must inspect or act inside another system. These layers can work together, but installing an integration does not remove the need for a clear task brief or qualified review.

• — Prompt: instructions and quality gates
• — Skill: reusable packaged method
• — MCP: connection to tools and data

04

Test the output instead of trusting the wording

Strong Prompts make the result testable. State what a passing output contains, how missing information is reported, which conditions cause the run to stop, and what a reviewer compares against the source material.

Test the package on a representative case, an incomplete case, and a failure case. Record the model, software, package version, corrections, and known limitations. Update the Prompt when repeated failures reveal an unclear instruction or missing gate.

• — Pass and stop conditions
• — Representative sample
• — Version record
• — Visible limitations

05

Search by the outcome you need

The most useful search path is person plus task plus software plus output: for example, a contractor estimating a patio canopy from photos, an architect reviewing a drawing set, a student learning a Grasshopper definition, or a developer connecting an AI assistant to Rhino.

archiPrompt organizes products and guides around those concrete combinations so a visitor can compare the required inputs, installation path, result, and review boundary before downloading anything.