AI for Architects: Where It Helps, Which Tools, and Where It Falls Short
A year ago, most architects were curious about AI. Now most practices are using it. The question has moved from "should we?" to "where does it actually help, and where does it quietly cause problems?"
This guide goes through a project from brief to handover and shows where AI earns its place for architects today, which tools do what, and the limits that still matter when your name is on the drawings.
The short answer: AI helps architects most in early design, for concept images, sketch-to-render visualisation, massing and site studies and fast layout options, and in the admin around a project, such as specifications, documentation and repetitive office tasks. It's weakest at dimensional accuracy, consistency across a set of views, codes and anything you're professionally liable for. In the RIBA's 2025 survey, 59% of UK practices were already using it.
Sketch
AI render
How many architects are using AI?
More every year, and fast. The Royal Institute of British Architects' 2025 AI report found that 59% of UK architecture practices now use AI, up from 41% in 2024. Adoption in large practices is above 80%, and nearly half of small studios (48%) have started too. Early design visualisation and specification writing were among the most common uses.
The same survey showed the worries: 67% were concerned that AI will increase the risk of their work being imitated. Both sides of that are worth keeping in mind as you read on.
Where does AI help across an architecture project?
| Stage | What AI can do | Examples of tools |
|---|---|---|
| Brief and research | Summarise precedent, organise client requirements, draft questionnaires | ChatGPT, Claude, Gemini |
| Site and massing | Sun, wind, daylight and noise studies; generate and compare site options | Autodesk Forma |
| Concept | Mood images, sketch-to-render, facade and material options | Nano Banana, ChatGPT, Midjourney, Veras |
| Layout | Generate residential layout options from a room brief | Maket and other plan generators |
| Visualisation | Restyle or enhance views from Revit, SketchUp or Rhino | Veras, image models |
| Documentation and admin | Specification drafts, sheet and file naming, registers, small scripts | Revit add-ins, Claude Code, agent workflows |
Site and massing
Autodesk's Forma is aimed squarely at the pre-design and schematic phases: it runs environmental analysis such as sun, wind and noise on early massing, and can generate and compare site options against your parameters. It's one of the clearest cases of AI answering a real design question rather than making a picture.
Concept and visualisation
This is where most practices start. Image models can turn a hand sketch, a massing model or a screenshot from your 3D software into a photoreal concept image in minutes. Veras, now owned by Chaos after its acquisition of EvolveLAB in 2025, does this inside Revit, SketchUp, Rhino, Vectorworks and Forma. Our guide to AI rendering, from sketch to render, walks through the method step by step.
Layouts and plans
AI floor plan generators produce layout options from a list of rooms, sizes and adjacencies, which is useful for residential feasibility and early client conversations. Image models can also turn an existing 2D plan into a furnished 3D view. See our guide to AI floor plan generators and 2D-to-3D plans.
The admin nobody went to architecture school for
Sheet naming, drawing registers, specification tables, renaming three hundred files the way your office names them: none of it is design, and all of it eats the week. This is where agentic AI and simple scripts written with tools like Claude Code often pay for themselves first, precisely because the task is repetitive and the result is easy to check.
Where does AI still fall short for architects?
- Consistency across a set. One great image is easy. Six views where the ceiling height, window mullions and cladding stay identical is hard, and clients notice when they don't.
- Dimensions. Image models don't measure. A render can't be trusted for a size, a clearance or a level.
- Codes and regulations. AI doesn't know your local building code, planning policy or fire strategy, and it will cheerfully draw something that breaks all three.
- Liability. Your professional responsibility doesn't transfer to a model. Anything that goes into a planning application or a construction set needs the same checking as before.
- Confidentiality. Check the terms before uploading client drawings, and use business accounts with training turned off for anything under an NDA.
None of this is a reason to avoid AI. It's the checklist for using it professionally. For a longer look at what this means for the profession, see will AI replace architects?, and for the newest tools, our test of GPT-6 Astra and AI 3D modelling, including what a photo-to-3D model actually contains.
How should an architecture practice start with AI?
- Pick one repeatable task. Early concept images or specification drafting are good first candidates, because the output is easy to judge.
- Start from what you already use. Screenshots from your Revit or SketchUp model make better AI inputs than blank prompts, and nothing in your current pipeline has to change.
- Write down the prompts that work. A shared prompt library turns lucky results into a studio method.
- Set rules for client work. Label AI images as concept visualisations, keep originals, and agree what can be uploaded where.
- Measure the time saved. If a task isn't faster after a month, drop it and try the next one.
Interior work follows the same pattern. Our guide to AI interior design covers rooms, styles and mood boards.
Want the workflow, not the tour? Our AI for Architects and Interior Designers course covers renders from your own drawings, BIM and Revit into AI, plans and elevations, client proposals, agentic AI for the admin and a portfolio site you build yourself. Ten modules, each worked on a real project, and a private 1:1 session at the end.
Frequently asked questions
How are architects using AI?
Mostly in early design and admin: concept images, sketch-to-render visualisation, site and massing studies, layout options, specification drafting and documentation tasks. The RIBA's 2025 survey found early design visualisation and specification writing among the most common uses.
What are the best AI tools for architects?
It depends on the task. Autodesk Forma for site and massing studies, Veras for AI visualisation inside Revit, SketchUp and Rhino, image models such as Nano Banana and ChatGPT for concept images, and residential plan generators such as Maket for layout options.
Can AI design a building?
It can generate images, massing options and layout ideas, but it can't take responsibility for a building. Structure, codes, services, coordination and sign-off still need an architect. AI speeds up the exploration; the design decisions and liability stay human.
Is it safe to upload client drawings to AI tools?
Only if the tool's terms allow it and the client would be comfortable. Use business accounts where your data isn't used for training, avoid uploading anything under an NDA to consumer tools, and agree an AI policy with clients for sensitive projects.
Start small, keep what works, and keep checking the output like you'd check a junior's first drawing. That's how AI becomes a studio tool rather than a novelty.