Your First AI Prompts for Design and Engineering
A useful AI prompt for an architect or engineer looks nothing like the one-line requests people use for casual chat. The difference between "give me a beam size" and a full, context-rich prompt is the difference between a dangerous guess and a usable starting point. In this lesson you will learn the PROJECT prompting framework built specifically for AEC work, plus six prompts you can paste into ChatGPT or Claude today and get genuinely useful results.
What You'll Learn
- The PROJECT framework for writing AEC-specific prompts
- How to give AI the right context about codes, jurisdictions, and project type
- Six ready-to-use prompt templates for daily architecture and engineering tasks
- How to iterate on an AI response without losing accuracy
The PROJECT Framework
For AEC tasks, most weak prompts are weak because they skip context. The PROJECT framework forces you to include what the AI actually needs:
- Project type and scale (residential, K-12, Class A office, 12-story steel, 120-bed hospital)
- Role you want the AI to play (structural engineer, architect, code consultant, spec writer)
- Output format you expect (CSI-formatted spec, bullet list, table, code citation with reference)
- Jurisdiction and applicable codes (IBC 2021, NYC amendments, California Title 24, NEC 2023, ASCE 7-22)
- Examples or constraints (existing conditions, budget, schedule, client preferences)
- Confidence check (ask the AI to flag what it is unsure about)
- Tasks broken down (what specifically do you want it to produce?)
You will not use all seven on every prompt, but the more you include, the fewer hallucinations you get.
A Weak Prompt vs a PROJECT Prompt
Weak prompt:
Size a beam for a 25 foot span.
That gives you a useless answer. The AI has to guess the load, the material, the code, the deflection limit, and the connection type.
PROJECT prompt:
Act as a licensed structural engineer (Role). I am designing a small commercial mezzanine in a Type II-B building under IBC 2021 and ASCE 7-22 (Project, Jurisdiction). Assume a 25'-0" simply supported steel wide-flange beam carrying 50 psf live load and 25 psf superimposed dead load, tributary width 10'-0" (Examples/constraints). Provide a trial size using AISC 360-22 LRFD, checking flexure, shear, and live-load deflection limit of L/360 (Tasks). Present it as a concise hand-calc-style summary table (Output). Flag any assumption you are not confident about (Confidence check).
The second prompt produces a trial size, the governing check, and explicit callouts for you to verify. You will still re-run the calculation yourself — but the AI gave you a defensible starting point in 30 seconds.
Six Ready-to-Use Prompt Templates
1. Spec Section Drafting
You are a CSI-trained specification writer. Draft CSI MasterFormat Section 09 29 00 Gypsum Board for a Type II-B office fit-out in {city}. Use three-part format (General, Products, Execution). Reference ASTM standards where applicable. Flag any fire-rated assembly notes I should verify against the code drawings.
2. Code Compliance Check
You are a plan reviewer. I am designing a {occupancy group} building in {jurisdiction}, {number of stories} stories, {square footage}. The building will be Type {construction type} with {sprinkler status}. Walk me through the IBC 2021 requirements for: (1) allowable building height and area, (2) required fire-resistance ratings, (3) means of egress width and travel distance, (4) occupancy separation. For each, cite the specific IBC section number and flag any items I need to verify against local amendments.
3. RFI Response
I received this RFI from the general contractor: "{paste RFI text}". The applicable drawings are {A-101 and A-501} and the contract specification is Section {07 84 13 Penetration Firestopping}. Draft a professional response that either (a) directs them to the existing information, or (b) flags the item as needing coordination with {consultant}. Keep the tone firm, cite the drawing or spec section, and do not commit to extra work without an RFI-to-COR path.
4. Quick Engineering Calc
Act as a mechanical engineer. I need to size a ductwork main for {CFM} at a maximum velocity of {fpm} using the equal friction method at {in wc per 100 ft}. Provide the recommended round and rectangular duct sizes, show the key calculation steps, and note the ACCA or SMACNA reference. Then suggest two alternatives that reduce fan horsepower without changing throw distance.
5. Drawing Review
I am going to upload a floor plan PDF (page 1) and a reflected ceiling plan PDF (page 2). As a senior architect reviewing for coordination, check for: (1) door swings vs. accessibility clearances, (2) light fixtures that conflict with diffusers or sprinklers, (3) ceiling heights that conflict with tall equipment or doors. Return a markup-style list with sheet, room name, and the exact issue. Do not invent issues that are not visible in the drawings.
6. Design Narrative
Write a 250-word design narrative for a {project type} at {address}. Audience: the owner's internal board. Emphasize {sustainability / community impact / material choice / phasing}. Use active voice, avoid jargon, and end with one sentence on how the design responds to the owner's stated priority of {priority}.
Iterating Without Losing Accuracy
Your first AI response is rarely your final answer. To iterate well:
- Pin what is right. Before asking for changes, tell the AI what to keep. "The beam size and deflection check are correct; do not change those."
- Ask for a diff, not a rewrite. "Show me only the changes you would make if we switched to a W14x22."
- Challenge, don't accept. When the AI gives a confident answer, ask "What would make this wrong? Under what conditions would this beam fail the drift check?"
- Force citations. For any code reference, ask the AI to give you the exact section and edition, and then open the code book to verify.
What to Keep Private
Never paste client-confidential drawings, unstamped engineering work from a consultant, NDA-covered material, or RFP responses into a public AI without confirming your firm's policy. Most firms now have a paid enterprise seat (ChatGPT Team, Claude for Enterprise) where data is excluded from training — use that one for project content.
Key Takeaways
- Use the PROJECT framework (Project, Role, Output, Jurisdiction, Examples, Confidence, Tasks)
- Always include the code edition, jurisdiction, and project type in engineering prompts
- The six templates cover most daily AEC AI needs
- Iterate by pinning what is correct, asking for diffs, and forcing citations
- Use your firm's enterprise AI seat for project-confidential content