For Architectss and Built Environment Professionals
40 Questions Architectss Should Ask Before Trusting AI
When Midjourney generates five massing options in seconds, your judgment about which deserves development is more important than ever. The questions you ask of AI outputs determine whether the tool expands your thinking or narrows it.
These are suggestions. Use the ones that fit your situation.
1Does this Midjourney render show a solution I would have arrived at through my own process, or does it represent something genuinely different from my initial thinking?
2What design moves has the AI generated that I do not fully understand or could not explain to my client?
3Am I refining the prompt to get closer to my original idea, or am I actually exploring new territory the AI has suggested?
4If I removed this AI-generated option from consideration, what would my hand-drawn or modelled concepts look like?
5Does the AI output ignore site constraints because those constraints were not clear enough in my prompt, or because the model has not learned them?
6Which visual qualities in this render came from the prompt I wrote, and which came from statistical patterns in the training data?
7Could a client mistake the sophistication of the render for the sophistication of the design thinking?
8What design vocabulary have I stopped using because Midjourney or Dall-E does not render it convincingly?
9Has this tool suggested a material or formal language that belongs to this specific site and brief, or does it look like every other project it has generated?
10Am I showing the AI output to the client because it is the best design, or because it is the easiest to present?
Structural Reasoning and Compliance
11Can I redraw the load paths that Autodesk AI or Speckle AI has proposed without referring back to the software output?
12What assumptions about material behaviour or connection detail has the AI made that I have not independently verified?
13Does the structural solution satisfy the brief constraints, or has the AI optimised only for the parameters I fed into it?
14If the AI has recommended a reduction in material or section size, have I calculated the stress concentrations at discontinuities?
15What happens to this structure under the load cases the software was not asked to test?
16Have I checked the AI calculations against my own hand calculations for at least one critical member?
17Does the structural form the AI proposed fit the architectural concept, or am I now designing around a structural output I did not anticipate?
18What constructability issues might a contractor encounter that the AI model did not account for?
19If I need to explain this structural solution to a structural engineer who reviewed my work, what parts would I struggle to justify?
20Does the cost of the structure the AI proposed match the budget you set, and if not, do you understand why the optimisation moved that direction?
Planning Analysis and Site Strategy
21When ChatGPT summarised the planning policy, did it miss local precedent or recent decisions that would shape what is actually permissible?
22Has the AI analysis of site constraints been checked against the actual planning officer's previous comments on similar applications?
23If I ran a planning analysis manually using the local authority documents, would I reach the same conclusions as the AI summary?
24What site observations have I made that the AI model would have no way to know about—microclimates, pedestrian behaviour, views from unexpected angles?
25Does the AI assessment of neighbouring character assume visual qualities that are actually present, or has it inferred them from the address alone?
26Have I tested the AI's site strategy recommendations against the specific councillors who will vote on this application?
27If the planning authority asks why you chose this approach over alternatives, can you explain the reasoning without reference to the AI output?
28What details of the local area did the AI get obviously wrong, and how much should that affect your trust in the rest of its analysis?
29Has the AI identified a planning opportunity that I genuinely missed, or has it simply restated the planning code in a new way?
30When the AI suggested a design approach based on planning policy, did it account for flexibility you might negotiate, or does it read policy as absolute?
Client Communication and Professional Judgment
31If I present this AI-generated visualisation, will the client understand what design decisions I made, or will they think the computer made them?
32Can I explain to the client why I chose this direction over the four other AI options in a way that shows my professional reasoning?
33Does showing AI renders early in the process help the client understand the concept, or does it suggest the design is already finished?
34What part of my design fee is now justified by prompt writing rather than design judgment?
35If the client asks me to change something about the AI-generated image, can I deliver that change, or am I now dependent on the AI to make revisions?
36Have I shown the client the iterative thinking that led to this solution, or only the final AI render?
37Does the render quality of the AI image set expectations about construction quality or cost that may not be realistic?
38Could the client assume that because the design looks photorealistic, the buildability and cost implications are also certain?
39When I present structural or planning analysis that came partly from AI, do I clearly separate my independent judgement from the tool's output?
40If something goes wrong with the building, can I clearly show which decisions were mine and which were delegated to software?
How to use these questions
Before you accept an AI structural output, hand-calculate one critical member yourself. The act of calculating rebuilds the intuition you need to spot errors.
Keep a sketchbook separate from your digital tools. Spend 15 minutes on paper before opening Midjourney. This protects your design vocabulary from narrowing.
When Autodesk AI or Speckle AI proposes a solution, first write down what you think the solution will be. Then compare. The gap between your prediction and the output shows where your understanding needs work.
Ask your client which design direction they find most convincing, then ask why. If their reasoning differs from yours, you have discovered something the AI render did not show.
Before presenting planning analysis from ChatGPT, call the planning officer and mention your scheme. Let them correct the AI's assumptions while you are still in the design phase.