Start by defining the type of research
Requests such as “research this market” or “study the site” are too broad. They often produce polished text that is difficult to use.
Divide early research into three practical categories:
- 01Context review: climate, regulatory framework and common architectural patterns.
- 02Needs analysis: user behavior, circulation conflicts and likely pain points.
- 03Option comparison: materials, structure, cost, schedule and operational trade-offs.
The important question is not only what you want AI to investigate. You must also define the output: a checklist, comparison table, list of assumptions or meeting brief.
A repeatable research workflow
1. Turn a broad topic into a verifiable question
Instead of asking for general ideas, define what can be checked.
- -“List five to eight design considerations related to X. Explain the reason for each and identify common exceptions.”
- -“Turn the relevant rules into a checklist and mark every item that requires confirmation.”
Asking AI to identify uncertainty makes the next verification step visible.
2. Ask for structure before detail
Work in two passes. First request the framework. Then expand only the sections that matter.
For climate-responsive design, the first pass might identify temperature, humidity, wind, solar exposure and rainfall. The second pass can ask for two or three strategies for each factor, together with suitable and unsuitable conditions.
This produces research notes rather than generic prose.
3. Specify a format your team can use
A useful comparison table may include:
- -strategy;
- -purpose;
- -cost implication;
- -risk;
- -required evidence.
A regulatory checklist may include the clause, affected area, available information and items still requiring confirmation.
The format matters because the output should become a working document, not remain inside a chat window.
4. Review every decision-critical claim
AI may present common knowledge as fact, overlook regional differences or generalize from incomplete conditions.
Use two firm rules:
- -Any number that can affect a design decision must be verified against its source.
- -Any regulatory issue should first become a checklist, then be confirmed by the responsible professional or consultant.
This is not distrust. It is a review workflow.
---
Example: early climate-response research
Consider a community housing project in a humid summer climate with cooler winters. The client wants a light visual character without sacrificing comfort.
The first task is to identify the climate factors that materially affect comfort and energy use. Ask AI to map each factor to possible architectural strategies and include the purpose, common approach, failure conditions and evidence required.
The resulting list may cover shading, combined ventilation strategies, envelope insulation and solar control.
Do not stop at the list. Convert it into a comparison table, then select two or three high-impact strategies for modeling, calculation or consultant review. AI has then produced decision candidates rather than a pile of information.
Example: circulation and operational screening
For an office renovation with a fixed structural grid, use AI to generate operational scenarios:
- -Where does peak-hour movement begin, pause and change direction?
- -Where can meeting and shared spaces create noise or visual conflicts?
- -Which layouts reduce future flexibility?
Use the output in two rounds. The first round identifies possible conflicts. The second checks each conflict against the floor plan and operational requirements.
AI is valuable here because it helps the team notice questions that may otherwise be missed. It still does not determine the final layout.
When the workflow is worth using
AI is most useful during the first part of research, when information is unstructured and the team needs a framework or a discussion draft.
It is less useful when verified research already exists or when the task requires exact legal wording, structural calculation or final compliance judgment.
AI can reduce the time spent organizing. It does not transfer professional responsibility.
Common mistakes
- 01Asking for an answer without asking for assumptions. Require AI to state the conditions behind its response.
- 02Accepting claims without traceable evidence. Decision-critical statements need a source.
- 03Treating AI as the designer. It can suggest strategies; the professional decides the trade-offs.
- 04Failing to bring the output back into project documents. Convert useful results into checklists, comparison tables or meeting records.
A practical principle
AI should provide candidates. Site conditions, professional knowledge and verified evidence should narrow those candidates into a decision.
Choose one current project this week. Ask AI for a structured strategy comparison, verify the critical points, and bring only the two most relevant options into the design discussion.
That is when AI becomes part of a working rhythm rather than another source of unreviewed information.
---
