Search with intent.
Create with direction.

“Find hotel precedents” is a search topic. “How do small urban hotels create a calm arrival sequence when the lobby also contains a bar?” is a design question. It gives the assistant a spatial tension to investigate and gives the designer a basis for judging the answer.

For a first workflow, separate discovery from interpretation. Ask for candidate sources, the specific feature each illustrates and any uncertainty about the source. Open the original material before treating a claim as project evidence. The assistant’s summary is a reading aid; the source remains the evidence.

A useful research prompt identifies the project context, question, exclusions and output structure. Ask for a concise comparison rather than a confident recommendation with no supporting material. Save successful instructions as a starting pattern, then adjust the project variables.

For image generation, write a separate creative brief. Describe the spatial character, camera, lighting, material palette and elements that matter. State which aspects are exploratory. Generate a small set of directions, critique them, and change one meaningful variable at a time so the next iteration has a reason.

A generated image can help discuss atmosphere without being a reliable representation of a buildable space. Review openings, circulation, joinery and material transitions before using it in a presentation. Keep the designer’s annotations and selection rationale alongside the shortlist.

Research and image generation can produce an expanding collection with no progress in the design. Decide the review question first: does this precedent explain a useful arrival strategy? Does this image communicate the intended material contrast? What is being accepted, rejected or developed?

Do not let a visual imply product availability, dimensions or performance that nobody has verified. Likewise, do not promote a generated precedent description into a factual reference. Those are different types of information and need different checks.

We work on prompt structure, iterative critique, source checking and visual direction using approved studio examples. Your team leaves with instructions it can explain and adapt, rather than a collection of prompts it must trust.

The next step is a repeatable project workflow: save the relevant brief and decisions, agree a review method and make the process usable by a colleague. That is where context and memory become important.