Writing & Interrogating the Brief

The first, obvious use is drafting. After a client meeting you usually have a page of half-sentences: wishes, numbers, a budget hint, a few strong feelings. An LLM turns that into an organised brief quickly. Give it your raw notes and a structure to hit - project background, client and users, goals and success criteria, functional requirements and spaces, site and constraints, budget and programme, aesthetic and experiential intent, and open questions - and it returns a clean, readable document you can edit.

The point of this is not that the AI writes a better brief than you - it is that a blank page is slow and a good first draft is fast to react to. You are far more productive editing and correcting a structured draft than composing one from scratch. It also imposes a completeness scaffold: by asking for a full structure, you immediately see which sections are thin, which is itself diagnostic.

Two cautions. Keep the draft grounded in what the client actually said - instruct the model to work from your notes and to mark anything it is inferring or inventing, so you do not silently absorb its assumptions as the client's wishes. And treat the draft as yours to own: every line will be read as your professional understanding of the project, so you edit it into truth. The AI produces a scaffold and a first pass; you make it accurate. Used this way, drafting is a genuine time-saver that also sets up the far more valuable step - interrogation.

One practical tip: give the model your structure explicitly rather than letting it invent one. A house, a workplace fit-out and a clinic need different brief scaffolds, and supplying the right headings up front both improves the draft and makes the missing-section diagnostic sharper. Keep a couple of reusable brief templates for your common project types and paste the relevant one in with your notes - the AI fills and organises, and you immediately see which rooms are empty. That combination of your template and its speed is far more useful than a generic brief it structures on its own.

This is where AI earns its place. A brief that reads smoothly can still be full of holes, and you - close to the project, keen to start - are the worst-placed person to spot them. An LLM, prompted well, is an excellent devil's advocate, precisely because it is not invested and does not get bored asking the twentieth obvious question.

Three kinds of weakness are worth hunting deliberately. Gaps: what a complete brief for this building type would normally specify that this one omits - accessibility needs, maintenance, acoustic requirements, future flexibility, servicing, who exactly the users are. Assumptions: statements taken as given that ought to be confirmed - "open plan", "low maintenance", "family home" all hide decisions. Contradictions and competing priorities: the tensions every real brief contains - a generous programme against a tight budget, maximum daylight against privacy, a heritage character against energy performance. Naming these early, before they collide in the design, is enormously valuable.

A strong interrogation prompt:

You are a critical, experienced design lead reviewing this brief before we
start. Do NOT praise it. Produce three lists:
1. GAPS - important things a brief for this project type should cover but
   this one omits.
2. ASSUMPTIONS - statements treated as settled that we should confirm.
3. TENSIONS - competing priorities or contradictions that will force a
   trade-off later.
For each item, give a specific question I could put to the client.

The questions it generates are the real deliverable - you take the good ones back to the client. Not every one will be sharp; you curate. But a brief that has survived this interrogation is markedly more robust than one that has not.

It helps to run the interrogation in rounds. Take the first list of gaps and assumptions, answer the ones you can from your notes, then feed the remainder back and ask the model to go deeper on what is still open - and to prioritise, so you are not handing the client forty undifferentiated questions. A second useful move is to ask it to reconstruct the brief from the client's point of view: "if you were this client, what would you assume the designer understood that is not actually written down?" That flips the lens and surfaces the tacit expectations - the unspoken "obviously the kitchen opens to the garden" - that cause the most painful late surprises when they turn out not to be shared.

Beyond finding holes, AI helps you sharpen a brief - especially the two things clients state most vaguely: priorities and success. Most briefs list a dozen desirable qualities as if all were equally attainable. They are not, and pretending otherwise defers the hard conversation to a worse moment. Use AI to force the ranking honestly.

Ask it to take the stated goals and construct the trade-offs: "if budget is fixed, which of these goals compete, and what would giving priority to each cost the others?" Ask it to draft a forced-ranking exercise you can run with the client, or to turn woolly aspirations into testable success criteria - "a calm home" becomes "low ambient noise in bedrooms, visual separation of living and work zones, natural light in every main room." Criteria you can later check are worth far more than adjectives you cannot.

You can also role-play stakeholders. Ask the model to respond to the brief as a cost-conscious client, a sceptical planning officer, a future occupant, or a facilities manager who will run the building for twenty years. Each lens surfaces different concerns cheaply, before any are expensive. The output is not truth - it is a set of provocations - and you judge which land. But this kind of structured stress-testing, done in an afternoon, routinely catches priority conflicts that would otherwise emerge in design development, where they cost real time and goodwill to resolve.

A caution belongs here too: the AI's stakeholder role-plays are educated guesses, not real stakeholders. A simulated planning officer may raise plausible concerns and miss the one that actually matters in your jurisdiction; a role-played client cannot tell you what your client truly values. Use the technique to widen your thinking and rehearse conversations, never as a substitute for asking the real people. Its provocations are cheap and occasionally sharp - which is exactly the right expectation to hold them to.

The whole value of a brief is that it captures this client, this site, this set of real constraints. AI, left unchecked, pulls toward the generic - a plausible average brief for the building type, smoothing away the specifics that make a project itself. Guard against that in two ways.

First, keep the human evidence primary. The brief must be anchored in what the client actually said, the site actually is, and the constraints actually are - not in the model's idea of a typical project. Use AI to structure and stress-test that evidence, never to supply content the client has not. When the model infers or suggests, mark it as a question for the client, not as an established requirement. Second, close every loop with people. The gaps and tensions AI surfaces are hypotheses; they become brief content only after you have taken them to the client and gotten real answers. The interrogation's job is to generate better conversations, not to replace them.

Done with that discipline, AI transforms the briefing stage. You draft faster, you interrogate more thoroughly than you would alone, and you arrive at design with a brief that is complete, honestly prioritised, and free of the buried contradictions that ambush projects later. The brief stays unmistakably the client's and yours - AI simply made it sharper. And a sharp brief is the single best investment you can make in everything that follows, including the programming we turn to next.

It is worth saying plainly why this stage repays the effort. Every ambiguity left in the brief does not disappear - it is merely deferred to a later, more expensive moment, where it surfaces as a redesign, a difficult client conversation, or a compromise nobody is happy with. An hour spent interrogating the brief with AI, then confirming with the client, is among the cheapest insurance a project can buy. The tool does not make the decisions or hold the relationship; it simply ensures the right questions get asked while they are still cheap to answer.