The problem

“Warm, welcoming and not too modern” is useful—but it is not yet a brief.

People usually know more about what they want than they think. They may say, “I like the wood,” “that feels too modern,” “keep the roofline,” “we need this to be easy for older family,” or “I like this version, but not that second window.” Those are real decisions. The problem is that they arrive as scattered comments.

ChatGPT can help turn those comments into a working document. The goal is not to let AI decide the style. The goal is to make your own decisions easier to see, preserve and test.

Step 1

Start with evidence, not a fresh prompt.

Bring the current photos, previous concepts, notes and family feedback into one conversation or working file. If you start every design session from zero, ChatGPT has more room to invent a new direction.

If you already completed Part 1, your Current Renovation Brief is the ideal starting point.

Step 2

Separate FIXED, PREFERRED, FLEXIBLE and UNKNOWN.

Not every statement has the same weight. A roofline that cannot change is different from a window shape you are curious about. Treating them equally creates drift.

This one move makes later visual comparisons much cleaner.

Step 3

Turn feelings into things you can actually see.

Words like warm, calm, welcoming or traditional are useful only if they eventually influence colour, material, light, shape, texture, spacing or layout.

For example, “warm” might lead to warmer light temperature, natural wood, softer contrast or earthier wall colour. It does not automatically mean all of them.

Step 4

Create the negative brief: what must this not become?

People often recognize the wrong direction faster than they can describe the right one. That information matters.

A negative brief might say: not sterile, not overly glossy, not crowded, not cold, not ultra-modern, not visually heavy. Those are examples—not universal rules.

This gives ChatGPT a boundary against the kind of polished drift that can make an image look impressive while moving farther away from the family.

Step 5

Put decisions in the right order.

Paint colour should not outrank structural reality. A decorative window should not outrank accessibility. A useful decision hierarchy keeps the exciting details from overpowering the important ones.

Step 6

Write one sentence that describes the direction.

A useful brief needs a simple test. One sentence will not capture every detail, but it should make the overall direction recognizable.

A good sentence can later become the question you ask of every new concept: does this still feel like that?

Step 7

Choose a small set of design principles.

Do not try to preserve twenty aesthetic rules. Start with three to five principles that are strong enough to guide multiple choices.

Those are examples of useful principles. Your own brief should be based on your own evidence.

The real case

Why the porch window decision matters more than it seems.

In the family porch project, the useful progression was not simply “AI generated several pictures.” The useful progression was that the family became more specific.

We compared one-window and two-window directions. A family reaction then narrowed the next experiment: preserve the preferred direction and test one oval or diamond-shaped window on the far right of the door.

That detail does not have to become the final answer. Its value is that it reveals a preference: the family may want a simpler façade with one controlled distinctive feature rather than multiple competing elements.

That interpretation should still be tested with the family. ChatGPT can notice the pattern, but it should not quietly convert an interpretation into an approved design rule.

Step 8

Build the actual Current Design Brief.

Now turn the decisions into one document that can travel from ChatGPT to family discussion to a contractor conversation without changing meaning every time.

Step 9

Ask ChatGPT to challenge the brief before generating anything.

A good AI assistant should not only obey the document. It should help you notice contradictions.

This is especially useful when several people have contributed feedback at different times.

Step 10

Freeze the brief before the next visual round.

Once the family agrees that the brief accurately represents the current direction, give it a version number and stop casually rewriting it inside every prompt.

The goal is not to make ChatGPT more creative. It is to make the next decision easier to judge.

Before you move to Part 3

You should be able to hand someone one document.

If a family member, designer or contractor asks, “What are we actually trying to do?”, you should no longer need to search through fifty prompts and six images to answer.

You should be able to show the Current Design Brief and say: this is the direction; these are the fixed realities; these are the preferences; these are the open questions.

Part 3 uses that brief to solve the next problem: how to compare AI concepts without letting each new version quietly redesign the whole house.

Continue to Part 3 · Compare Without Drifting →