You type a prompt. The AI responds. That is how most people think about working with AI.

But there is a second mode — one that most people never discover — where the flow reverses. Claude asks you the questions. Not vague, open-ended questions like “what would you like?” but structured ones with clear options, visual previews, and tradeoffs explained. You pick from a menu. Claude builds from your answers.

This is reverse prompting. And once you start using it, you will wonder why you ever tried to describe everything upfront.

The problem with prompting

When you write a prompt, you are making dozens of decisions at once. You are choosing colours, layout, tone, structure, features, and behaviour — all in one breath. Some of those decisions you care about. Most of them you do not even realise you are making.

The result is predictable. Claude fills in the gaps with safe defaults. You get something that works but feels generic. Then you spend the next twenty minutes saying “no, not like that” and “can you change the…” — slowly uncovering the decisions you did not know you needed to make.

Reverse prompting solves this by surfacing those decisions before Claude builds anything.

What it looks like

Instead of you describing everything, Claude presents you with structured questions. Each question has a few clear options with brief explanations of what each one means.

An example. You say:

Build me a portfolio site.

Instead of immediately building something with its default assumptions, Claude asks:

Layout: “How should your work be presented?”

  • Grid — equal-sized thumbnails, clean and organised
  • Masonry — varied sizes, more dynamic and editorial
  • Single column — one piece at a time, maximum impact

Tone: “What feeling should the site have?”

  • Minimal — lots of white space, the work speaks for itself
  • Editorial — magazine-like, with type and layout as design elements
  • Warm — inviting, personal, like visiting a studio

Navigation: “How should visitors move through the site?”

  • Single page — everything scrolls, no clicking needed
  • Multi-page — separate sections for work, about, contact

You pick your answers. Claude builds exactly what you chose. No guessing, no defaults, no twenty minutes of corrections.

This is not a trick or a prompt hack. Claude Code has a built-in mechanism for this — it can pause what it is doing, present you with a structured question and a set of options, wait for your answer, and then continue. The questions are real interface elements, not just text in a chat. Claude uses this tool whenever a decision needs your input.

Why this is better

Three reasons.

1. You make decisions you did not know existed.

Most people do not think about whether their portfolio should use a masonry layout or a grid. They have never considered whether the tone should be “editorial” or “warm.” These are real decisions that affect the result — but they are invisible until someone asks.

Reverse prompting surfaces these decisions. It turns “build me a website” into a brief creative direction session. The result is better because the inputs were better.

2. Claude builds with confidence instead of guessing.

When you say “make it look nice,” Claude has to guess what “nice” means to you. Sometimes it guesses right. Often it does not. But when you say “minimal tone, grid layout, single page” — there is nothing to guess. Claude builds with full context.

The difference is the difference between asking a contractor to “make the kitchen nice” and handing them a mood board with specific choices. Same contractor, wildly different outcomes.

3. You discover your own taste.

This is the surprising one. When Claude presents options you have never considered, you discover preferences you did not know you had. “Actually, I do want masonry layout. Actually, I hate multi-page navigation. Actually, editorial tone is exactly what I have been trying to describe for months.”

The questions teach you about yourself. That is not something a one-directional prompt can do.

How to trigger it

There are several ways to get Claude to ask you questions instead of guessing.

In conversation

The simplest way. Just tell Claude to ask before building:

Build me a portfolio site. Before you start, ask me about my preferences — layout, colours, tone, navigation. Give me options to choose from.

Or even simpler:

Build me a portfolio site. Interview me first.

Claude will ask structured questions with clear options before writing a single line of code.

In your project memory

If you want Claude to always ask before building, add this to your CLAUDE.md:

## How we work

Before starting any new page, component, or significant visual change,
ask me structured questions about my preferences. Present 2-4 options
for each decision with brief descriptions. Do not assume — ask.

Now Claude will ask every time, without you needing to remember to request it.

In a skill

Skills can instruct Claude to ask structured questions as part of their workflow. For example, a design review skill might always begin by asking “what matters most to you right now — visual polish, accessibility, or performance?” Your answer determines what the skill focuses on. The interview becomes a repeatable process — every time the skill runs, it asks the right questions in the right order.

Sketching the options

This is the part most people never see.

When Claude asks you a question about layout or visual structure, it can sketch each option — a quick ASCII mockup that shows you what the choice actually looks like. Not a description. The thing itself, drawn in the question.

An example. Claude asks: “Which header layout do you prefer?”

Option A shows:

┌─────────────────────────────────────┐
│  LOGO              Home  Work  About│
│─────────────────────────────────────│
│                                     │
│         Your Name                   │
│         Subtitle here               │
│                                     │
└─────────────────────────────────────┘

Option B shows:

┌─────────────────────────────────────┐
│                                     │
│            YOUR NAME                │
│                                     │
│       Home    Work    About         │
│                                     │
└─────────────────────────────────────┘

You see the difference immediately. No description needed. You pick the one you prefer and Claude builds it.

This works for layouts, component structures, colour schemes, navigation patterns — anything where seeing is faster than reading.

Use cases

Reverse prompting is not just for building websites. It works anywhere that Claude would otherwise be guessing.

Creative direction

Create a logo for my photography business.

Before designing, Claude asks about style (typographic, iconic, abstract), colour palette (monochrome, warm earth tones, high contrast), and mood (serious and refined, playful and approachable, minimal and modern). Three questions, nine options, and Claude has a complete brief.

Project setup

Start a new project for my client’s restaurant.

Claude asks: what type of restaurant? What is the target audience? Is this a single-page site or does it need reservations, menu, and events? Does the client have brand colours, or should we choose? Four questions and the project starts with real context instead of assumptions.

Debugging and decisions

The page loads slowly on mobile.

Instead of immediately optimising everything, Claude asks: “Which matters more — keeping the current visual quality, or maximum speed? Do you want me to investigate first and show you what is slow, or just fix the most common causes?” Your answer determines whether Claude spends ten seconds or ten minutes, and on what.

Choosing between approaches

Add a dark mode toggle.

Claude asks: “Should the toggle respect the system setting by default, or start in light mode always? Where should the toggle live — top-right corner, in a settings menu, or floating at the bottom? Should the transition be instant or have a smooth crossfade?” Each choice affects the implementation. Getting them right upfront saves three rounds of “actually, can you move that.”

The interviewer pattern

The most powerful version of reverse prompting is the interviewer — a skill or instruction that conducts a structured conversation before doing anything.

Add this to any skill or project memory:

## Before building

Conduct a brief interview. Ask 3-5 questions, one at a time.
Each question should have 2-4 clear options with descriptions.
Include a visual preview when the question is about layout or visual structure.
After all questions are answered, summarise the choices and confirm before building.

The interview pattern turns Claude from an assistant that takes orders into a collaborator that helps you think. It is the difference between “here is what I told you to build” and “here is what we decided to build.”

When not to reverse prompt

Not everything needs an interview. If you know exactly what you want, say it directly. “Make the heading larger” does not need a question about how much larger. “Fix this bug” does not need a preference survey.

Reverse prompting is most valuable when:

  • You are starting something new and the decisions are not yet made
  • The outcome depends on taste, not just correctness
  • There are real tradeoffs between approaches
  • You want to explore options you have not considered

For everything else, just tell Claude what to do. The directness of a clear prompt is its own kind of efficiency.

Getting started

Try it in your next session. Instead of describing everything upfront, say:

I want to build [thing]. Interview me about my preferences before you start. Give me options to choose from.

Or just click the microphone and say it out loud. Voice input is one of the most natural ways to work with Claude — and answering structured questions by speaking feels like a real conversation.

Watch what happens. Claude will ask questions you would not have thought to answer. Your answers will be more specific than your prompts ever were. And the result will be closer to what you actually wanted — because you told Claude what you wanted, one decision at a time.

The best conversations with AI are not monologues. They are dialogues.

Art & Algorithms publishes guides, tutorials, and prompt packs at the intersection of art and code. Subscribe for the full archive.