Bible Network Crypto DeFi Onchain RWA AI Agent Stablecoin Chain SAFU CryptoTax DeFAI AGI Claude Me Claude Skill Claude Design Claude Cowork
Independent Media
Not affiliated with any project
Turn Ideas Into Interactive Visuals with Claude Design
claudedesign-me.com
LATEST
Where AI-Generated UI Breaks Accessibility: It's Not Just Color — Semantic Structure Is the Real Problem  ·  Case Study: Redesigning a Cluttered Dashboard — Where It Went Wrong and How AI Helped Untangle It  ·  Figma vs. Canva vs. Claude Design: Breaking Down Where Each One Actually Fits  ·  The Checklist to Run Before You Hit Generate: Five Questions, Good vs. Bad Examples  ·  Output Example: A One-Page Portfolio Site, From Prompt to Finished Result  ·  Five Prompt Templates for Common Design Tasks — Copy, Swap a Few Words, and Go
prompt-examples

The Checklist to Run Before You Hit Generate: Five Questions, Good vs. Bad Examples

30-Second Version · For the impatient
Anything you care about but didn't say explicitly is a decision handed over to the AI — these five questions are the checklist for saying what you care about.

Full Explanation +
01 · Why did this happen?

Of these five questions, if you could only build one habit first, which should be the priority?

If only one habit could be built first, question one (is the task itself clearly stated) is the priority — because it's the foundation everything else rests on. If what type of thing is even being made isn't clear, then style, constraints, and format requirements have nothing solid to attach to afterward; conversely, even if style and constraints are described with perfect precision, if the task itself is vague, the AI might still generate something with beautifully considered detail pointed in entirely the wrong direction.

A practical priority order: make sure question one is solid first, then fill in the other four as the situation calls for. Not every prompt needs all five questions fully addressed (a simple task might only need questions one and two), but question one is almost never optional — it's the baseline every prompt should confirm first.

02 · What is the mechanism?

Question three mentions ruling out approaches you already know cause problems — where should this "already known" list come from?

The most direct source is issues you've actually run into in your own past generation attempts — if your buttons last time all came back saying something vague like "Submit," rule that out explicitly this time; if the AI picked a color last time that you didn't like, rule out that specific color or color family directly this time. This list is inherently personal, and it grows more complete the more you generate with AI — there's no one-size-fits-all "must-exclude list" that applies universally.

A more efficient habit is jotting down what you changed each time you revise a generated output — over time, this builds into your own personal exclusion list that you can reference directly in future prompts, rather than re-deriving the same fix from scratch every time.

03 · How does it affect me?

Question four emphasizes format specificity, but some tasks genuinely need AI creativity — does every format need to be locked down?

No, and this is worth clarifying specifically: format specificity and limiting creativity are two different things. Format rules govern presentation (how many items, how long, what structure), not what the content itself should say — locking format down actually lets the AI concentrate its creativity on content, instead of having to juggle both "what should the content say" and "what should the format look like" at the same time.

If a task genuinely needs exploration and doesn't yet have a clear direction, the better move usually isn't giving no format at all, but giving a loose one (like "give me three different directions, two or three sentences explaining each"), letting the AI's content run free within a clear presentation frame — rather than leaving both format and content completely open, which usually produces a result that's hard to compare and hard to build a further discussion around.

04 · What should I do?

Does the order of these five questions carry particular meaning, or can the checking order be swapped around freely?

The current ordering reflects, to some extent, a "foundation before ornamentation" logic: the task (question one) determines what's being built, style (question two) determines visual tone, constraints (question three) rule out known failure modes, format (question four) determines presentation, and reference material (question five) gives everything before it something concrete to follow. If the earlier items aren't confirmed, the later ones tend to become detail built on shifting sand.

But this order isn't an absolute rule — what matters more in practice is that all five questions get thought through at some point, not rigidly following the checking sequence. If you already have reference material on hand (question five) and the task is simple, leading with that reference and then quickly confirming the other four is entirely reasonable too. The order is a memory aid, not a ritual that must be followed exactly.

Full Content +

Rather than discovering after generation what the prompt failed to specify, it's worth spending ten seconds running through a checklist before hitting send. This article collects five questions, each paired with a vague-versus-specific prompt example so you can feel the difference directly instead of memorizing rules.

Question One: Is the Task Itself Clearly Stated?

Vague: "Make me an interface." Specific: "Create a login screen for a mobile app, including an email input field, a password input field, a login button, and a forgot-password link." The vague version doesn't even specify what type of thing is being made, so the AI has to guess from the most common possibilities; the specific version lists both what's being built and which components it needs, leaving little room for the AI to freelance an unexpected result.

Question Two: Does Style Have Specific Adjectives, or Just Abstract Words?

Vague: "Make the style a bit modern." Specific: "Style: generous white space, sans-serif typeface, colors limited to black and white plus one accent color, avoid shadows and gradients." A word like "modern" has enormous room for interpretation — a hundred people might picture a hundred different things; the specific version breaks "modern" down into a handful of directly executable visual rules, giving the AI something concrete to follow instead of having to guess what "modern" looks like in your head.

Question Three: Have You Explicitly Ruled Out Anything You Already Know Goes Wrong?

Vague: (no exclusions at all) Specific: "Don't use vague terms like 'empower' or 'unlock your potential,' don't use gradient backgrounds, avoid the word 'Submit' on buttons — use a specific action description instead." If you've already run into an approach you disliked in previous output, naming it directly and excluding it is far more reliable than hoping the AI avoids it on its own — the AI has no way of reading "don't do that again this time" out of your head unless you say it.

Question Four: Is the Format Requirement Clear, or Left to the AI's Judgment?

Vague: "List a few key points about this feature." Specific: "List the key points about this feature as three bullet points, each under 15 words, no full sentences — use short phrases." Without a clear format spec, the AI might give you a full paragraph or ten bullet points, depending on what it judges appropriate in the moment; the specific version locks down the number of items, word limit, and presentation style, so what comes back doesn't need to be reformatted.

Question Five: If There's Existing Material to Reference, Is It Attached, or Only Described in Words?

Vague: "Colors should match our company style." Specific: (attach a recently published document or existing design system screenshot) "Read the colors and typography from this document, and keep everything that follows consistent with this style." Describing "match our company style" in words is abstract to the AI — it has no way to reverse-engineer what your company's style actually looks like from that sentence alone; attaching real reference material directly gives the AI something concrete to point to, far more reliable than a text-only description.

What This Means for Your Design Work

These five questions are actually the same thing underneath: any detail you care about but didn't state explicitly hands the decision over to the AI, which can only fill it in using the statistically most common approach — and that result isn't guaranteed to be what you wanted. Spending ten seconds checking a prompt against these five questions before generating is usually far faster than fixing it after the fact. This isn't so much a set of rules as a reminder: whatever's left vague will need fixing eventually — fixing it early is cheaper than fixing it late.

Diagram
Pre-Generate Checklist五問題檢查清單圖:任務、風格、限制條件、格式、參考素材,清單確保的是「有沒有講清楚」而非「內容對不對」Pre-Generate Checklist1. Task clear?what type of thing, what components2. Style: specific adjectives?not just "modern"3. Known pitfalls excluded?negative constraints4. Format spelled out?count, length, structure5. Reference material attached?not just described in wordsChecklist confirms claritynot whether the answer itself was rightthat still needs your own judgmentClaude Design Me · claudedesign-me.com
Feel free to share. Please credit the source.
Ask a Question
Please enter at least 10 characters
Related Articles
Five Prompt Templates for Common Design Tasks — Copy, Swap a Few Words, and Go
prompt-examples · Aug 14
Where AI-Generated UI Breaks Accessibility: It's Not Just Color — Semantic Structure Is the Real Problem
advanced · Aug 15
Case Study: Redesigning a Cluttered Dashboard — Where It Went Wrong and How AI Helped Untangle It
cases · Aug 15
Figma vs. Canva vs. Claude Design: Breaking Down Where Each One Actually Fits
comparisons · Aug 15