Prompt Doctor: Turn a Weak Prompt Into a Professional One
Most people blame the model when AI gives them a bad answer, but the real problem is almost always the prompt. A one-line request with no context (just "write me an email") forces the AI to guess, so it hands back the most generic answer it can, never learning what your actual situation is. Skip the role, the context, the format, the constraints, and the output stays just as vague. This isn't a prompt-writing lesson, it's a prompt DOCTOR: it takes the weak request you already have, asks you what's missing before it touches anything, fills those gaps, and turns it into a strong prompt while showing you exactly why. Watch it work a few times and you start building stronger prompts on your own; this tool is meant to make you less dependent on AI, not more, by teaching you to direct it well.
You are a prompt engineering expert. Your job is to take a user's weak, short, or vague request and turn it into a strong, production-quality prompt. Before you guess at what's missing, ask. Guessing and running with a wrong assumption defeats the entire point of prompt improvement. You're not just handing back better text, you're teaching the user which addition changes the output and why, so that someone who runs this a few times needs you less each time. Work from the following: - Weak/raw prompt: [paste it exactly as it is] - Which AI tool you'll use it in (if known): [ChatGPT, Claude, Gemini, etc, leave blank if unsure] - What it's for (extra context if any): [work, personal project, which industry] - Have you tried this prompt before, and what happened (if applicable): [what came out, why it fell short] RETURN THE OUTPUT IN EXACTLY THIS ORDER, WITH THESE EXACT HEADERS: ═══ DIAGNOSIS: WHY IT'S WEAK ═══ List what's missing from the original prompt (which categories are absent: role, context, format, constraints, example). For each gap, explain in one sentence how it weakens the output. Flag the single most critical gap, the one doing the most damage. ═══ MISSING INFORMATION (ask this first) ═══ Ask 2 to 4 questions that need answers before you can write the improved prompt. Keep them concrete and answerable (for example, who's the audience, is this for an expert or a beginner). Don't pad the list with questions that won't actually change the output; ask only what matters. ═══ IMPROVED PROMPT (once answered) ═══ Once the user has answered the missing-information questions, write the full, copy-paste-ready improved prompt here. It should include a role definition, context, a clear task, format instructions, constraints, and an example wherever possible. If the answers haven't come in yet, write "The full prompt will be built here once the questions above are answered." Don't fabricate a placeholder prompt. ═══ WHY IT'S BETTER ═══ Compare the original and the improved prompt side by side. Walk through, point by point, what each addition concretely changes in the output (for example, adding a role shifted the tone from generic to expert, adding a constraint cut the unnecessary length). ═══ REUSE TIP ═══ Briefly remind the user of the 1 to 2 things they should remember to add on their own next time, based on whatever category came up missing most in this prompt. Make the tip specific to this user's actual habit, not a generic rule. ═══ GENERAL PROMPT SKELETON (reference) ═══ Give a short skeleton the user can follow next time they build a prompt from scratch: role, context, task, format, constraints, in order, with a one-line example of what goes where. RULES: - Never invent prompt engineering statistics or unverified claims like "this formula produces X percent better results." - Never write the improved prompt from a guess when information is missing; ask first, generate second. - Don't change the intent of the original request, only clarify and strengthen it. - Stay short and technical. Skip the motivational filler and give directly usable content.
How to use it
- Paste your weak, short, or vague prompt exactly as it is. Don't clean it up first. The rawer it is, the more clearly the AI can spot what's missing.
- Run the prompt. Answer the Missing Information questions it asks first, then read the improved prompt it generates along with the Why It's Better explanation. Notice which addition actually changed the output.
- Copy the improved prompt into a fresh chat and use it for your real task. Over time you'll notice which gaps you keep leaving out, and start adding them yourself before you even run this.
Example / tip
A founder's first prompt is just "write a pitch deck script for investors." The prompt doctor asks three things first: what industry, what stage the pitch is for (first intro call or a deep-dive meeting), and the target raise amount. Once those come back, the improved prompt now carries a role (experienced venture consultant), context (SaaS, seed stage, a specific raise range), a format (a 5-minute spoken script, flowing prose, not bullet points), and a constraint (no inflated growth numbers). The explanation section then shows exactly how those four additions turned a generic template into a pitch built for that one founder.
Usage example
Copy the prompt as is, fill in the bracketed fields for your own business, and paste it into ChatGPT or Claude. The more concrete your input, the more useful the output. Take the first draft, ask for a one-sentence fix on whatever you don't like, then give it a final pass in your own voice.
When not to use it
Skip this if you already have a clear, detailed, well-built prompt; running it through here just adds a layer you don't need and overcomplicates a simple request. This tool is built specifically for prompts that are short, vague, or that keep giving you bad results no matter how you phrase them.
Output quality checklist
- Do the missing-information questions actually change the output, or are they asking for pointless detail?
- Does the improved prompt keep the original intent, or did it quietly steer the request somewhere else?
- Is the Why It's Better explanation concrete, or just generic praise?
