Good Prompts vs Bad Prompts: Examples and Rewrites
Summary
- Bad prompts fail because they are vague, missing constraints, or ask for the wrong output format.
- Good prompts specify role, goal, audience, inputs, constraints, and a clear definition of “done.”
- This guide shows bad prompt examples, then rewrites each into a stronger prompt you can reuse.
- You will also learn quick “prompt triage” checks to fix a weak prompt in under a minute.
- If you reuse prompts frequently, saving proven rewrites can help reduce repeated prompt editing across tools.
Bad prompts waste time by producing generic, off-target answers; good prompts get you usable output by stating the goal, context, constraints, and the exact format you want.
Good prompts vs bad prompts: the practical difference
A bad prompt is unclear about what you want, who it is for, what inputs matter, and what the output should look like. The model fills in the blanks, and those guesses are where quality drops.
A good prompt reduces guessing. It gives the model the information it needs to choose the right approach and produce an answer you can actually use (not just read).
What “good” looks like (a quick checklist)
- Goal: What are you trying to accomplish?
- Audience: Who will read/use the output?
- Inputs: What facts, text, data, or constraints must be used?
- Output format: Bullets, table, JSON, email draft, outline, etc.
- Constraints: Length, tone, reading level, do/don’t include, compliance rules.
- Quality bar: What makes the answer “done” (examples, edge cases, steps, assumptions)?
Bad prompt examples (and better rewrites you can copy)
Below are common “bad prompt” patterns, why they fail, and rewritten prompts that are more likely to produce usable results. You can adapt these to ChatGPT, Claude, Gemini, or other AI tools.
Example 1: Vague request
Bad prompt: “Write a blog post about email marketing.”
Why it’s bad: No audience, angle, length, goal, or structure. The output tends to be generic.
Rewrite (good prompt):
“Write a 900–1,100 word blog post for US small business owners who sell services (not ecommerce) explaining how to start email marketing in 7 days. Include: (1) a short intro, (2) a step-by-step plan with daily tasks, (3) one example welcome email, (4) common mistakes to avoid, and (5) a simple checklist at the end. Tone: practical, not hype. Avoid mentioning tools by name unless necessary.”
Example 2: Missing context (the model can’t read your mind)
Bad prompt: “Respond to this customer complaint.”
Why it’s bad: The model needs the complaint text, your policy constraints, and what you can offer.
Rewrite (good prompt):
“Draft a customer support reply email. Use a calm, accountable tone. Constraints: do not offer refunds; we can offer a replacement or store credit. Ask one clarifying question if needed. Here is the complaint: [paste complaint]. Here is our policy excerpt: [paste policy]. Output: subject line + email body.”
Example 3: “Make it better” without a target
Bad prompt: “Make this copy better: [text]”
Why it’s bad: “Better” could mean shorter, clearer, more persuasive, more compliant, more formal, etc.
Rewrite (good prompt):
“Rewrite the following landing page paragraph for clarity and conversion. Audience: US parents shopping for an after-school tutoring program. Goal: increase demo sign-ups. Keep it under 90 words, 6th–8th grade reading level, and keep these claims unchanged: [list claims]. Provide 3 variations with different hooks. Text: [paste text].”
Example 4: Asking for “everything” (scope explosion)
Bad prompt: “Tell me everything about prompt engineering.”
Why it’s bad: Too broad; you get a shallow overview that doesn’t solve a specific need.
Rewrite (good prompt):
“I’m a marketer who uses AI to draft blog outlines. Teach me prompt engineering only for outlining. Give me: (1) a simple framework, (2) 5 reusable outline prompts for different intents (how-to, comparison, listicle, troubleshooting, opinion), (3) a checklist to diagnose weak outlines, and (4) one worked example using the topic ‘home office ergonomics.’”
Example 5: No output format (hard to use)
Bad prompt: “Analyze this survey feedback and summarize it.”
Why it’s bad: You might need themes, counts, quotes, and action items, but the model will guess.
Rewrite (good prompt):
“Analyze the survey feedback below and return a table with columns: Theme, What users are saying (1–2 sentences), Representative quote, Suggested action, Confidence (High/Medium/Low). Then list the top 5 actions in priority order with a one-line rationale each. Feedback: [paste responses].”
Example 6: Hidden constraints (compliance, brand, or policy)
Bad prompt: “Write ad copy for our supplement.”
Why it’s bad: Regulated categories require careful wording; without constraints, you may get unusable claims.
Rewrite (good prompt):
“Write 10 ad headlines and 5 primary text options for a supplement brand. Constraints: avoid medical claims, avoid ‘cure/treat/prevent,’ avoid before/after language, and do not mention specific diseases. Tone: confident, not sensational. Audience: US adults interested in daily wellness routines. Include one disclaimer line that can be appended to any ad.”
Example 7: Asking for citations you can’t verify
Bad prompt: “Give me statistics proving remote work increases productivity.”
Why it’s bad: You may get fabricated or unverifiable numbers if you don’t control sourcing.
Rewrite (good prompt):
“Help me write a balanced section on remote work and productivity without using specific statistics. Provide: (1) key arguments on both sides, (2) practical factors that influence outcomes (role type, management, tooling), and (3) suggested placeholders where I can insert verified stats later. Keep it neutral and avoid claiming ‘research shows.’”
Example 8: One-shot prompt for a multi-step task
Bad prompt: “Create a full marketing plan for my business.”
Why it’s bad: A plan depends on positioning, budget, channels, timeline, and constraints. One-shot prompts skip discovery.
Rewrite (good prompt):
“Act as a marketing strategist. First, ask me up to 10 questions to clarify my business, audience, offer, budget range, and timeline. Then propose a 90-day plan with: weekly milestones, channel mix, messaging pillars, and 5 experiments. Keep recommendations realistic for a small team. Output as a structured outline.”
Prompt triage: how to fix a weak prompt fast
When you get a bad answer, you do not always need a brand-new prompt. Run this quick triage and patch what’s missing.
Step 1: Add the missing “frame” in one line
- Role: “You are a [role]…”
- Task: “Your task is to…”
- Audience: “This is for…”
Step 2: Provide inputs (or say what to assume)
- Paste the text/data to use, or
- State assumptions explicitly: “If details are missing, assume X and list assumptions first.”
Step 3: Lock the output format
- “Return a table with columns…”
- “Give 3 options, each under 120 words…”
- “Output JSON with keys…”
Step 4: Add constraints that prevent unusable output
- Length limits
- Do/don’t include lists (claims, banned phrases, required points)
- Tone guidance (“direct,” “empathetic,” “technical,” “plain English”)
Step 5: Ask for a self-check
Add one line like: “Before finalizing, check that you followed every constraint; if something is missing, ask me one question.” This can reduce back-and-forth when your prompt has hard requirements.
Good prompts vs bad prompts: patterns you can reuse
Use these as “prompt skeletons” you can fill in quickly.
1) Rewrite prompt skeleton
Template:
“Rewrite the text below for [goal]. Audience: [audience]. Keep: [non-negotiables]. Constraints: [length/tone/claims]. Output: [format]. Text: [paste].”
2) Extraction prompt skeleton
Template:
“Extract [fields] from the text below. Output as a table/JSON. If a field is missing, return ‘Unknown’ and do not guess. Text: [paste].”
3) Planning prompt skeleton
Template:
“First ask clarifying questions (max [N]). Then produce a plan with [sections], tailored to [constraints]. Output format: [outline/table].”
4) Critique prompt skeleton
Template:
“Critique the draft below against these criteria: [criteria]. Provide: (1) top issues, (2) suggested edits, (3) a revised version. Draft: [paste].”
Where prompts break: limitations to expect
- Missing or wrong inputs: If you do not provide the key text/data, the model may fill gaps with assumptions.
- Ambiguous goals: “Make it better” can produce changes you did not want (tone shifts, removed details, added claims).
- Verification needs: If you need exact facts, quotes, or citations, you must verify them outside the model or provide the source text to work from.
- Policy and compliance constraints: Regulated or sensitive topics require tighter constraints and careful review.
- Long tasks: Complex work benefits from multi-step prompting (questions first, then output), rather than one giant request.
Choosing a way to save and reuse your best prompt rewrites
If you are collecting “good prompt” rewrites (like the ones above), you will likely want a reliable way to save them, find them later, and reuse them across different tools and documents. There are a few approaches, and the right choice depends on how you work.
Decision answer (by user type)
- If you work on Windows and want a local-first way to save copied text, favorite key snippets, and separately save reusable prompts: CopyCharm can fit this workflow.
- If you prefer a simple, manual system and do not need an app: a plain text file or notes document can be enough (but you will manage naming and searching yourself).
- If you want prompts available inside a specific AI platform: consider that platform’s native features (availability and behavior can vary by account and product), but those features may not help you reuse prompts in email, docs, or other apps.
A concrete workflow with CopyCharm (save, find, reuse)
Disclosure: CopyCharm is our product.
CopyCharm is a Windows desktop app and local-first context workbench for copied text. A practical way to use it for “good prompts vs bad prompts” is to treat each rewritten prompt as a reusable asset:
- Save: When you write a prompt rewrite you like (for example, the “Rewrite prompt skeleton” filled in for landing pages), copy it and save it as a Saved Prompt. Separately, you can favorite important copied text clips (like a brand voice paragraph or compliance constraints) so they are easier to pull back later.
- Find: Later, search your past clips or saved prompts when you need “bad prompt examples” to teach a teammate, or when you need your proven rewrite for “extract fields to JSON” or “ask clarifying questions first.”
- Reuse: Copy the saved prompt back into your destination tool (ChatGPT, Claude, Gemini, Cursor, email, documents) and fill in the bracketed parts. For those apps, the verified workflow is manual: search/retrieve in CopyCharm, then copy/paste into the other app.
When ChatGPT access matters: authenticated connector vs manual reuse
If you want ChatGPT to pull in your saved prompt text without you manually pasting it each time, CopyCharm offers an authenticated ChatGPT connector backed by optional AI Access sync. After you sign in with the account for an eligible active CopyCharm purchase, authorize the CopyCharm Desktop connection, enable and complete sync, and authorize the connector, ChatGPT can search and retrieve supported Synced Data. Supported categories depend on what you enable (Favorite Clips, Saved Prompts, and optional Other Clips within your selected time range). ChatGPT cannot access unsynced local CopyCharm data.
This is useful when you want to ask ChatGPT something like: “Find my saved prompt for extracting requirements into a table,” then retrieve the full text and run it with your new inputs.
One compact decision table (neutral)
| Need | What to look for | Why it matters for prompt rewrites |
|---|---|---|
| Fast reuse across many apps | Quick search + easy copy/paste | Lets you drop proven prompts into docs, email, and different AI tools without rewriting. |
| Separate “prompts” from “snippets” | A distinct place to save reusable prompts | Keeps your prompt library from getting buried in general clipboard clutter. |
| Optional ChatGPT retrieval | An authenticated connector with clear sync boundaries | Helps you retrieve saved prompts inside ChatGPT when you have authorized sync; avoids assuming local data is accessible. |
| Control over what is synced | Category and time-range selection | Helps you limit what leaves your device when you only need a subset available to ChatGPT. |
Why choose CopyCharm (and when to choose something else)
- Choose CopyCharm if: you are on Windows, you repeatedly refine prompts, you want to save reusable prompts separately from general copied text, and you want the option for ChatGPT to search/retrieve only the supported data you explicitly sync.
- Choose a simpler alternative if: you only need a handful of prompts and do not mind manual organization in a notes file, or you do not want any connector-based retrieval workflow and prefer to keep everything in one document you control.
Download CopyCharm to save prompt rewrites, search them later, and reuse them across your workflow: https://copycharm.ai/download
Frequently Asked Questions
FAQ 1: What is a “bad prompt” (in practical terms)?
Answer: A bad prompt is one that leaves key decisions to the model: unclear goal, missing context, no constraints, and no output format. The result may be readable but not usable for your specific task.
Takeaway: Bad prompts create guesswork; good prompts remove it.
FAQ 2: What is the fastest way to rewrite a bad prompt into a good one?
Answer: Add three lines: (1) who the output is for, (2) what “done” looks like (format + length), and (3) the constraints (must include/must avoid). If the task depends on source text or data, paste it in the prompt.
Takeaway: Audience + format + constraints fixes many weak prompts quickly.
FAQ 3: Can you give a bad prompt example for rewriting text, and a better version?
Answer: Bad: “Make this better: [text].” Better: “Rewrite the text below to increase clarity and sign-ups. Audience: US homeowners comparing quotes. Keep these claims unchanged: [list]. Limit: 80–100 words. Output 3 variations with different openings. Text: [paste].”
Takeaway: Define “better” with a goal, constraints, and output shape.
FAQ 4: Why do vague prompts produce generic answers?
Answer: When you do not specify audience, context, or constraints, the model defaults to broad, safe coverage. That can be fine for brainstorming, but it often misses your real intent (tone, compliance, depth, or format).
Takeaway: Specificity is how you trade generic output for targeted output.
FAQ 5: Should I ask for a role (“act as…”) in every prompt?
Answer: Not always. A role helps when you need a particular perspective (editor, analyst, tutor, support agent) or a specific style of reasoning. If your prompt already has clear constraints and format, adding a role may not change much.
Takeaway: Use roles when they meaningfully narrow the approach.
FAQ 6: How do I prompt for structured output like tables or JSON?
Answer: Name the structure explicitly and define the fields. For example: “Return JSON with keys: summary, risks, next_steps. If a value is missing, use ‘Unknown’ and do not guess.” For tables, specify column names and what each column should contain.
Takeaway: Structure is a prompt requirement, not a hope.
FAQ 7: How do I avoid hallucinated facts and fake citations in outputs?
Answer: Constrain the task to your provided inputs (“Use only the text below”), ask the model to label assumptions, and avoid requesting precise statistics unless you can verify them. If you need citations, provide the source passages and ask for quotes from those passages rather than invented references.
Takeaway: Verification comes from controlled inputs and careful review.
FAQ 8: How can CopyCharm help me reuse good prompt rewrites without retyping them?
Answer: You can save your proven rewrites as Saved Prompts and later search and copy them into any app. If you want ChatGPT to retrieve them inside ChatGPT, you can use CopyCharm’s authenticated connector after eligible account authorization and AI Access sync; ChatGPT can search/retrieve only supported synced data, not unsynced local CopyCharm content.
Takeaway: Save prompt rewrites once, then retrieve and reuse them when you need them.