Back to home
3 real scenarios · before vs after

See how PromptOpt turns “close enough” into “ready to ship”

Each case shows: user's raw input → PromptOpt-style optimized prompt → key changes. Copy into ChatGPT / Claude and feel the difference in 5 minutes.

Case 1 · Content

Xiaohongshu post · Sunscreen review

Beauty blogger · ship a 300-word post in 10 minutes

😣 The pain

Asking ChatGPT "write me a Xiaohongshu post" returns something that's either too corporate, the wrong word count, or stuffed with emojis but no hook. You spend 3-5 rounds editing before it's publishable.

Before · user's rough input
write a Xiaohongshu post about summer commute sunscreen
After · PromptOpt style
You are a senior Xiaohongshu beauty blogger. Over the past 3 years you have written 200+ viral posts, with a single post reaching 500K views.

# Task
Write a Xiaohongshu post reviewing "summer commute sunscreen".

# Audience
25-32 year-old urban women, full-time office workers. Pain points: morning commute sun exposure, lunchtime reapplication, and humid subway rides after work.

# Must include
1. Opening hook: "Girlies!"-style strong emotion (avoid "Fam!")
2. 3 real usage scenarios: morning commute / lunch reapplication / subway ride home
3. At least 1 concrete usage detail (e.g. "no white cast, 5-second dry finish")
4. 1 contrarian insight (e.g. "Higher SPF isn't always better")
5. Closing call-to-action: "Which one do you use? Drop a comment for the link!"

# Style
- 300-500 words, never exceed 500
- At least 3 emojis
- Banned words: "literally", "iconic", "no cap", "bussin", "slay"
- Conversational; at most 3 sentences between "girl" / "hun"

# Output
Give the publishable Xiaohongshu post directly, no preface or explanation. End with 5 precise hashtags.
✨ What changed (and why it works)
  • Specifies a role ("senior blogger") + credentials ("200+ posts, 500K peak views") so the AI enters "expert mode"
  • Lists 3 real usage scenarios to defeat "vague scenario = vague output"
  • Sets word cap + banned words so the AI never produces filler
  • Demands "no explanation" output format so you don't need to delete preambles
Try it in ChatGPT Want AI to auto-generate this?
Case 2 · Programming

Production-grade Python scraper

Backend engineer · ship code on the first try

😣 The pain

Asking "write a Python scraper" gets you a half-baked tutorial: no error handling, no rate limiting, no tests. Run it in production and you get IP-banned within 10 minutes.

Before · user's rough input
write a Python script to scrape web pages
After · PromptOpt style
You are a senior Python backend engineer specializing in web scraping and anti-ban strategies. Over the past 5 years you have built production scrapers for 10+ companies.

# Task
Build a production-grade Python scraper that crawls a specified news site's homepage article list.

# Input parameters
- target_url: URL to scrape (e.g. https://news.example.com)
- max_pages: max pages to crawl (default 1)
- rate_limit: seconds between requests (default 1.5)

# Output format
JSON list, each record contains:
- title (str): article title
- url (str): detail page URL
- publish_time (str): ISO 8601 timestamp
- summary (str): first 200 chars

# Required capabilities
1. Error handling: 404 / timeout / encoding / JSON-parse failures → catch + log, never crash
2. Rate limiting: gap between requests must be >= rate_limit seconds
3. Respect robots.txt
4. UA spoofing: random pick from a 10-real-UA pool
5. Full English docstring (one line per function + inline comments on key steps)

# Constraints
- Only use requests / beautifulsoup4 / lxml
- No Selenium or browser automation
- Python 3.10+ compatible
- PEP 8 style

# Deliverables
1. scraper.py (with shebang + if __name__ == '__main__')
2. test_scraper.py (3 pytest cases: happy / 404 / rate-limit)
3. README.md (run instructions + dependency list)
✨ What changed (and why it works)
  • Role + credentials ("5 years, 10+ companies") → no more "tutorial code"
  • Explicit input parameters → clear, reusable interface
  • JSON output schema (field names + types) → no more parsing random strings
  • Lists 6 required capabilities + constraints → error handling, rate limit, UA spoofing all baked in
  • Demands test_scraper.py → drops straight into CI
Try it in ChatGPT Want AI to auto-generate this?
Case 3 · Business

SaaS competitive analysis · Q4

Product manager · 1 hour to a board-ready deck

😣 The pain

Asking "analyze my product vs competitors" returns a fluffy SWOT with no data sources, no comparison matrix, and no actionable recommendations. The CEO says "redo it".

Before · user's rough input
analyze my product vs competitors
After · PromptOpt style
You are a top-tier SaaS industry analyst. Over the past 10 years you have produced competitive analyses for 30+ SaaS companies.

# Task
Based on the inputs below, produce a Q4 competitive analysis report that is directly board-ready.

# Inputs
- **Our product**: PromptOpt (AI prompt optimizer, priced US$5/29/99 one-time)
- **Main competitors** (3):
  1. AIPRM (subscription, $20/month)
  2. PromptPerfect (subscription, $16/month)
  3. ChatGPT Official Prompt Library (free)

# Analysis dimensions (each dimension must include specific observations + source citations)
1. **Pricing model**: lifetime vs subscription LTV differences (cite public financials)
2. **Feature gaps**: each competitor's 3 unique features + 3 missing features
3. **User segments**: power users vs one-off users
4. **Market positioning**: summarized from official sites + G2/Capterra reviews
5. **Moat**: technology moat vs marketing moat

# Output format
- 2000-word analysis report
- Must include 1 Markdown comparison matrix table
- End with "3 differentiation opportunities for PromptOpt", each actionable

# Data source constraints
- Prefer: G2 / Capterra reviews / official sites / public funding news
- Cite a source URL for every data point
- Unknown = "not disclosed", never fabricate

# Style
- Objective and neutral, no vendor bias
- Bold key conclusions
- Paragraphs <= 5 lines
✨ What changed (and why it works)
  • Role + credentials ("10 years, 30+ companies") → AI enters senior analyst mode
  • Lists specific competitors + pricing + business model → no generic "industry overview"
  • 5 dimensions + each demands "source citations" → no hand-waving
  • Source constraint ("unknown = not disclosed") → AI can't fabricate financials
  • Word cap + mandatory matrix + 3 actionable recommendations → CEO can decide on the spot
Try it in ChatGPT Want AI to auto-generate this?

Don't want to write them by hand? Let PromptOpt do it

For the same 3 scenarios, sign in and type your “rough idea” — get a PromptOpt-style prompt in 5 seconds. New accounts get 10 free credits, no credit card.

Start free ← 中文版
中文