Before you pay, tell apart information anxiety from a course that actually fits you.
Fill in four fields and hand a "course due diligence" to ChatGPT, Gemini, Claude or Perplexity. It digs through PTT, Dcard, Threads, Vocus and YouTube comments for real reviews, checks out the instructor, and ends with a clear buy or don't-buy. The prompt is in Chinese on purpose: it searches Taiwanese communities.
The more specific, the sharper the verdict
Prompt version
Log in to the platform first
Prompt preview (Chinese)0 characters
Loading…
01 · Why I built it
A course's review wall shows 100 glowing reviews because the seller picked them. The most important rule: if it can't find criticism, it says so, and explains why (too new? the community lives in a private Discord?).
02 · How to use it
Course URL, your background, the problem you want solved, your budget. The more specific, the sharper the verdict.
Full works with any AI; Lite fits in a URL and fills itself in; Threads pairs with Meta AI to dig up scattered reviews.
Every button copies to your clipboard first. Platforms that accept a prefill get it automatically; for the rest, open a new chat and paste.
| Version | Length | Use with | Behaviour |
|---|---|---|---|
| Full | ~2,100 chars | Any | Seven-part report. Too long for a URL, so the button copies it; paste into a new chat |
| Lite | ~750 chars | ChatGPT / Claude / Perplexity | Short enough for a URL, so the question fills itself in |
| Threads | ~1,400 chars | Meta AI | Collects evidence only, no verdict; copy and paste |
03 · What it checks
Price, one-off or subscription, refund policy, number of lessons, last update, and whether any lesson is still marked "in production" (you'd be buying futures).
PTT, Dcard, Threads, public Facebook groups, Vocus, blogs, YouTube comments. It searches both positive and negative keywords.
Two columns: facts that can be cross-checked, and claims only they make. Then one question: do they practise this field, or mainly teach it?
Whether free resources are enough. An instructor's free content is usually the floor of the paid course.
04 · Evidence grades
The most common way AI gets reviews wrong: it finds the seller's "100+ happy students" wall and concludes the course is well loved.
Most trustworthy
Read the trend, not single posts
Treat as ads
Not evidence of reputation
05 · Nine red flags
Each marked ✅ fine, ⚠️ watch out, or 🚩 high risk.
06 · Honesty rules
The most important part of the prompt: no criticism found is not the same as a good reputation. No made-up reviews or links, a real URL for every quote, facts kept apart from inference, and if the price isn't published, it says so instead of guessing.
07 · FAQ
No and no. Free, no sign-up, open source. Just open it and use it.
The tool is a static page with no backend. What you type stays in your own browser for next time; when you send it, it goes to the AI you picked.
Prefill parameters are a convention, not a public API, and Gemini and Meta AI don't have one. So every button copies to your clipboard first; the prefill is a bonus, and nothing breaks if it stops working.
No. Check the results yourself and open the cited links. That's why the prompt demands a real URL for every quote.
There are reviews on Threads that outside search engines can't reach, and the honest ones are often in the replies. The Threads version, used with Meta AI, only collects them without a verdict; paste the list back into the full report as grade B evidence.
Use it however you like, and share your changes too. The source is on GitHub.
Who made this
Former Senior Technical Manager (E10) at MediaTek, two-time founder. I want to help people who, like me, feel stuck at work and want to find their passion, use AI to try things and discover work that makes them happy.
Opens Substack to confirm. One more click and you're in.