From Benchmark Dissection to Original Copy
Parse one benchmark creator's five scripts with the Social Media Assistant, break down and iterate the style prompt, then write an original piece of your own.
This assignment is the Lesson 2 workflow run end to end: pick a benchmark creator → parse five of their scripts → break down the style and iterate the prompt → write your own original. You submit four items, and missing any one means the assignment is not complete: the Feishu parsing table, the style analysis, the original copy, and course feedback.
Assignment goals
- Run the full find-benchmarks → break-structure → fill-content workflow instead of just watching;
- End up with a style-cloning prompt for your own niche that you can reuse afterwards;
- Produce your first genuinely original piece — the style is dissected, the story is yours.
Three parts
Part 1 · Pick a creator, parse five scripts
Find a benchmark creator you like, and use the Social Media Assistant + a Feishu multi-dimensional table to parse five of their video scripts in bulk.
| Field | My answer |
|---|---|
| Benchmark creator (account + platform) | |
| Why this creator (data / niche / style) | |
| The five videos parsed (titles or topics) | |
| Link to my Feishu parsing table |
Reminder: choose the creator by data (likes, completion, comments), not by I think they shoot well. Remember to normalise the transcripts before analysis — dirty data in, skewed frameworks out.
Part 2 · Break the style, iterate the prompt
Use the in-class style prompt (the viral reverse-engineering and original-writing agent) to analyse the expression traits of these five scripts, then adjust and iterate based on the results.
| Field | My answer |
|---|---|
| First-3-second hook type (pain contrast / denying common sense / fear of loss / cognitive flip) | |
| Sentence-length range and rhythm | |
| High-frequency catchphrases / connectives (at least 3) | |
| Emotional curve (how each stage rises and falls) | |
| My iterations on the prompt (what I changed and why) |
Evidence, not feelings: every field needs a quantified result (length range, counts, ratios). I feel their writing is conversational does not count as complete.
Part 3 · Write about yourself, finish the original
Combine your real experience and stories to generate an original piece that belongs to you.
| Field | My answer |
|---|---|
| Video type (business acquisition / industry exposé / persona story) | |
| Target user and reasons to buy | |
| Real material used (experience / cases / stories) | |
| Original copy (full text attached) |
Two self-checks: one, does the writing carry the benchmark's rhythm (short sentences, conversational, textured); two, is every piece of content from your own business and experience — replicate the method, never the content.
Grading
| Level | Standard |
|---|---|
| Pass | All four items complete, the parsing table has real data, the analysis is quantified, the original copy is complete |
| Good | On top of pass: prompt iterations have clear evidence (what changed and why), and the original shows traces of the style transfer |
| Excellent | On top of good: the style-cloning prompt is reusable standalone, the original weaves in specific reasons to buy and real stories, and it reads with no AI flavour |
Submission
- Upload screenshots of your finished work to the website for submission;
- Pick the piece you are most satisfied with and share it in the course group so everyone can see your workflow.
Deadline: Parts 1 and 2 within 48 hours while the workflow is fresh; Part 3 within 7 days — leave yourself time to polish the original. All four items (Feishu parsing table, style analysis, original copy, course feedback) are required; missing any one means not complete.
Tools and Workflow Consolidation: From One Hit to a Content Asset
Connect the breakdown sheet to real filming or AI-generated footage, reuse the cloned style, and fix the method into a weekly workflow one person can run.
Everyone a Product Manager: Build Your Own Tools with AI
Six questions, one checklist, one working first version — you make the judgement calls, AI writes the code, and you verify it yourself at the end.
Tutorials