Social Media & Marketing Copy
Campaign plans, A/B copy, poster scripts, and post-mortems
Campaign plans, A/B copy, poster scripts, and post-mortems
A reusable work template with realistic input, a defined output format and a human review point.
Run one real task, verify the critical facts and record the before-and-after time.
AI can handle "fast drafting + multi-version comparison + post-campaign review," making social/marketing output faster and more consistent.
1) Campaign Design
- Prompt example:
You are a marketing planner. Generate a campaign plan for [campaign goal / audience / channel].
Output: 1) Core selling points and hooks (3 options); 2) Copy direction by channel (email / WeChat Moments / Weibo / Xiaohongshu / community); 3) Timeline and asset checklist; 4) Risk and compliance reminders.
- Ask AI for "main storyline + backup creative angles" to enable A/B testing.
2) Multi-Version A/B Copy
- Have AI output 3-5 versions, each labeled with "angle / tone / length / best-fit channel."
- Email/tweets: also provide "subject line / preview text / CTA variations."
- Community/social posts: require "strong hook in first two lines + clear action point + link placement advice."
3) Visual / Poster Scripts
- Prompt:
Generate poster copy: title / subtitle / selling point bullets / CTA / notes. Then generate a shoot or design script (shots / duration / script lines). - For video, ask for "1-minute talking-head outline + storyboard + subtitle copy."
4) Asset Organization & Scheduling
- Have AI output "asset checklist + owner + deadline + reference links" with a calendar schedule.
- Turn recurring templates (launch announcements, discounts, livestream teasers) into reusable prompt snippets.
5) Post-Campaign Review & Iteration
- Review prompt:
Based on the following data/feedback, output: best and worst performing versions, likely reasons, and 3 improvements to try next time. - Ask AI to flag "channel differences / audience differences" to avoid one-size-fits-all conclusions.
6) Risk & Compliance
- Define banned words and compliance boundaries (exaggerated claims, medical/financial sensitive language, unauthorized brand assets).
- Before publishing external assets, have AI run a "risk scan + rewrite suggestions" pass.
7) Practice
Pick an upcoming campaign. Have AI output "3 copy sets + channel adaptation + asset checklist + schedule," then run a review template to create an improvement plan for next time.
8) Worked Example: From Campaign Brief to Review
Suppose you are promoting a two-hour, in-person AI workshop. Do not begin with “write a social post.” Lock one factual brief first:
Campaign facts:
- Audience: professionals who want to use AI in daily work
- Format: a two-hour, in-person practical session
- Participant outcome: one reusable workflow
- Do not claim: guaranteed promotion, instant mastery, or scarcity without evidence
Before writing copy, return:
1. Three questions this audience is likely to ask
2. One content angle for each question
3. Missing facts marked [TO CONFIRM]
After confirming the facts, adapt the message by channel:
| Channel | Reader context | Content priority | Appropriate CTA |
|---|---|---|---|
| Xiaohongshu | Looking for practical experience | Scenario, process, before/after | Save the checklist or view details |
| Evaluating professional value | Decision, method, evidence | Read the case or join the discussion | |
| Already knows the brand | Time, fit, concrete takeaway | Reply or open registration |
For the review, provide real performance data and comments. Require AI to separate evidence from interpretation:
Below are impressions, dwell time, clicks, registrations, and comments for three variants.
Separate:
- conclusions directly supported by data
- hypotheses that require another test
- one variable to change next round
Do not present correlation as causation.
9) Human Publishing Gate
- Names, dates, prices, locations, and inclusions match the source of truth
- No unsupported “best,” “guaranteed,” or scarcity claims
- The copy sounds specific to this brand and audience
- Images, music, quotes, and user content have usage rights
- There is one primary CTA and the destination works
- The published variant is recorded so metrics can be traced back
10) Failure Modes and Fixes
| Failure | Root cause | Fix |
|---|---|---|
| Five variants only swap adjectives | Angles were never defined | Test conflict, benefit, evidence, and story as distinct hypotheses |
| Fluent copy does not sound like the brand | No real language samples | Extract sentence length, rhythm, and vocabulary from 3–5 approved posts |
| A weak result triggers a total rewrite | Too many variables change together | Test only title, visual, CTA, or audience in one round |
| AI invents a reason for lower conversion | No user evidence | Return to comments, enquiries, and landing-page behaviour; mark known vs inferred |
11) Chapter Deliverable
Leave with a reusable Campaign Pack: factual brief, three channel variants, publishing gate, version record, and review template. Continue to AI Governance and Metrics to make content production measurable and auditable.