AI 办公提效Extended Use Cases
Chapter 21
21 / 23

Social Media & Marketing Copy

⏱️ 25 min

Campaign plans, A/B copy, poster scripts, and post-mortems

CHAPTER PRACTICE GOAL
01Work problem

Campaign plans, A/B copy, poster scripts, and post-mortems

02Reusable output

A reusable work template with realistic input, a defined output format and a human review point.

03Definition of done

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:

ChannelReader contextContent priorityAppropriate CTA
XiaohongshuLooking for practical experienceScenario, process, before/afterSave the checklist or view details
LinkedInEvaluating professional valueDecision, method, evidenceRead the case or join the discussion
EmailAlready knows the brandTime, fit, concrete takeawayReply 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

FailureRoot causeFix
Five variants only swap adjectivesAngles were never definedTest conflict, benefit, evidence, and story as distinct hypotheses
Fluent copy does not sound like the brandNo real language samplesExtract sentence length, rhythm, and vocabulary from 3–5 approved posts
A weak result triggers a total rewriteToo many variables change togetherTest only title, visual, CTA, or audience in one round
AI invents a reason for lower conversionNo user evidenceReturn 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.