From Brief to Scheduled: Building One Week of AI-Assisted Social Content
Objective: Run a real content brief through the lesson's 5-stage playbook end to end: build a one-page brand voice guide, generate and prioritize ideas, draft platform-specific captions, and produce a ready-to-schedule week.
You're the marketing associate at Snowflake, the Nasdaq-listed (SNOW) cloud data platform, building next week's LinkedIn and X content around a new AI feature launch with a same-day turnaround.
Write a one-page brand voice guide, generate 15 ideas and cut to 5, draft each in its platform's format, and self-review every draft against the Stage 4 checklist.
Before you start
What you'll need
Free path (everything below is enough to finish)
Free, structured, and easy to share for a second review pass
Free tier covers basic AI-assisted graphics for social posts
The process
3 steps
Step 01 of 03
A one-page voice guide (3 adjectives, 3 banned words, 1 gold-standard post, a 2-sentence persona, and banned formats) pasted into every prompt prevents the most common AI failure: generic-sounding content.
Snowflake's brand voice is technical but not jargon-heavy. What 3 adjectives and 3 banned words would you set, and what's your gold-standard sample post?
Procedure
- List 3 voice adjectives, for example 'precise, direct, unshowy'
- List 3 banned words or phrases, for example 'game-changer', 'revolutionize', 'seamless'
- Paste in one real high-performing post as the gold standard
- Write a 2-sentence audience persona
- List 2 banned formats, for example no emoji-only openers
SNOWFLAKE VOICE GUIDE v1 Adjectives: precise, direct, unshowy Banned: 'game-changer', 'revolutionize', 'seamless' Gold-standard post: [pasted LinkedIn post, 180 words] Persona: Data engineers and platform leads who are skeptical of hype and want the technical detail first. Banned formats: no emoji-only hooks, no rhetorical-question openers
Healthy
Every caption drafted this week reads consistent with the guide without heavy editing.
Unhealthy
Each caption needs a full rewrite because the guide was skipped or too vague to constrain the model.
What this means
The guide is a constraint, not a suggestion; specificity in the guide is what saves editing time later.
So what do I do about it?
| Symptom | Action | Effort |
|---|---|---|
| Drafts keep drifting into generic SaaS voice | Tighten the banned-words list with the exact phrases you keep deleting | 5 min |
Step 02 of 03
Stage 2 feeds a structured brief (platform, tone, pillars, cadence) into an LLM, then triages ideas into write now, write later, or cut rather than scheduling everything generated.
The AI returns 15 ideas for the AI-feature launch. Your cadence this week is 3 LinkedIn posts and 2 X posts. How do you cut 15 down to 5?
Procedure
- Generate 15 ideas from a structured prompt covering platform, tone, pillars, and cadence
- Tag each idea by content pillar
- Mark 5 as 'write now' matching this week's cadence and launch relevance
- Mark the rest 'later' or 'cut'; do not schedule everything generated
IDEA TRIAGE (15 -> 5) WRITE NOW (5) 1. 'Why we built X the way we did' - LinkedIn, technical pillar 2. 'The 3am alert this feature kills' - LinkedIn, customer-pain pillar 3. '90 seconds: what changed' - X 4. 'A customer's before/after query time' - LinkedIn 5. 'One config, not five' - X WRITE LATER (7) / CUT (3) ...10 more rows
Healthy
5 ideas selected map directly to this week's cadence and the launch.
Unhealthy
All 15 get scheduled, diluting focus and burying the launch post.
What this means
Volume from AI ideation is a filtering problem, not a scheduling problem.
So what do I do about it?
| Symptom | Action | Effort |
|---|---|---|
| Calendar has 15 posts queued for one launch week | Cut to cadence capacity before drafting, not after | 5 min |
Step 03 of 03
Stage 3's format table sets a different length, hook style, and CTA per platform; a LinkedIn draft copy-pasted to X will fail the character limit and the format expectations.
Your LinkedIn draft is 240 words with a stat-led hook. What has to change to make it work as an X post?
Procedure
- Draft the LinkedIn version first, 150-300 words with a stat or bold-claim hook
- Cut and reformat the same idea for X, under 280 characters with a punchy statement
- Run every draft through the Stage 4 checklist for voice, accuracy, specificity, and cliches
- Flag any draft taking over 10 minutes of editing as a brief problem, not an editing problem
DRAFT SET (excerpt) LinkedIn (idea 1, 210 words): 'Query times dropped from 4.2s to 0.6s...' X (idea 3, 61 chars): '90 seconds. That's the whole demo now.'
Healthy
Each platform version respects its own length and hook convention.
Unhealthy
The same 210-word draft is posted unedited to X and gets truncated.
What this means
Reformatting is mandatory per platform, not optional polish.
So what do I do about it?
| Symptom | Action | Effort |
|---|---|---|
| X posts get cut off mid-sentence | Draft platform-native from the idea instead of copy-pasting and trimming | 5 min |
Final deliverable
A one-page brand voice guide plus 5 platform-formatted, checklist-passed posts ready to schedule for launch week.
See a reference example
Instacart, AI-feature launch week (excerpt) VOICE GUIDE: direct, warm, no-jargon LINKEDIN (idea 2, 190 words): 'Our on-call engineer used to get paged at 3am for...' X (idea 4, 240 chars): 'One customer's query time: 4.2s -> 0.6s. No new infra. Just a smarter default.'
Success criteria
You're done when you can:
- Voice guide includes all 5 required parts
- 5 posts map to this week's cadence and launch relevance
- Each post is reformatted per platform, not copy-pasted
- Every draft passes the Stage 4 checklist