Most social media teams don't fail because they lack ideas. They fail because the pipeline between the idea and the published post is manual, inconsistent, and impossible to audit. A rubricator lives in someone's head, the content plan lives in a spreadsheet nobody opens, posts are written the night before, and engagement reports are a screenshot with a vague "reach is down, we need more Reels."
Prompts don't fix that by magic. But a structured prompt collection — organized by task, not by vibe — turns an LLM into three different colleagues: a strategist who builds the rubric, a copywriter who fills it, and an analyst who reads the numbers back. The key word is structured: the same prompt that produces a brilliant post on Tuesday produces a bland one on Thursday if the input schema drifts.
Below are 15 prompts organized by difficulty. Each one is a working template with a defined input, a defined output shape, and an example result. Treat the output schema as the contract — that's what makes the collection reusable instead of a pile of one-off chats.
Basic level: get the rubricator and the post skeleton right
1. The rubricator generator
Most accounts don't have a rubricator — they have a habit. This prompt forces a finite set of content pillars with a purpose for each.
You are a content strategist. Build a rubricator for the account below.
Account: {niche}
Audience: {who they are, what they fear, what they want}
Goal: {leads / retention / brand}
Return a table: pillar | purpose (educate/prove/sell/engage) |
format | posting frequency per month | example topic.
Rules:
- 4 to 6 pillars, no more.
- At least one pillar must be purely proof (cases, numbers, screens).
- No pillar may exist only to sell.
Example result: for a B2B accounting service, the model returns five pillars: "myth vs. law," "client case in 3 screens," "checklist," "founder's voice," "offer." The useful part isn't the pillars — it's the frequency column, which forces the team to admit they can realistically produce 2 cases a month, not 8.
2. The content plan for a month
Using the rubricator above, produce a 30-day content plan.
Format: date | pillar | format | working title | hook (max 12 words) | CTA.
Constraints:
- No two consecutive posts from the same pillar.
- Fridays are engagement posts (poll, question, hot take).
- Leave 4 empty slots marked [REACTIVE] for news.
Why the empty slots matter: a plan that fills all 30 days breaks on day 3 when something newsworthy happens, and the whole document loses authority. Reserving reactive capacity is the cheapest way to keep the plan alive.
3. The post skeleton
Write a post from this plan line: {title, hook, pillar, CTA}.
Structure: hook (1 line) → tension (2-3 lines) → payload (3-5 lines) → CTA (1 line).
Tone: {tone}. Length: {n} characters.
Do not use emojis. Do not start with a question.
4. The hook multiplier
Here is a post: {text}.
Generate 8 alternative first lines. Each must use a different device:
number, contradiction, mistake, before/after, direct address,
mini-story, unpopular opinion, question-free curiosity gap.
Do not change the body.
This is the single highest-leverage prompt in the set. In most accounts the hook decides reach more than the body does, because the body is only read by people who stopped scrolling.
Advanced level: make the model produce assets, not text
5. The carousel decomposer
Turn this article into a 7-slide carousel.
Per slide: headline (max 6 words) | body (max 25 words) | visual note.
Slide 1 = promise, slide 7 = CTA. No slide repeats the previous idea.
The trap: models default to splitting text evenly, producing seven slides that each say "and here's more context." Add an explicit rule — each slide must be able to stand alone if someone screenshots only that one.
6. The visual brief for an image model
For each carousel slide, write an image prompt for {Midjourney/Flux}:
subject, composition, palette (hex), lighting, style, negative prompt.
Keep one consistent visual system across all 7 slides.
7. The engagement analyst
Here is a CSV of posts: date, format, pillar, reach, likes,
comments, saves, shares, follows.
Compute per format and per pillar: median reach, save rate,
comment rate. Identify the top 3 and bottom 3 posts and explain
what they share. Flag any post where reach is high but saves are low.
The non-obvious insight: saves and shares behave differently from likes. A post with high reach and low saves is usually entertaining but not useful — it buys attention you can't convert. A post with modest reach and high save rate is your evergreen asset, and it deserves reposting in a different format, not replacing.
8. The rubricator auditor
Compare the last 60 posts against the rubricator.
Output: pillar | planned share | actual share | verdict.
Then answer: which pillar is overproduced relative to its engagement
contribution, and which is starved?
9. The comment-to-content converter
Here are 40 real comments and DMs.
Cluster them into recurring questions. For each cluster:
question | frequency | suggested format | draft answer (max 40 words).
Ignore one-off complaints.
This turns the comment section into a research channel — the cheapest source of content topics you will ever have, and the one most teams never read systematically.
Expert level: pipelines, guardrails, and reporting
10. The brief-to-pipeline prompt
You are a content ops lead. Given the brand brief below, produce:
1) a rubricator, 2) a monthly plan, 3) a prompt set for writers,
4) a QA checklist, 5) a weekly reporting template.
Every artifact must reference the same pillar names.
The value is naming consistency: if the plan says "case" and the report says "success story," your analytics are broken before they start.
11. The brand voice lock
Rewrite this post in our voice. Voice rules:
- sentences under 15 words;
- no adjectives that can't be proven;
- no "we are proud";
- numbers instead of superlatives.
Return the rewrite plus a list of every rule violation you removed.
Asking for the violation list is what makes this auditable. Without it, you can't tell whether the model followed the voice or just got lucky.
12. The failure-mode reviewer
Review this draft as a skeptical follower.
List: claims that need a source, promises the brand can't keep,
lines that read as AI-generated, and one reason to scroll past.
Do not rewrite. Only diagnose.
Separating diagnosis from rewriting is the difference between an editor and a text generator. A model asked to "improve" a post will smooth away the sharp edges that made it work.
13. The engagement report generator
Input: last month's CSV and last month's plan.
Output a report: what we planned vs. what we published,
top 3 insights with the numbers behind them, 2 experiments
for next month, 1 thing to stop doing. Max 400 words.
No adjectives without a number next to them.
14. The A/B hypothesis generator
From the engagement data, propose 5 testable hypotheses.
Format: hypothesis | variable (hook/format/time/CTA) |
metric | minimum sample | how we'll know it's noise.
15. The monthly retrospective prompt
Write a retrospective: what worked, what didn't, what we learned
about the audience, what we'll change. Reference specific posts.
Keep it honest — no motivational filler.
Two trade-offs worth naming
Speed vs. distinctiveness. A locked prompt set lets you publish five times faster, and it also makes you sound like everyone else using the same templates. The fix isn't fewer prompts — it's keeping one pillar that is deliberately un-promptable: the founder's voice, the raw case, the screenshot with a real number. That pillar is your moat.
Automation vs. taste. Every step you hand to the model removes a decision you no longer make consciously. The prompts above are designed so the model proposes and the human selects — because selection is the part that still requires knowing your audience. For a deeper look at how this plays out on the marketing side, see our breakdown of marketing and copywriting prompts, and if your output feeds a publishing workflow, the fact-checking workflow is worth reading before you scale.
A realistic first week
Don't deploy all 15 at once. Start with three: the rubricator generator, the content plan, and the engagement analyst. Run them for one month. If your plan and your report use the same pillar names, you'll see for the first time which content actually earns its place — and which pillars you keep producing out of habit.
That's the whole promise of SMM prompts, honestly stated: not creativity on demand, but a repeatable pipeline where strategy, production, and analysis share one vocabulary. The model doesn't make you a strategist. It makes the strategist's decisions visible — and visible decisions are the ones you can improve.