Most SMM content plans die on day four. Not because the ideas were bad — because there was no grid. A list of 30 clever post ideas feels productive until you realize nobody assigned them to funnel stages, nobody checked whether three carousels in a row is too many, and nobody planned what happens when a Reel flops.
The myth is that a content plan is a creative artifact. The working model is simpler and far less romantic: a content plan is a grid — rubrics × goals × frequency × formats — and an LLM is very good at filling grids when you give it the parameters. It is not good at "inspiring" you, and you should stop asking it to.
This roundup walks through six prompts that turn ChatGPT (or Claude, or any capable model) into a planning instrument: a rubric generator, a 30-day grid builder, a format-specific post writer, a hook editor, a repurposing engine, and a weekly metrics reviewer. Each comes with the task, the prompt itself, and a realistic example of what comes back.
Why the grid model beats the idea model
A rubricator (USDрикатор) is the skeleton: 4–7 recurring content categories tied to what the audience needs and what the brand can credibly say. A grid is the calendar view: each day gets a rubric, a funnel goal (awareness, consideration, conversion, retention), and a format (carousel, Reel, story poll, static, text).
When you brief an LLM with those three axes, it stops hallucinating "10 viral hooks" and starts doing combinatorics — which is what planning actually is. You gain speed and consistency; you pay for it with the need to define the axes precisely up front. Vague inputs produce a vague grid, and a vague grid is just an idea list with dates attached.
The most common failure mode: the model quietly collapses all rubrics into "educational post" because that is the statistically safest output. You will spot it immediately if you count rubric distribution before approving anything.
1. Rubric generator: build the skeleton before the calendar
Task: turn a niche and audience into 5–6 non-overlapping rubrics, each with a purpose and a format bias.
Prompt:
You are an SMM strategist. Build a content rubricator for [niche] targeting [audience description].
Constraints:
- Exactly 5 rubrics, no overlap in purpose.
- For each: name, one-sentence purpose, which funnel stage it serves (awareness / consideration / conversion / retention), preferred formats, and a realistic posting frequency per month.
- Add one "risk" rubric the brand should use sparingly and explain why.
Output as a table. Do not invent statistics about the niche.
Example result for a specialty coffee roastery targeting home baristas:
| Rubric | Purpose | Funnel stage | Formats | Frequency/month |
|---|---|---|---|---|
| Bean science | Explain origin, roast, extraction | Awareness | Carousel, Reel | 6 |
| Brew tutorials | Teach a repeatable recipe | Consideration | Reel, carousel | 8 |
| Equipment picks | Help choose gear | Consideration | Static, carousel | 4 |
| Behind the roast | Show process and people | Retention | Stories, Reel | 6 |
| Offers & bundles | Drive purchase | Conversion | Static, story poll | 4 |
| Hot takes | Provoke discussion (risk rubric) | Awareness | Text, Reel | 2 |
The "hot takes" rubric is the risk one: it earns reach but can alienate the core audience if used more than a couple of times a month.
2. The 30-day grid: assign rubrics, goals, and formats
Task: distribute the rubricator across a month without clustering.
Prompt:
Using the rubricator above, build a 30-day content grid for [month].
Rules:
- No rubric appears twice in a row.
- At least 40% of posts serve awareness, 30% consideration, 20% conversion, 10% retention — adjust if the rubricator makes this impossible and say why.
- Each row: day, rubric, funnel goal, format, working title, one-line angle.
- Mark 4 days as "flex" slots for reactive content.
Output as a table with 30 rows.
Example result (excerpt):
| Day | Rubric | Goal | Format | Working title | Angle |
|---|---|---|---|---|---|
| 1 | Bean science | Awareness | Carousel | Why your Ethiopian tastes like berries | Flavor compounds explained simply |
| 2 | Brew tutorials | Consideration | Reel | The 4-minute V60 | Timing mistakes |
| 3 | Offers | Conversion | Story poll | Which bundle? | Two-option vote |
| ... | ... | ... | ... | ... | ... |
| 14 | Flex | — | — | — | Reserved |
The grid is the deliverable. If the model returns prose instead of a table, the prompt was too loose — tighten the output format line.
3. Format-specific writer: carousel, Reel, story poll
Task: turn one grid row into a ready post in the right format.
Prompt:
Write a [format] post for the grid row: "[working title]", rubric [rubric], goal [goal].
If format = carousel: 7 slides, each with a headline (max 6 words) and one supporting sentence; slide 1 is the hook, slide 7 is the CTA.
If format = Reel: 30-second script with hook (first 3 seconds), 3 beats, and CTA; include on-screen text suggestions.
If format = story poll: one question, two options, and a follow-up story idea based on each answer.
Tone: [tone]. No hashtags. No emojis in headlines.
Example result for a carousel on "Why your Ethiopian tastes like berries":
- Slide 1: "Your coffee isn't fruity. Your roast is hiding it."
- Slide 2: "Light roasts preserve volatile aromatics. Dark roasts burn them off."
- Slide 3: "Ethiopian beans carry floral and berry compounds — if you let them survive the drum."
- ...
- Slide 7: "Save this before your next order. Full guide in bio."
The value here is not creativity — it is that the post conforms to the format rules you set, which is what makes a grid executable by a team.
4. Hook editor: rewrite the first line until it earns the scroll
Task: take a flat opening and produce five variants with different mechanisms.
Prompt:
Rewrite this opening line five ways: "[original]".
Use these mechanisms, one each: a specific number, a contrarian claim, a mistake the reader is making, a question that implies stakes, a before/after contrast.
Keep each under 12 words. Do not use clickbait that the post cannot deliver on.
Example: original — "Coffee brewing is important." Variants: "Most home brews are 10°C too hot." / "Your grinder is why your coffee tastes muddy." / "You are probably over-extracting — here's the tell." / "What if the problem is your water, not your beans?" / "Same beans, two brews, opposite results. Here's why."
The trade-off: sharper hooks raise click-through but also raise the expectation bar. A hook the post cannot satisfy costs more in retention than it gains in reach.
5. Repurposing: one post into five formats
Task: extract maximum value from one piece of content without rewriting from scratch.
Prompt:
Take this post: "[paste]." Repackage it into five formats:
1. A 60-second Reel script.
2. A 5-slide carousel.
3. A story poll with two options.
4. A short text post under 280 characters.
5. A newsletter paragraph under 100 words.
Keep the core claim identical across all five. Change only structure and length.
Example result for the Ethiopian carousel: the Reel opens with the same hook, the text post compresses it to "Light roast or nothing — your berries are burning," and the newsletter expands the slide 2–3 argument into a paragraph.
This is where LLMs save the most real hours, because repurposing is mechanical: same claim, different container. If the model changes the claim between formats, you have a consistency bug, not a creative win.
6. Weekly metrics review: correct the plan, don't just report it
Task: feed last week's numbers back into the grid and adjust the next week.
Prompt:
Here are last week's metrics by post: [paste table: post, format, rubric, reach, saves, replies, link clicks].
Identify: (a) the rubric with the highest save rate, (b) the format with the lowest reach, (c) one post that underperformed its rubric average.
Then propose three concrete changes to next week's grid — swap a format, increase a rubric's frequency, or retire an angle. Justify each change with the numbers above. Do not invent metrics I did not provide.
Example result: the model notes that brew tutorials have the highest save rate but the lowest reach, suggests converting one tutorial into a Reel, and flags a static offers post that underperformed its rubric average — proposing a story poll instead.
The limit: with one week of data the model is pattern-matching on noise. Feed it three to four weeks before trusting its recommendations, and keep a human in the loop for anything that changes the rubricator itself.
What this actually changes
A grid built this way is auditable: you can point to day 14 and say which rubric, which goal, and which format it serves. That is the difference between a plan and a wish list. The prompts above do not make the model creative — they make it consistent, which is what a monthly content plan needs.
Start with prompt 1, build your rubricator, then run prompt 2 and count the distribution before you write a single post. If you want more prompt collections for practical work, browse our other guides — for example, 12 LLM prompts for mobile release day or prompts for salary negotiations.