The prompt that plans 30 days of content from your 3 best posts
Hey, it's Cindy 🌱 You give it your 3 best posts with their real numbers. It works out why they took off, builds your pillars from that, and hands you a 30-day calendar.
btw the waitlist for AI System Club is now open! If you're keen on using AI to automate your content, video editing, job search, business or build AI agents, feel free to sign up (only if you want to hehe). Ps. the waitlist gets first dibs:)
The contract, in one line. You provide 3 winning posts with their numbers, one flop and 3 captions that sound like you. The prompt diagnoses, profiles your voice, builds pillars and then plans. You receive a full 30-day calendar plus your first week written out and ready to film.
📊 Your 3 best posts, with the real numbers. Pick them by saves and comments, not by the ones you personally liked. On Instagram, open the app, go to your profile, tap the menu and open Insights, then Content you shared, and sort by saves. Copy the caption or script of each one plus its views, saves, comments and roughly when it went up.
📉 One post that flopped. This is the input almost everyone skips and it is the one that sharpens the whole diagnosis. Without a flop the model can only tell you what your winners have in common. With one, it can tell you what is actually different about them.
✍️ 3 captions that sound most like you. Not your best-performing ones, your most you ones. This is what the voice profile is built from, so pick the ones a friend would recognise.
📅 Your platform and how many days a week you post. Four days a week across 30 days is about 17 posts, and the calendar will be that many rows. Be honest here rather than aspirational, because a plan you cannot keep is the same as no plan.
💰 What you are promoting this month, and the dates. If you give it an open and close date, it builds toward that date instead of sprinkling promo evenly across the month. If you are promoting nothing, write "nothing" and it plans a pure value month.
the prompt
Paste this into Claude
Open a new chat at claude.ai, paste the whole thing in, and fill in the square brackets before you send. It is long on purpose. Every step is doing a job that a short prompt skips, and the two steps people usually cut, the diagnosis and the self-check, are the two that make the output usable.
🗓️ the 30-day content planner
You are my content strategist. You plan from evidence, not from guesses, and you write in my voice, not yours.
MY INPUTS
- Niche and who I help: [e.g. AI tips for busy marketers]
- Platform and posting days per week: [e.g. Instagram, 4]
- My 3 best posts. For each one: the caption or script, the format, views, saves, comments, and how long ago it went up:
[post 1] [post 2] [post 3]
- 1 post that flopped, same details, if I have one: [paste, or write "none"]
- 3 captions that sound most like me: [paste]
- What I am promoting this month and the date it opens or closes: [offer, or "nothing"]
STEP 1. Diagnose. For each top post, quote its opening line word for word, then name the hook mechanism (curiosity, contrarian, outcome, scenario, list), the format, the promise it made, and why it earned saves rather than only views. List the patterns all three share. If I gave you a flop, say what it did differently, quoting both sides.
STEP 2. Voice. Write a 5-line voice profile from my captions: my average sentence length, the words and phrases I actually use, the words I never use, how I handle humour, and how I sign off, including how I capitalise my CTAs.
STEP 3. Pillars. Turn the patterns into 3 or 4 content pillars. Name each one in plain language, say what it is for, which of my top posts proves it works, and what share of the month it should own.
STEP 4. Calendar. Build a table covering the next 30 days, one row per posting day: Day, Date, Pillar, Format, Hook type, Hook (in my voice, ready to say out loud), the 3 key points, and the CTA written exactly the way I write mine. Never run the same pillar or the same hook type two posting days in a row, and keep promo to no more than 1 post in 5. If I gave you an offer with a date, build toward that date instead of spreading the promo evenly.
STEP 5. Check your own work before you show me. Find every day that anyone in my niche could have written and rewrite it so it could only come from me and my evidence. Find every hook that is a question with an obvious yes and rewrite it. Then list the 3 days you are least confident about and say why.
STEP 6. Give me the first week in full: for each post, the hook, the opening 3 lines, and the CTA, ready to film or type.
If an input is missing, ask me for it before you start, and never invent numbers for my posts.
what each step is doing
Why the long version wins
If you ever want to change this prompt, change it knowing what each part is holding up. Here is the job every step is doing.
Step 1 diagnoses before it plans. Asking it to quote your opening lines word for word is what stops it summarising your posts into a vague theme. Once the real hooks are on screen, the patterns it names are grounded in something you can check. The flop is the control group.
Step 2 makes it sound like you, not like a brand. Three captions is enough for a usable voice profile. Asking for "words I never use" matters more than it looks, because that list is what keeps the calendar free of unlock, game-changer and transform.
Step 3 keeps the pillar count sane. Three or four is the working range most strategists land on. Sprout Social puts it at "at least around 3-5 pillars at any one time", and warns that too many dilutes the message. Each pillar is tied back to the post that proves it, so none of them are invented.
Step 4 is the only part most people ask for. The Hook type column exists so you can see the rotation with your own eyes. Without it the no-repeats rule is a promise you cannot check, and the model will quietly run three outcome hooks in a row.
Step 5 is the step that earns the reel. Asking a model to go back and find its own generic work is a self-check, and it is one of the techniques Anthropic's own guidance recommends. In testing this rewrote three days that were recycled takes anyone in the niche could have posted.
Step 6 gets you filming today. A calendar you have to re-prompt every Monday tends to die by week two. The first week comes back with the hook, the opening three lines and the CTA written out, so there is nothing between you and the camera.
what it actually gives back
A real run, start to finish
This is what came back when the prompt was run on a marketer posting four days a week on Instagram, with three winning posts, one flop and a $47 offer opening and closing on set dates.
The diagnosis quoted the real openers It pulled each post's first line word for word, named the mechanism (outcome, list, scenario), and separated why the carousel earned a 16% save rate while the scenario reel earned the comments.
The voice profile was uncomfortably accurate Sentence length of 6 to 14 words, a list of words never used, and the note that the sign-off is peer to peer rather than a call to action.
Four pillars, each tied to a post Not four themes invented out of nowhere. Every pillar named the post that proved it and the share of the month it should own.
Seventeen planned posting days Four a week across 30 days, with the promo posts clustered at cart open, mid-launch and close rather than spread thinly through the month.
The self-check is worth watching. In that run it flagged three of its own days as recycled takes, showed the before and after, and then named the three days it was least confident about and why. One of them was a hook that assumed the audience were freelancers rather than in-house, which is exactly the kind of thing you want flagged before you film it.
if it comes back flat
4 fixes that almost always work
The hooks sound like a brand, not like you Your three captions were probably your best-performing ones rather than your most natural ones. Swap them for three you would have sent to a friend and re-run Step 2 only.
Every day feels interchangeable You skipped the flop. Add one, then say: rerun Step 1 with the flop included, and tell me what my winners have that it does not.
It repeats the same hook shape Check the Hook type column. If three of the same type run together, reply: rebuild the calendar so no hook type repeats within three posting days, and keep everything else.
The month feels like one long ad Reply with: cap promo at one post in five, move the rest of the promo weight into the three days before the cart closes, and rewrite the value posts so they set up the offer without mentioning it.
keeping it alive
The Monday top-up prompt
A 30-day plan built in one sitting starts drifting the moment real results come in. Run this one on a Monday, in the same chat, so the plan updates from what actually happened rather than from what you assumed in week one.
🔁 the weekly top-up
Here is what actually happened last week: [paste each post you published with its views, saves and comments, and note anything unusual, like a post that went out late or a week you were away].
Update the plan, do not rebuild it. First, tell me which pillar is outperforming and which one is underperforming, using the numbers rather than your impression, and quote the specific hook that did best. Then rewrite the next seven posting days: keep anything that is still working, replace any day whose pillar is underperforming, and add one post that pushes further in the direction of the best hook from last week. Keep the same table columns and keep the CTA style unchanged. Finish by telling me the one thing you would test next week that I have not tried yet, and why you think it would work for this audience specifically.
good to know
Plans, limits and the ChatGPT version
🆓 The free plan is enough to run this. You do not need a paid account to paste a long prompt and get a calendar back. What a paid plan buys you is room to keep going in the same chat when you want the top-up prompt and a few rounds of edits, rather than hitting a limit halfway through week three.
📎 You can upload instead of paste. If your captions live in a doc or an export, attach the file rather than pasting. Claude takes up to 20 files per chat, and screenshots of your insights count, so you can drag in the numbers instead of typing them.
💬 It runs in ChatGPT with no changes. The structure it uses, a role, then the inputs, then numbered instructions, is the same shape OpenAI's own prompting guide recommends. Paste it as it is. The output formatting differs slightly, the plan does not.
📌 Keep it as a project, not a chat. Put your voice profile and your three winning posts into a Claude project, and every future content chat starts already knowing them. That is the difference between running this once and having it available every month.