How to Plan a Week of Content With AI

I wrote this in August 2026. Menus move, so if a button isn't exactly where I say it is, look nearby, because the idea stays the same.

The worst part of my week used to be the Sunday content calendar, when I'd open a blank page and try to invent 5 ideas out of my own head while a toddler pulled on my leg.

It's hard because you're doing two jobs at once. You're trying to figure out what people want to hear and also trying to write it, and your brain is bad at doing both at the same time.

So split them up. Let Claude go read what already worked in your niche first, and then let it write. You need two free accounts and one connection you set up once, and after that it's about 30 minutes on a Sunday.

Is this just copying other people?

This is the first question people ask, so let's deal with it up front.

You're pulling public posts to see what worked, and "what worked" is really two separate things.

One is the topics that made people stop and type something. If 4 accounts in your niche all had a post about pricing blow up in the last 2 weeks, that tells you the women following those accounts are thinking about pricing right now.

The other is how those posts were written, meaning what the first line did, how the caption was built and whether the posts that got 300 comments opened with a question or an admission. That's craft, and you're allowed to study it (every writer you like learned from someone).

So here's my rule: learn how a post was built, then write your own and don't copy the words. A copied caption is easy to spot, and it won't work for you anyway.

Your people followed you, so sound like you.

What you need before you start

You need two accounts, and both are free.

1. Claude. You already have this, or you can make one at claude.ai.
2. Apify. It's a service that reads public web pages for you. You don't need a credit card to sign up, and it gives you $5 of usage every month.

Apify goes and gets the posts, and Claude reads them and tells you what's in them. They don't work together until you connect them, which is step 2.

Do the setup on a laptop. It takes about 10 minutes and it's the only part that needs one, because everything after it runs from your phone.

1. Make the Apify account

Go to apify.com and sign up for the free plan (no card needed).

You get $5 in credit every month. It resets on your billing date and doesn't roll over, so an unused month is just gone, which is fine because this workflow costs about $2 a month to run. I break down that math further down the page.

You don't have to poke around in there or learn what an Actor is. Just make the account and close the tab, because you'll do everything else through Claude.

The mistake that makes it fail: signing up and then trying to run the scraper yourself inside Apify. You can, but it's a lot of settings and you'll quit. Connect it to Claude instead and you get to ask for what you want in plain words.

2. Connect Apify to Claude

You do this once and never think about it again. Nothing else on this page works until it's done.

The exact path:

1. Go to claude.ai on a laptop.
2. Click your initials in the bottom left, then Settings, then Connectors.
3. Look for Apify in the connector directory first. If it's there, click it and skip to number 6.
4. If it isn't, click Add custom connector.
5. Paste this as the URL: https://mcp.apify.com
6. Claude sends you to Apify to sign in and approve the connection. Read that screen instead of clicking through it, then approve.

That's it, and now Claude can pull posts for you.

What plan you need: custom connectors work on the free plan, with one catch that's worth knowing before you start. Free accounts get 1 custom connector total, so if you're on free, Apify has to be the one you spend it on. Pro and above don't have that limit. On Team or Enterprise, an Owner has to add it for the organization before you can switch it on.

The mistake that makes it fail: typing the URL instead of pasting it. The address is mcp.apify.com (not www.apify.com), and one wrong character gives you an error with no useful explanation, so copy it.

3. Pick your 7 to 10 accounts

This step matters more than any other one on the page.

Pick 7 to 10 Instagram accounts in your niche and write the usernames in a note on your phone. You'll reuse this list every week, so it's worth 10 minutes of thought now.

What makes a good list:

• Your actual niche. If you teach AI to moms, pull accounts that teach AI to moms, because general business accounts with 2 million followers won't help you.
• Sized like you, or a step ahead. A post from an account with 800,000 followers does well partly because the account is already that big, and what works for them can do nothing for you at 2,000. Accounts one step ahead of you are the useful ones.
• People who post a lot. An account that posted twice in 2 weeks gives you almost nothing to read.
• Public. Private accounts won't return anything.

The mistake that makes it fail: picking 10 accounts you personally enjoy. This list is research, and some of the most useful accounts on my list are ones I'd never follow.

4. Pull the last 2 weeks

Open a new chat in Claude and ask for it in plain words, like this.

Use Apify to pull Instagram posts from these accounts:
[paste your 10 usernames]

Only posts from the last 14 days. Up to 20 posts each.

Claude asks permission before it goes, so you'll see it happen. Then it takes a minute or two and comes back with the posts, the captions, and the like and comment counts for each one.

Two weeks is the right window. A month is too much to hold in one conversation, and the older posts pull your answer toward what was happening 3 weeks ago, and one week is too thin to see a pattern.

The mistake that makes it fail: asking for everything. "Pull all their posts" gets slow and expensive, and it makes the next step worse instead of better. The setting that works is 14 days with a cap of 20 per account.

5. Ask what actually got people commenting

Look at comments instead of likes. A like is a reflex (someone double-taps going down the stairs), and it tells you nothing about whether they cared. A comment takes effort, because somebody had to stop and open the keyboard to type it.

Then there's a second step, and it's the part most people skip.

Compare every account against itself. Raw comment counts mean nothing across accounts of different sizes. A post with 200 comments on a 400,000 follower account might be one of their worst weeks, while a post with 30 comments on a 3,000 follower account might be the best thing they've made all year. If you just rank all 200 posts together, the big accounts fill every spot on the list and you learn nothing except who has the most followers.

So you work out each account's own average first, then look at what beat it.

For each account separately, work out its average
comments per post across these 14 days.

Then find every post that beat its own account's
average by 1.5x or more. Those are the winners. Don't
compare accounts to each other, only to themselves.

For each winner, tell me:
- the account, and how far above its average it landed
- what the post was about
- what the first line was doing
- how the caption was built: roughly how long, whether
  it asked a question, what it asked for at the end
- the format (reel, carousel, single image)

Then tell me what the winners have in common that the
below-average posts from those same accounts don't.

That last line is what makes the setup worth it. You're asking it to compare winners against losers from the same account, in the same 2 weeks, with the same audience, which is about as close to a fair test as you can run from the outside.

The answer is usually about how the post was written, like posts that admitted something, named a specific number or started with a question the reader was already asking.

Then push once more, because the first answer is usually the obvious one:

Now show me the 3 winners that don't fit that pattern.
What were they doing instead?

The outliers are usually where you find an idea the rest of your niche hasn't run yet.

One honest limit: comment counts come back for every post, so the averaging above works fine on the free plan. Reading the comments themselves is more limited, because on free usage you get roughly the 15 newest comments per post instead of the most-liked ones. That's enough to get the feel of a thread, but it isn't a full read of 400 comments, so don't build a conclusion on it.

The mistake that makes it fail: letting it rank all 200 posts in one pile, because the biggest accounts take every slot at the top and you end up with "post more reels." Make it work out the per-account average first, and if the answer still comes back that generic, tell it to redo the comparison inside each account one at a time.

6. Turn it into 5 reels in your voice

Now it gets to write, but first it needs to know how you sound. Skip this next line and everything it writes will sound like an ad.

Read the last 20 captions I posted so you know how I
sound. Then turn the strongest 5 topics into 5 reels,
using the hook shapes that worked above.

For each one give me: the hook, the text on screen,
and the caption.

If you can't point it at your own captions, paste 5 or 6 of them straight into the chat and you'll get the same effect. It needs to see how you write before it can write like you.

If what comes back still reads stiff, that's a voice problem and it has its own guide. Make Claude Sound More Human goes deeper on this than I can fit here, and it's the fix for reels that come out sounding like an ad.

What comes back is a draft, and you still film it your way. I usually keep the topic and the structure, rewrite the hook and throw out about one in five.

That's still the hard part done, because editing on a Sunday is a completely different task than staring at nothing.

The mistake that makes it fail: pointing it at captions that don't sound like you. If your last 20 are all one-line "link in bio" posts, that's what it's learning from. Pick 20 you'd be happy to have written and say so.

7. Put it on a schedule

Everything above works fine by hand, and this step just means you stop having to remember to do it.

Claude can run a saved prompt on a repeating schedule, and it runs on their computers instead of yours, so your laptop can be shut. You wake up Sunday and the work is already waiting for you.

In the Claude desktop app, click Scheduled in the left sidebar, then New task. Give it a name, paste the prompt, and set it to weekly. Scheduled tasks can use the connectors you've already switched on, so it reaches Apify the same way you do in a chat.

Every Sunday, pull the last 14 days of Instagram posts
from my 10 accounts.

Work out each account's average comments per post, then
find the posts that beat their own account's average by
1.5x or more. Tell me what those winners have in common
that the below-average posts don't.

Then write me 5 reels in my voice using those findings.

What plan you need: this one needs a paid Claude plan, because scheduled tasks are on Pro, Max, Team and Enterprise and not on free. If you're on free, everything else on this page still works and you just run it yourself on a Sunday, which takes the same 30 minutes you're spending now anyway.

The mistake that makes it fail: scheduling it before you've run it by hand a few times. The prompt needs tuning against your actual list first, so wait to automate it until it's giving you good answers.

What this actually costs

This costs less than people expect, and the math is small enough that I can just show you.

Apify charges for what it pulls. On the free plan the Instagram scraper runs $2.70 per 1,000 posts, and you get $5 of credit a month.

You can nearly double the list before you'd owe anything, and Claude's free plan covers the connector. The only thing you'd ever pay for is the Sunday scheduling in step 7, and that's a Claude subscription you might already have.

Keep an eye on it anyway. When the $5 is gone, Apify stops your runs instead of charging you, so the worst case is a scraper that won't start (and no surprise bill).

When it doesn't work

There are five things that usually go wrong, and all of them are quick to fix.

Claude says it can't find the tool. The connector is added but not switched on for that chat, so go back to Connectors and check the toggle. On free you only get 1 custom connector, and if you added something else earlier, that's your slot.

An account comes back empty. Usually it's private, and sometimes the username is wrong by one character. Check it by opening the profile on your phone, because if you can see the posts while logged out, so can Apify.

The run stops halfway. You're out of credit for the month, which you can check on your Apify dashboard. Either wait for the reset or cut your list down.

The reels sound like a robot. You skipped the "read my last 20 captions" line, or you pointed it at captions that aren't really your voice. This is the most common one by far, and it's a one-line fix.

The analysis is generic. If you get "Post consistently and use trending audio," that's what a first pass over too much data looks like. Narrow it to 14 days, and ask what separates the top from the bottom instead of what the top have in common.

The fastest path to a win

1. Make the Apify account, which takes 2 minutes and no card.

2. Connect it to Claude with that URL. You only do it once, on a laptop, and it takes 5 minutes.

3. Write your 10 usernames in a note, since this is the part worth thinking about.

4. Pull 14 days of posts and look at comments instead of likes.

5. Compare each account against its own average, because that one instruction is what makes the answer worth anything.

6. Point it at your own captions before you ask it to write a word.

None of this needs you to be techy, I promise. The hardest thing on this page is pasting a URL into a settings screen.

If you only do one thing: sort by comments instead of likes. You can do that by hand today with no accounts and no setup (just scrolling), and it'll change what you make next week.

Every prompt on this page

Copy these somewhere you'll find them, and swap the brackets for your real list.

1. Use Apify to pull Instagram posts from these accounts: [10 usernames]. Only posts from the last 14 days. Up to 20 posts each.
2. For each account separately, work out its average comments per post across these 14 days. Then find every post that beat its own account's average by 1.5x or more. Those are the winners. Don't compare accounts to each other, only to themselves. For each winner, tell me: the account and how far above its average it landed, what the post was about, what the first line was doing, how the caption was built (roughly how long, whether it asked a question, what it asked for at the end), and the format. Then tell me what the winners have in common that the below-average posts from those same accounts don't.
3. Now show me the 3 winners that don't fit that pattern. What were they doing instead?
4. Read the last 20 captions I posted so you know how I sound. Then turn the strongest 5 topics into 5 reels, using the hook shapes that worked above. For each one give me: the hook, the text on screen, and the caption.
5. Every Sunday, pull the last 14 days of Instagram posts from my 10 accounts. Sort by comment count. Take the top 15, tell me what they have in common, and write me 5 reels in my voice.

The blank page stops being the first thing you deal with on a Sunday, because when you show up there's already something to react to. That's the win.