An AI content calendar workflow in Notion

Why most AI planning falls flat

Ask an assistant for “30 content ideas for a skincare brand” and you get thirty ideas that would suit any skincare brand. Nothing in the answer knows that your carousels outperform your Reels, that Tuesday posts quietly die, or that the one behind-the-scenes post you made in August got three times your usual saves.

That knowledge exists. It is usually scattered: the plan in Notion, the results in an analytics tab, and a rough memory of what worked. The workflow below puts it in one place first and only then brings the AI in.

An AI plans as well as the data it can see. Give it your results, not your niche.

Before you start: one database that holds plan and results

You need a content calendar where each post is one row: a title, a date, the caption, the media, a status, and — this is the part most calendars are missing — the results once the post is live. If you are starting from scratch, the content HQ guide covers the columns worth having.

With Ninecast connected, the results columns fill themselves. Each published post gets its link written back, and every night its Instagram likes, comments, shares, saves, reach and views are written into the same row. After a month, your calendar is also your report, and that is exactly what an AI needs to read.

Step 1: Gather the last 30 to 60 days

Filter your database to posts published in the last 30 to 60 days. In Notion, Notion AI can read the database where it sits; with Claude, connect Notion and point it at the database; with anything else, export the view as CSV (Notion and Airtable both offer it) and attach the file.

Here are my posts from the last 60 days, one per row, with the caption, post type, date and results (likes, comments, shares, saves, reach, views).

Before anything else, tell me how many posts there are and which columns you can see, so I know you are reading the right data.

That second paragraph matters. Asking the AI to describe the data back to you catches a missing column or a half-exported file before it builds a plan on top of it.

Step 2: Ask what worked, and why

Reach on its own rewards whatever the platform happened to push. Saves and shares are the better signal: they are what people do when a post was worth keeping or worth sending. Ask the AI to rank by those.

Rank these posts by saves plus shares, relative to reach.

For the top five and the bottom five, what do they have in common? Look at post type, topic, the first line of the caption, length, and day of the week. Only point out patterns that appear in at least three posts.

The “at least three posts” rule keeps the AI honest. One great post is a moment; three is a pattern you can plan around.

Step 3: Turn the patterns into next month’s angles

Now the AI is useful for ideas, because it is working from your evidence rather than from the internet’s average.

Using only the patterns you found, suggest 12 post ideas for next month.

For each one give: a working title, the post type, the pattern it builds on, and one sentence on why it should work for this audience. Mix in two ideas that test something we have not tried yet, and label them as tests.

Paste the ones you like into your database as new rows with the status set to Idea. Notion AI can add the rows for you; otherwise it is a quick copy and paste.

Step 4: Write the captions yourself, then let AI tighten them

This is the step to keep human. Write a rough first draft of each caption in the row, in your own words, however messy. Then use AI as an editor rather than an author.

Edit this caption without changing my voice or adding anything I did not say.

Make the first line stop the scroll, cut anything repeated, and keep it under 150 words. Show me the edit and list what you changed.

Asking for a list of changes means you approve each one rather than accepting a rewrite wholesale. Over a few weeks the edits get smaller, because the AI is learning what you keep.

Step 5: Check the grid before anything is scheduled

A month of good individual posts can still make a messy profile: three dark images in a row, two carousels side by side, a launch buried between quotes. Ninecast shows the planned posts as a real Instagram grid inside the same Notion page or Airtable base, built from your rows, so you can see the month the way a new visitor to your profile will.

Drag a post onto another and they trade dates, written straight back into your database. This is the one check AI cannot do for you, because it is a judgement about how your feed looks.

Step 6: Give each row a date and let it post

Set the date on each finished row. When the date arrives, Ninecast publishes it to every platform you connect — or only the ones named in the row’s Platforms column — and writes the post’s link back. The results start coming in that night, and next month’s Step 1 is already half done.

Plan from last month’s results, post, and let this month’s results land in the same rows. That loop is the whole workflow.

Where AI helps, and where it should not

• Helps: reading a month of results in seconds, spotting patterns across dozens of posts, turning patterns into angles, and tightening a caption you wrote.
• Should not: decide what your brand stands for, write in a voice it borrowed, or judge how your grid feels. Those are the parts your audience follows you for.

Setting up the posting side takes about ten minutes with the setup guide. For the details of each half, see a Notion content calendar with analytics and scheduling posts from Notion, or from Airtable.