How to build a content engine that tells you what to post

Short answer: list the competitors worth watching, have AI read every post they published last week, run the same analysis over your own posts, compare the two to find what wins for you specifically, then generate next week's ideas from that plus your real client conversations. The point is that the ideas come from evidence instead of from whatever occurs to you on the day.

Here is the whole thing, step by step.

Why guessing fails

Most content calendars are a list of things someone thought of in a meeting. There is no feedback loop, so you repeat whatever felt good rather than whatever worked.

The fix is not more creativity. It is a system that looks at what already performed, for you and for people like you, and tells you what to make next.

How to build a content engine, in six steps

Six steps, and the first one is the only one you do by hand.

1. List the competitors worth watching

Put them in one place with a tick box next to each. Mine is a Notion database. Ticking a name adds them to the next run and unticking removes them, which means the list can change week to week without touching anything else.

Pick people slightly ahead of you in the same niche, not the biggest accounts in your industry. You want a set you could realistically match.

2. Read every post they published last week

For each account, pull the last seven days: the caption, the link, the transcript, the opening line, and the engagement numbers. Transcripts matter more than captions, because the hook is spoken in the first three seconds and is often nowhere in the caption.

3. Run the same analysis over your own posts

This is the step most people skip, and it is the one that makes the rest useful. The same numbers, the same hook breakdown, applied to your own account.

Without it you have a report about other people.

4. Compare on the right measure

Raw view counts tell you who has the bigger audience, which you already knew. Compare each post to the median post of the account that published it. A post doing three times its own account's normal is a signal. A big account's average day is not.

Do the same to your own posts, and you find out which of your formats actually work rather than which got lucky.

5. Add the things only you know

Your client calls and your inbox are full of the questions people actually ask you. Feed those in. A theme that two clients raised this week is worth more than a competitor's view count, because those are people who already pay you.

6. Have it write the ideas, with the evidence attached

The output is not a list of topics. Each idea should carry the hook, the talking points, the call to action, and why it is being suggested: whose post, which number, which client asked.

If an idea cannot explain itself, it is a guess with extra steps.

How long does it take to set up?

An afternoon for a rough version that runs, and then a few weeks of correcting it. Mine had two faults that made it useless for months and produced a tidy report the whole time, which is covered in what happens when an automation quietly returns nothing.

What do you need to connect?

A way to read social data, which for most people means Apify, covered in connecting Apify to Claude. After that it is whatever you already use to store things.

Use Claude Code. This one calls a scraping service, keeps a week of results, and runs on a schedule without you. That is Code's job. You can still read the output wherever you like.

You need a Claude account with Code, an Apify account for the social data, somewhere to keep the competitor list and the output, and a list of four to eight accounts worth watching.

1. Write your competitor list by hand, with a tick box, so it can change week to week without touching anything else.
2. Connect Apify so Claude can read real numbers instead of guessing at them.
3. Pull one week of their posts manually and read it yourself before you automate anything.
4. Add your own account to the same analysis, because without it you have a report about other people.
5. Feed in your client calls and your inbox, which is the part nobody else can copy.
6. Set it to run weekly and have it write ideas with the evidence attached.

Build me a weekly content engine. Run it every [DAY] at [TIME] and write the result to [WHERE YOU WANT IT].

About me:
- I do [WHAT YOU DO] for [WHO YOU SERVE]
- I post mainly on [PLATFORMS]
- Accounts worth watching: [4 TO 8 HANDLES, SLIGHTLY AHEAD OF ME, SAME NICHE]

Each week:
1. Pull the last 7 days of posts from those accounts: caption, transcript, opening line and engagement numbers
2. Do the same for my own account
3. Rank every post against the MEDIAN post of the account that published it, not by raw views, so a big account's ordinary day does not look like a win
4. Pull the themes out of my client calls and emails from the same week
5. Give me 10 ideas for next week. For each one: the hook word for word, three talking points, the call to action, and WHY you are suggesting it, naming the post, the number or the client question behind it

Rules:
- An idea that cannot explain itself does not go in the list
- Never name a client
- Tell me plainly if a week of data looks wrong or empty rather than producing a tidy report anyway

Ask me for anything you still need. Then build steps 1 and 2 only and show me the output before you build the rest.

Expect an afternoon for something that runs, then a few weeks of correcting it. That last rule matters more than it looks, because an engine that quietly returns nothing still produces a report that looks fine.

Where to start this week

Do step one by hand. Write down four accounts and look at their last seven posts yourself. Note the opening line of each and what it promised.

You will learn more in twenty minutes than from another month of posting on instinct, and you will know whether this is worth automating before you build anything.

If you want help building it around your business, have a look at how I work.