Find formats that work

There is a version of "content strategy" that is really just taste: post what feels good, notice what gets likes, do more of that. It works for some people and it is impossible to learn from, because nobody can say afterwards why one post did well. This chapter is the other version. It is slower for the first two weeks and much faster after, because every post teaches you something you can write down.

Formats, hooks and ideas are three different things

Keep them separate in your head and in your log, because you can only test one at a time.

• An idea is what the post is about: "most freelancers undercharge for revisions".
• A format is the shape it takes: a numbered list, a before/after, a single-sentence opinion, a screenshot with one line of context, a 30-second talking-head video, a carousel.
• A hook is the first line, or the first second of video: the thing that decides whether anyone reads line two.

The same idea can be published as five formats with five hooks. That is not repetition; that is the experiment. And it is where the ideas list from chapter 1 stops being a constraint — twenty ideas times four formats is eighty posts.

Start with formats you can make on a bad day

Pick three formats for the first test cycle, and choose them by cost, not by what you think performs. A format you can produce in ten minutes on a day when everything else went wrong is a format you will still be using in month three. A format that needs a camera, good light and an edit is a format you will abandon at the first busy week, and an abandoned format has no results at all.

Reasonable first three for most founders who sell something:

• The mistake list. "Three things I got wrong about X, and what I do now." Numbered, plain text, no image needed. Works on X, Bluesky, LinkedIn, Threads and Mastodon with almost no editing.
• The build note. One screenshot of something you made or changed, with two lines on why. Works everywhere an image is accepted; on Instagram it needs the image to carry the meaning, since the caption is below the fold.
• The specific question. One line that asks your buyer something only they can answer. "What do you do when a client asks for a fourth round of revisions?" Cheap, and the replies are next month’s ideas.

The experiment log

A spreadsheet. One row per post per network. The columns matter more than the tool.

Fill it in at the same time every day, for the posts from two days earlier. The 48-hour window is a convention, not a rule: pick any window and keep it, because the comparison is what matters.

What a fair test looks like

The trap in week two is drawing a conclusion from one post per format. Networks are noisy: the same post published twice can get very different reach depending on the hour, the day, and who happened to be online. A few rules keep the log honest:

• At least three attempts per format before you compare formats. Five is better. This is why the plan in chapter 1 rotates formats over two weeks rather than testing one per day.
• Change one thing at a time. If you change the format and the posting time and add a link, you have learned nothing you can name.
• Post at the same time of day for the whole cycle. Timing might matter; you can test it later, on its own.
• Judge by the median attempt, not the best one. One outlier makes every format look like a winner once.
• Keep the loser. A format that reliably does average is worth more than a format that did brilliantly once and badly three times; you can schedule the reliable one for weeks.

How to read the numbers without fooling yourself

Every network reports what it wants to. Here is how to read the common ones in the light of what the networks themselves say.

• Impressions or views tell you the network showed the post; they say nothing about whether anyone wanted it. On a small account they can swing a lot for reasons unrelated to the post, such as who happened to be online.
• TikTok’s own explanation of the For You feed says a strong indicator of interest, such as watching a longer video to the end, gets more weight than a weak one, and lists shares among user interactions; Instagram names the likelihood of a reshare among the most important Reels predictions. If a network gives you watch-through or shares, it is reasonable to weight them above likes.
• Saves appear in Instagram’s own list of activity signals for Feed, Reels and Explore, and they mean something practical: someone intends to come back. A post that collects saves is worth treating as a template candidate even when its likes are ordinary.
• Replies are the one you should read by hand. Ten replies from strangers who match your buyer are worth more than a hundred from friends. The log has a column for this because no network will count it for you.
• Follower count is not the score. TikTok states it is not a direct ranking input, and on every network a follower who never buys is a vanity number. Watch it, do not optimise for it.

When a format wins: scale it, then vary it

Once a format has three or more attempts that do reliably better than the others on a network, three things happen in order.

1. It becomes the default. Two of every three posts on that network use it, with new ideas from the list.
2. The hooks that worked get a list of their own. Re-use them verbatim with new ideas — "Three things I got wrong about [new topic]" — because a hook is a promise about the post’s shape, and the audience that liked the shape once will read it again.
3. The third post in every three tries something new: a new format, or the winning format with a different media type. That keeps the log growing without risking the reliable slot.

And once a post has clearly outperformed, republish it. Not the next day; a few weeks later, to the same network, possibly with a sharper hook. Feeds are ranked per viewer, so there is no guarantee your followers saw it the first time — and republishing is easy to test: compare the second run with the first in the log.

When nothing wins

After a full cycle, sometimes every format sits at the same low level. That usually means one of three things, in order of likelihood: the posts are not for a specific enough person (go back to the one-sentence buyer); the hooks are all the same shape (try the specific-question format, which forces a different first line); or the network is wrong for this buyer (check the buyer-replies column across networks and move effort to wherever it is highest, even if the impressions are lowest). What it almost never means, this early, is that you should post more. Post differently first.