Why my posts now start with five bullets

The old pipeline: elegant, automated, and empty

I used to feed a topic into Make and get back a 1,200-word post that sounded like everyone else.

The automation itself was neat. A new row in a Google Sheet would trigger a Make scenario that did everything: draft the article, edit it, and post it to Ghost. No manual copy-paste, no formatting, no metadata. From a systems point of view, it was clean.

The problem was the input. I would drop in something like:

"What six months of weigh-ins taught me"

That was it. A topic. The model then had to invent the lessons, the voice, and the specifics. It did its job: the output was grammatically fine and structurally tidy. It had an intro, a few sections, and a conclusion that tied everything up.

It also had very little to do with what actually happened. The model filled the gaps with generic observations that could have been written by anyone who has ever stepped on a scale. It could not know about a 1.5 kg daily weight swing or any real pattern, because none of that was in the input. The post read like a product spec for “blog content,” not a record of something that actually occurred.

The new rule: no topic without five bullets

The fix turned out to be boring and effective: I do not trigger the automation unless I have written five bullets in a Notes cell first.

Those bullets are not polished. They are usually sloppy, sometimes half in Dutch, sometimes just numbers and fragments. A typical set might look like:

• 1.5 kg swing between morning/evening
• stopped panicking after “bad” weigh-ins once I saw the pattern
• Excel graph: weekly average vs daily noise
• one week travel = weight spike, back to normal in a few days
• biggest mistake: reacting to single data points

They are real. They contain the actual observation, the real mistake, the specific number. The AI cannot invent my 1.5 kg daily weight swing or my Tuur lineup error, because those are now in the input. Its job is to write from the bullets, not around them.

The scenario still runs the same way on the outside: Google Sheet row in, Ghost draft out. The difference is that the row now holds a topic and five bullets of evidence.

I wrote about this earlier in What My Automated Archive Exposed About My Writing.

From product to translation

The biggest change is not the prose. It is the feeling of ownership.

A post generated from a topic feels like a product. It has the shape of a blog post, but the content is generic. You can sense the template under the surface: broad intro, three lessons, optimistic wrap-up. The model is doing what it was asked to do: produce something that matches “a blog post about X.”

A post generated from five bullets feels like a translation. The bullets carry the fingerprint of my actual day: the frustration, the specific tool, the surprising outcome. The AI wraps them in sentences, but the insight is already there. It is turning notes into paragraphs, not ideas into filler.

That shift is easy to miss if you only look at word count or grammar. The paragraphs can look similar from a distance. The difference is that one post could have been written about anyone, and the other clearly could not.

This is also why “improve my writing style” prompts never fixed the old pipeline. The problem was not style. The problem was that there was nothing personal to style. You cannot edit specificity into a blank.

The five-bullet filter: thought vs post

The five-bullet rule became a filter against thin ideas.

If I cannot write five concrete bullets about a topic, I do not have a post. I have a thought. A thought can live as a tweet, a line in a logbook, or a future seed. It does not deserve a full article and a publish date.

This filter has killed more automation triggers than any editorial review. I open the sheet, type a topic, try to write the bullets, and sometimes stall at two vague lines. That is the signal. The automation does not run on speculation; it runs on evidence. The bullets are the evidence.

The topics that survive are easier to write and more fun to read. If you can list five specific things that actually happened, you already have structure. Each bullet is a section. The post becomes “here are five real things that occurred and what I learned,” not “here is my opinion about a concept.”

More on this in my article GEO vs SEO in 2026: What Actually Moved For My Clients.

Two minutes in, twenty minutes saved

Writing five bullets takes longer than typing a topic. It costs a couple of minutes instead of a few seconds. That looks like friction in an automated setup that is supposed to be fast.

Those minutes remove an entire revision cycle later.

With the old pipeline, I would get back a generic post and then spend time rewriting it to inject personal details. I would replace vague claims with real numbers, swap invented mistakes for actual ones, and remove sentences that sounded like a content mill.

Now, the personal details are the foundation. The AI cannot smooth them away because they are structurally load-bearing. If the third bullet is “Tuur batting 3rd was my lineup error,” the generated section has to talk about that, not a fictional coaching mistake. The model is constrained by the notes, which is the point.

So the trade-off is simple:

• Old way: almost no time to type a topic, lots of time to fix a generic post.
• New way: a bit more time to write bullets, almost no fixing.

In an automated pipeline, the cheapest editorial investment is at the start. Once the machine is running, it will happily produce polished nonsense at scale if you let it.

Automation is not the enemy, vague input is

I did not abandon the automation. I fixed its input.

Read also Building A Personal Longevity Dashboard With bypass-longevity.

The system still handles the parts I do not want to spend time on: formatting, SEO meta, Ghost upload. It still generates the slug, the excerpt, the featured image placeholder, and the scheduled publish time. Those are mechanical tasks that do not require judgment.

What changed is the boundary between my job and the system’s job.

• My job: capture five real bullets with concrete evidence.
• System’s job: turn those bullets into a readable post and publish it.

That boundary keeps the automation in its lane. It is not allowed to invent my experience. It is only allowed to express it.

If you are running something similar and your posts feel oddly generic, the problem may not be the model, the prompts, or the tools. It might just be that you are asking the system to do the part that only you can do: noticing what actually happened and writing it down, even if it is in five ugly bullets.