How I publish every day without writing every day

I publish content nearly every day. On more than one network and in more than one format.

I don't do it because I have time to spare. I do it because I built a process that doesn't depend on me having a good idea that day.

Here's the whole process.

The real problem with publishing daily

Almost nobody stops publishing because they don't feel like it. They stop because they run out of material.

On Monday you have three ideas. By Thursday you've used all three. On Friday you write something you don't like and post it anyway. Two weeks of that and you quit.

What's missing isn't consistency. It's somewhere to pull ideas from that isn't you at eight in the morning.

Where the ideas come from

Mine come from three places, and none of them is inspiration.

The work. Every week I work with a client, something turns up that I didn't know. A new objection. A different way of explaining the same thing. A number. I write it down the same day, raw, without thinking about how I'd publish it.

A field document. I keep a file with what I see in real companies: what worries them, how they say it, in which words. That document is what stops me writing like a brochure. When a line of mine sounds like a consultant, it means it didn't come from there.

What gets published in the sector. Regulation, launches, data. I use AI heavily here, and I explain how below.

Those three sources give me a bank of ideas. Publishing stops being invention and becomes choosing.

How I use AI to gather

This is what people ask me about most and where I got it most wrong at the start.

At first I asked AI to write posts for me. What came out was correct and empty. It sounded like everyone else.

Now I use it before writing, not to write. Like this:

• Finding and summarising what has happened. I ask it to search what's been published on a specific topic in a specific period. It gives it back with the source next to it. No source, no use.
• Checking what I'm about to claim. If I'm going to quote a number, I verify it. And it goes in whole: if a fine has bands, you name the right band or you don't name the figure.
• Getting the idea out of my head. Sometimes I talk to it for five minutes and ask it to give me back the three ideas buried in there. That helps me far more than asking for a post.

The rule I use: AI searches, checks and sorts. The idea and the final words are mine. Opinion on its own is noise. Opinion with a verifiable fact behind it is another thing entirely.

How each piece gets written

I have a style document. I wrote it after reading thirty of my own posts back to back and seeing that half of them didn't sound like me.

It has two levels.

Level 1: make it clear. Short sentences. One idea per sentence. Plain language. If a word can be swapped for a simpler one, it gets swapped. The reader shouldn't have to reread anything.

Level 2: make it believable. This one came later and it lifted the quality most. Three rules inside it:

• Drop the absolutes. "Nobody", "always", "everyone". They sound strong and get knocked down in one comment. Swapping them for something hard to argue with sounds the same and holds up.
• Three levels, not two. Almost everyone does belief and twist. The third is missing: why it happens. "You don't have time" followed by "what you don't have is something that moves without you watching it" is fine. What finishes it off is explaining the mechanism: every decision comes back to your desk because it never fully left it.
• Scenes, not concepts. "Having a good year isn't having a system" gets forgotten. "If May was a record and September depends on ringing the same person again to ask how she did it" gets remembered.

I give that document to the AI every time. Not so it writes for me, but so the rewrite goes in the right direction.

The checks

This is where the system stops being a nice idea.

Every batch of pieces goes through automatic checks before it gets published:

• No sentence over 25 words. A long sentence is nearly always two ideas stuck together.
• Average under 16 words per sentence.
• A blacklist of words. "Synergy", "ecosystem", "paradoxically" and about twenty more. Every time a new one slips through, I add it.
• Different rhythms within the same batch. Ten posts in a row with the same structure show.
• Overlap between networks. If the version for one network and the version for another share six words in a row, one gets rewritten.

That last check turned up something I wasn't expecting: simplifying the language increases the overlap. Using more common words makes the two versions converge. So the overlap check always goes last, after the style passes. One batch took seven rounds.

The videos

I produce the videos with Hyperframes.

The reason is the calendar, not laziness. Filming on camera means I need the place, the light and the free hour. When any of those three fails, there's no video that day. And a hole in the calendar costs more than the difference in quality.

What doesn't change is the script. The text goes through the same process as a post: idea, structure, checks. A well produced video with a weak script is still a weak video.

The log

I keep a document per week with what got published: what went out, on which network, on which day, which idea it used.

It's good for three things. Not repeating an idea ten days later. Seeing which topics are exhausted. And knowing what's left to say about an idea that worked.

It's the least glamorous part of the system and the part that holds it up.

What actually makes it work

It isn't the tool. It's having one idea you always come back to.

Mine is this: companies don't have an artificial intelligence problem, they have a memory problem. Meetings, documentation, holidays, onboarding, processes: all of those are forms of memory that fail.

Without an idea like that, you publish scattered observations anyone could have signed. With it, each piece adds to the last one.

The system is what lets you publish on a bad day. The idea is what makes it worth publishing.