Insights

/

16 Nov 2026 · 3 min read

Note

AI content creation gets better when fixes become rules

We taught an AI to make our content. It wasn't right on day one. It got better because every fix became a written rule it reads next time.

AI content creation gets better when fixes become rules

One new card in the box. The next post reads it first.

Most advice on AI content creation is about the perfect prompt. Find the right words, and good posts come out. We tried that. The first posts were fine. But each one needed the same fixes by hand, and the same misses came back the next week.

The fix was not a better prompt. It was what we did after each post that missed. We wrote the fix down, so the AI would read it next time. Our short video shows the whole thing from start to end.

Do it by hand first

We didn’t start with a playbook. We made one week of posts together, one step at a time. Every choice got said out loud. Why this idea. Why this first line. Why this picture.

That week was slow. It was also the only way to find out what good means for us. You can’t write down a standard you haven’t seen yet.

Three steps turn a week like that into a playbook:

01Make it by hand, and say each choice out loud
02Write down what worked, one short playbook per job
03Correct it, and turn each fix into a new rule

Write down what worked

The steps from that week became short playbooks. Each one does one job. One finds ideas. One writes captions. One makes the short videos. One writes the weekly email. One puts posts on the calendar.

The AI reads the right playbook before it starts, the way a new hire reads the job sheet before the first call. So each job starts from the same steps.

Every correction becomes a rule

This is the part that matters most. When a post looks wrong, don’t just fix that post. Fix the playbook. Then the next one starts from the corrected version.

Three real notes from October, and the rule each one became:

  • Reels stopped on the problem. People were left hanging. Now every reel ends on the fix and one clear next step.
  • The voice changed from line to line. We compared the settings by ear and picked one. Every clip uses it now.
  • The text was too small for a phone. Now headlines are sized for a phone, on every video.

You don’t write the perfect playbook on day one. You correct it until it sounds like you.

None of these was a big change. Fixed in one post only, the same miss can come back next week. Fixed in the playbook, the next post starts from the fix. That’s the difference between editing AI content and teaching it.

Does AI content still need a person?

Yes. A person still says OK twice a week. Once to the plan, with every day, every post and every first line on it. Once to the finished posts, all on one page, before anything goes out. When a choice is a matter of taste, there are a few versions to pick from.

That check is also where the next rule comes from. Every “not like that” becomes a new line in a playbook. It’s the same idea as not typing everything twice, only for the way you sound.

In Income Mavericks, the finished posts can then go out on their own, each at its own time.

The fix is ordinary. Make it once by hand. Write down what worked. Then treat every bad post as a rule you haven’t written yet.

More on this topic

AI

30 Sep 2026 · 4 min read · 4 min

Explainer

A WhatsApp AI chatbot should end in a booking

A WhatsApp AI chatbot that only answers questions leaves the job where it found it. For a service business, the chat has to end in a booked slot, a record and a named person.

→

AI

17 Sep 2026 · 4 min read · 4 min

Explainer

An AI employee answers. A person still decides.

An AI employee handles the first minute of a conversation. It books, answers and logs. What it cannot do is judge an exception, and that is worth knowing before you buy one.

→

Rather talk than read

Bring us the process that keeps going wrong

Tell us what is breaking and we will tell you what it would take to stop it.

Get in touch