Let AI check AI, but keep the final call
The minimum sequence: one AI writes the brief, another checks it against the facts, and you approve it.
Almost every step in our marketing involves AI, and it saves so much time that most of the marketing engine can be run by a single person. The one problem I keep running into is that AI makes very confident mistakes. That’s why everything it produces has to be reviewed before it’s published anywhere.
Three roles, one rule
Our content production setup uses three AI roles where the first acts as a strategist and recommends tasks based on its data, to the second AI, which carries out the production of the content or the actions. That AI hands the result to me for review. If it’s OK, I pass it to the third AI. But before that, I normally rewrite a large part of the content to adjust the tone to the situation, since tone is something AI handles poorly in my experience.
The third AI is told explicitly not to just obey and its job is to check every statement against the facts, question it, and say so before anything changes. It then carries out the finished product, or comes back to me with proposed adjustments. Sometimes when the AIs disagree, I make the call, and more than once my call has overruled one of them.
It doesn’t always look like this, but for anything a little more sensitive, this is normally how we work. And even then, small faults can slip through. That’s why I strongly question companies that rely entirely on one AI agent to handle everything by itself.
The minimum sequence
In content production, the minimum sequence should be: one AI writes the brief, another checks it against the facts, and you approve it.
Content creation has turned very cheap with AI and the work has therefore turned to checking it, until I can trust it enough.
Step 09 in AI search notes: Check every AI draft against its source