The editor that grades my AI's work

I don't trust a first draft from a person either. So I gave my AI an editor.

Why the writer can't grade itself

When the same tool writes and checks its own work, it mostly agrees with itself. Left alone, I'd be the editor for everything.

That doesn't scale for one marketer, and it wastes the part of me that's worth something: taste.

A reviewer that only finds problems

A separate reviewer whose only job is to find what's wrong. It never writes or fixes anything. It reads each piece with fresh eyes, as if it had never seen the draft, and scores it against a written bar for that kind of work.

Its default is suspicion. A review that finds nothing gets looked at twice.

If a piece falls short, it goes back to whoever made it with a specific list of fixes. That can happen up to three times. If it still isn't good enough, it comes to me with a plain explanation of why, instead of lowering the bar.

What it checks

• FactsEvery claim has to trace to something real.
• VoiceIt has to sound like mode40, or like the client it's written for.
• Hard rulesThe naming and writing rules I never want broken.
• ClarityPlain words over clever ones.

Every verdict is logged

I can see what the editor catches and what it lets through.

How my edits teach it

When I rewrite or reject a piece, the system keeps what I changed. The version I rejected becomes an example of what not to do. A version I clearly like becomes an example of what good looks like.

The scoring bar itself only changes when I review it. My taste adjusts what the team aims at, and I decide when the bar moves.

The principle

The one who writes a piece should never be the one who grades it. A reviewer whose only job is to find the flaw catches what the author can't see.