I was reading my own brand brief out loud, the way I tell clients to, and I stopped on one line.
“Listing alert automation, built in Make.com, for a realtor.”
I sat there a second. Because I never wired that automation. It was planned. It never got built.
A quick backstage tour
Before the rest of this makes sense, here's how my setup works. I run my one-woman business with AI doing a lot of the writing. I keep files that describe my business. Who I serve, what I offer, what I've done. My AI reads those files before it writes anything, so it sounds like me and knows my story.
And because I don't trust “sounds good” as a standard, I built a grader. The grader is another AI whose only job is to score every draft against my writing rules, 1 to 10, like a teacher with a red pen. Nothing gets published under an 8. I built it because I'm too close to my own writing to judge it.
How a plan becomes a fact
Now. That line about the realtor automation. I did not write it as a lie. Months back it was a plan. I wrote it down the way you write down things you mean to do. And once it was in the file, it was quietly true forever, the way things in your own files are. Like writing “learn Spanish” on a to-do list and finding “fluent in Spanish” on your resume a year later. Nobody lied. Nobody checked.
Then my AI got hold of it.
It spread into nine documents. Two of them I only found after I thought I was done. It became a pre-written subject line, queued and ready to send. It became the reason a strategy doc told me to build for realtors first, because I “already had a case study.”
And here's the part that got me. The grader, the teacher with the red pen, never flagged it. It couldn't. That line was sitting in the file the grader reads to learn what good proof looks like. I had accidentally written the lie onto the answer key.
Two different jobs, one automated
The grader was never built to catch this. It checks whether writing is good. It cannot check whether writing is true. Spell-check works the same way. It catches “definately” every single time. Type “I definitely won an Olympic gold medal” and it says: looks great.
The one question
So here's your move, and it takes ten minutes. Open the file your AI reads about you. The brand doc, the About page, the prompt you paste every time.
Read it out loud. At every specific claim, be the kid who answers everything with “prove it”: can I point to it?
If you can point to it, keep it. If you cannot, cut it. Do not soften it. Cut it.
I cut the line. What is left is shorter, and it is true.
The machine writes fine. The job now is being the last honest reader in your own system. Nobody else in there is checking.
Janet V.