Our name is a rule. Everything we make with AI gets checked before it goes out, by a person looking at the real output, not at the plan. Here is the checklist, in the order we run it.
1. Facts against the source
Prices, hours, phone numbers, service names and addresses are checked against the business's own booking system or website, every time. AI tools are confident about details they have never seen. A reel that quotes last year's price is worse than no reel.
2. Nothing invented
No made-up clients, reviews, testimonials, before-and-afters or results. No "customers saved 40%" unless we can show the receipts. If a piece needs a person, it uses a real photo of the real person, with permission. If we use an illustration, it is labelled as one.
3. Disclosures where the law and the platforms want them
If a post earns a commission, it says so on screen and in the caption, in plain words, before the link. Regulated topics such as debt relief and credit repair get their partner's written approval before anything runs, and no earnings or outcome claims, ever.
4. Look at the frames
For video, we export a contact sheet of frames from every scene and look at it. This is where we catch overlapping text, a caption cut off at the edge, a misspelled word, or the wrong photo. On one recent build this step caught that a client's website was still using a stand-in portrait instead of a real photo of the stylist. Real photos replaced it the same night.
5. Listen to the levels
Voice should sit clearly above the music without shouting. We measure loudness instead of guessing: around -18 LUFS for the voice, music ducked underneath it. Our first attempt at -15 sounded like the narrator was yelling.
6. The ending matches everywhere
The call to action says the same thing in the narration, on the end card and in the caption: who to follow, for what, and how to book.
7. Watch it run
An automation is not finished when it is built. It is finished when it has run on its own and we have checked the result: the post went out, the email arrived, the row was marked done. Every workflow also gets a failure alert, so a broken login does not hide for a week.
8. Write down what changed
When a client corrects something, the correction goes into the recipe for that brand so it never comes back. That is how "the voice sounds like he's yelling" becomes a number in a file instead of a mistake we repeat.
None of this is glamorous. It is the difference between AI content that quietly costs a business trust and AI content that earns it.