Ad images and videos made by AI from one written brief, sized for every ad placement. AI makes the images and videos, a person chooses them, and they are uploaded to your ad account switched off until you approve.
A short description of the ads to make. It is drafted from last month’s customer messages, your reviews and your price list, so the selling points come from what customers actually said.
The brief
Roofers, Phoenix. The missed call is the job. Your own number, as iMessage. Four angles: speed, trust, money, relief.
SpeedTrustMoneyRelief
What it becomes
36 imagesfrom one brief
9:16 · 4:5 · 1:1every placement
Directedby a person
Pauseduntil you say go
from: last month's threads · 3 reviews · the price list
Last month’s customer messages
Searchmissed call
2:18 PM · missed call from a Meta ad
Hi, thanks for calling ABC Home Services. Sorry we missed you, we're on a job. What do you need?
AC stopped blowing cold. How soon can someone come?
An illustration. The business and the customer are made up.
Three reviews
The price list
The five stages
From one brief to paused ads, in five stages.
01
The selling points
Four selling points (angles): speed, trust, money, relief.
Each selling point is made in three sizes, nine images at a time. They come from the brief, not from stock photos.
Examples
1Last month’s message conversations are searched for the words customers use, and those words become the selling points (the angles).
2The price list is read so that no ad promises a price you did not approve.
3One brief produces a month of ad images and videos. To change them, you edit the brief, not thirty-six files.
1–9Speed
10–18Trust
19–27Money
28–36Relief
02
Images
AI image models make the images.
Four selling points in three sizes: thirty-six images from one brief. Every file is stored and named.
Examples
1One product photo becomes a month of still-image ads in an afternoon.
2Each image is stored with its selling point and size, so the best performer can be made again in new versions.
3Nothing is uploaded to your ad account until a person picks.
36images · from one brief
Nine at a time
SpeedTrustMoneyRelief
9:164:51:1
03
Video
AI video models make the videos, in a size for every place the ad appears.
AI video models make videos from the best-performing selling point. A person directs them. The videos are either a person speaking to camera or moving graphics, made with AI video models.
Examples
1The best-performing still image becomes a vertical 9:16 video for Instagram Stories and TikTok.
2The script uses the selling point that is already getting results, not a guess.
3You can test thirty selling points a month instead of three.
BBright RoofingSponsoredEvery missed call, answered.Learn more
9:16Story · TikTok
An illustration. The business is made up.
04
Sizes
Each place an ad appears gets its own size.
Vertical 9:16, portrait 4:5 and square 1:1. The resizing runs on a Mac, and each run is written to the Ledger (the system’s record of every request and response) like every other action.
Examples
1One original video becomes every size without anyone opening an editing app.
2Captions are added into the video for people who watch without sound.
3When a different ad starts performing best, its new sizes are made in minutes.
9:16One original
9:16Story · TikTok
4:5Feed
1:1Everything else
05
Uploaded, paused
The chosen ads are uploaded to your ad account, paused.
Thirty-six files and twelve ads are uploaded to your Meta ad account, switched off (paused). No money is spent until you tell it to start.
Examples
1Uploading, naming and attaching each file is one action in the Directory (the table of every action the system can run).
2Ads are always created paused, so none can start running by accident.
3You text “Go” from your phone and they start running.
Meta ad account36 files · 12 ads
you text, from your phoneGo
The permissions on one action
Make 12 ad variants from the brief
limit: 36 renders per run
write: Meta, this account
limit: paused until you say go
uploaded, paused
Every action the system can run is a row in the Directory. Each one runs through dispatch, which checks limits like these before it runs, and each run is written to the Ledger.