Ad copy variants per product and platform, scored and kept in a library.
Describe the product once, pick the platform, and review five candidates with a headline, primary text and call to action.
What you get
- Five variants per request by default. You can ask for anything from 1 to 10.
- Each variant has a headline, primary text, a call to action and a quality score from 0.0 to 1.0.
- Copy follows the format of the platform you pick: Meta, Google or TikTok. A batch request writes for all three at once.
- Languages: English, Portuguese (Brazil), Spanish (Latin America), French, German and Italian.
- The library sorts by score, then by date, and you can filter by platform.
What it does not do
- It does not publish. Nothing goes to an ad account unless you use it yourself.
- It does not promise results. The score is an estimate made by the model.
- It does not read your competitors or your past results. It works from what you describe.
Usage is metered
Every AI call is recorded in a usage ledger, so you can see spend per customer account.
Questions
Which model writes the copy?
An Anthropic Claude model. Your product description and settings are sent to it to generate the variants, as the privacy page states.
Does the score predict performance?
No. It is the model's own estimate of quality on a 0.0 to 1.0 scale. It sorts the library, nothing more.
Is anything published to an ad account?
No. Variants are stored in a library for you to review and use. Nothing reaches a live campaign without you.