Skip to content

The rulebook

The AI writes.The rules decide what ships.

Generation is the easy half. Evaluation is where the money is. These are enforced, not suggested.

Rule 01

Broad match is banned

Not off by default. Banned. There is no setting to turn it back on.

Rule 02

Phrase and Exact on every keyword

Every curated keyword ships as both. You do not choose, and it does not drift.

Rule 03

Neighbouring businesses blocked before launch

The categories that share your vocabulary and none of your customers are mapped and excluded during the build, not discovered in month three from a search terms report.

Rule 04

Career, DIY and definition searches never ship

Stripped during curation before you ever see them. You never get a padded list of searches that look commercial and never convert.

Rule 05

Nothing runs unapproved

No push, no proposal, no budget change reaches your account without a click. If the model fails, the system stops and tells you. It does not fall back to something uncurated and hope.

Rule 06

Every ad is graded before it ships

Thirty-one checks run on every headline and description, from mechanical things like length and casing through to the ones that actually matter: is this grounded in a real keyword, does it say something only your business could say, would a person read it twice. Fail any and it goes back to the model to be rewritten. Copy that cannot pass does not ship.

How the AI is built

Anyone can point an AI at your ad account.The hard part is teaching it what not to ship.

Connecting a language model to the Google Ads API is a weekend project. That is why there are forty of them. The difference is what happens between the model producing something and that thing reaching your money.

We built that layer first.

The AI writes everything. It just gets marked first.

It builds your ad groups, curates your keywords, writes every headline and description, generates your extensions and your landing pages. Then all of it is graded against thirty-one checks before it is allowed anywhere near your account. Copy that cannot pass goes back to be rewritten.

Generation is cheap. Judgement is the product.

Ask any model for fifteen headlines and you will get fifteen headlines. Ask it for fifteen headlines that are each grounded in a keyword a real customer typed, that no competitor could have written, and that survive a scoring pass, and you have a very different problem. That problem is what this platform is.

It runs on your account every day, inside a fence you can see.

It reads performance continuously and acts on waste automatically. What it is allowed to do on its own is deliberately narrow and completely visible, every action is logged with its reason, and anything it does by itself can be undone.

You approve what ships.

Every keyword, negative, ad and budget suggestion arrives as a card with a yes and a no on it. Until you click, it does not exist in Google Ads. Not because the AI cannot be trusted, but because it is your money and you should see the reasoning.

It watches the account constantly. It touches it only when you say.

Connect and review a build