For whoever builds it.

Everything below is the parts: the folders, the tools by role, the steps and what it costs to run.

Ready now. 16 October 2026

Build order: One campaign firstTurn on Google's new setting for one steady campaign first. Test it on half the traffic and leave the other half as it was.
Time to first result: UnknownNot measured yet
What you need from the owner: The off-limits listBefore you start, the owner writes down the pages and words Google may never use, whether ads show for the store's own name, and each product's margin.
Monthly cost: Ad spend plusAd spend, the assistant, a weekly read
What breaks first: PagesWith page picking on and no off-limits list, paid clicks land on pages the store would never choose.
Folders: 9paid-search/ in the middle, eight around it

1Build it

One prompt. Paste it into your AI and it draws your business on this chart and makes the nine folders the system runs from, with your rules written in. About 15 minutes. Nothing sends to anyone.

What to do

1Hit the button. It copies the instructions.
2Open Claude Code in the folder where this system should live. Paste, and send.
3Answer its questions, one at a time. When it shows your chart, say go and it builds the first square.

Written from this chart's research. Nothing to edit.

Pro adds the finished folders for this chart in Claude Code: every step, every prompt, an example setup. See the plans.

2The nine folders

One folder for the owner in the middle, one workspace per part beside it. The Google Ads account and the product feed hold the state; the folders read them and change nothing without the owner's yes.

paid-search/

The owner's folder: what the store will and won't bid on, the pages it allows and forbids, the words its ads may never carry, its brand lists, its margins and conversion values, and every change dated with why.

You can see what this folder is for. The full folders for this chart are not published yet.

3What it runs on

Tools by role. Prices are public list prices read in October 2026, as examples, not recommendations.

Reaches Search ads on Google (Search, AI Max for Search, Performance Max) and the store's own landing pages. A landing-page builder from $31 a month billed annually ($39 monthly).

Thinks An AI session working from the store's domain folder, beside Google's own matching and text writing. The model is not priced in this chart.

Runs The product feed sync, inventory-aware serving, Smart Bidding, conversion tagging and the weekly pull (Google Ads scripts or the API; Google Ads Editor for bulk changes). A feed tool from $64 a month; server-side sales tracking from $225 a month.

Knows The domain folder; the shared brand lists; the negative keyword and URL exclusion lists in the account. Not priced in this chart.

Feeds The search terms and landing pages reports, asset reports, conversion data with value, the product feed, stock and price. A PPC reporting tool from about $209 a month billed annually ($299 monthly).

What it costs to run

4Build order

Start here: Try Google's new setting on one steady campaign as a 50/50 experiment. Read the first weekly note. Change nothing on less than two weeks of data.
Then: Set up the eight workspaces beside that folder; each drafts and proposes, and changes nothing in the account without the owner's yes.

Open the folders

5What it costs

The ad spend, the usage fees of the AI assistant that does the reading and drafting (not priced here), and the owner's weekly time. Paid parts a store may add, at public prices: a catalog feed tool from $64 a month, sales tracking from $225 a month, a reporting tool from about $209 a month billed annually.

Itemize it each month: ad spend, the model, and any feed, tracking or reporting tool. The public prices are under What it costs to run.

See the sources

6The five rules

Where the work lives

The rules live in one folder the owner keeps: what the store will and won't pay for, the pages and words that are off-limits, whether ads show for the store's own name, the margins, and every change made, dated, with why. Each of the eight parts reads from it, so next week's proposals carry what this week's note found.

Nine folders handle outreach. A whole business needs them to work together.

Your AI does better work when it reads the right instructions before it acts. These are real folders on a computer. Each holds plain text files of instructions for one kind of work, which your AI reads before it starts.

These nine handle outreach. A business has many areas: sales, delivery, finance, content, the owner's own week. Each needs its own folders, and they have to work together. When they don't, your AI reads the wrong instructions, two folders say different things about the same fact, and every new conversation with your AI starts from zero.

Five rules keep them working together:

One file read first. A single file tells your AI which folder handles which kind of request.
One home for each fact. A rule lives in one folder. Every other folder reads it there, and only that folder changes it.
Hand work over, don't reach in. Work meant for another folder is dropped into a folder inside it called inbox, and the work is done there, by that folder's rules.
Write down where things stand. Each folder keeps a file of its decisions and progress, which your AI updates as it works, so the next conversation starts where the last one stopped.
Nothing runs unseen. Anything set to run on its own, like a scheduled email or a nightly update, is listed with when it runs and how to stop it.

Large companies already run this way: one home for each fact, written rules read before the work starts, a record of every change. AI can read those rules and work from them.

When Expert opens, it teaches you to build this for your business: a full course on setting up the folders and rules, from your first folder to all of them working together, Nick's group coaching while you build, and a community of people building theirs. It includes everything in Pro.

7Sources

For the owner: decide Previous Next All charts