What is a model here, and what is not.

One thing: it plans a product’s week and writes the drafts. That is a single request to a Claude model from Anthropic, given the product’s name, its facts, its rules and a short summary of how earlier pieces performed.

The model also writes a reason for each piece it plans. You can read those reasons in the queue.

We call the parts of Campaign Fleet agents because each has one job and writes down what it did. Most of them are not AI models.

Reading your site and proposing facts is fixed code. So is checking a draft for figures, superlatives and promises that no confirmed fact supports. So is applying your rules, deciding whether a draft goes to you, and recalling pieces when a fact changes.

That is deliberate. A check done by code gives the same answer every time and cannot be talked round by a well-written draft. The part that writes is a model; the part that judges is not.

You confirm facts. Nothing can be cited until you do, and no model or code confirms one for you.

You set the rules, and you choose for each channel whether drafts wait for you or may be approved on your behalf. Four cases come to you whatever you chose: the first piece on a channel, a fact being used for the first time, anything a rule marks as regulated, and anything that spends money.

You publish. Campaign Fleet does not post to social networks, and it never sends email to a list: it prepares a draft and you press send.

They catch a claim that rests on nothing. They do not catch a fact you confirmed that was wrong, and they do not judge tone, taste or whether a piece is any good.

Drafts can be dull, clumsy or unsuitable. Read every piece before you publish it.

Much of the code, and many of the words on this website, were drafted with an AI coding assistant, directed and decided by the person who runs Campaign Fleet.

The examples on this site use a made-up shop. The sample receipt is an illustration, not a customer’s record.