Store copy

Stage ① Copy writes your listing text in one metered call.

What it reads

• Your premise — title, genre, audience, description.
• Your story codex — characters, places, terms.
• Your chapter arc.
• The opening of chapter one.

That last one matters more than it looks. It is the only input that carries your actual voice, and a blurb written without it reads like stock copy for a book in your genre rather than copy for your book.

What it withholds

The last two chapters' goals are deliberately excluded from the request.

Otherwise your ending leaks into a store description, which is a mistake you cannot take back once it is live.

What you get

A blurb, a long description, keywords, category headings, and A+ content copy.

You can regenerate individual sections rather than the whole thing — so a description you are happy with survives another attempt at the keywords. Either way it is one metered call.

What it structurally cannot invent

There is no field for comp titles, review quotes, awards or bestseller claims.

This is structural, not a prompt instruction. A prompt that says "do not invent reviews" is a request; a schema with no review field is a guarantee. All four of those are false advertising if fabricated, and the safest way to not fabricate them is to have nowhere to put them.

Keywords are filtered on the server as well, so the same guarantee is not resting on the model's cooperation.

Categories and BISAC codes

The model returns category headings only — human-readable subjects.

Codes are then resolved locally from those headings. If nothing matches confidently, you get nothing, plus a note to pick one in the store.

That is the right answer: a BISAC code is nine characters of apparent precision, and a wrong one is indistinguishable from a right one until your book is filed under the wrong subject.

Before you generate

The call reads the saved version of your book, so pending edits are flushed first automatically. You do not need to do anything, but it does mean the copy reflects what you have actually written, not what you wrote a minute ago.

What to watch for

• Read it before you use it. It is a strong first draft of copy about your own book — you are still the one who knows whether it lands.
• A parse failure is not refunded. If the model returns something unusable the call still cost what it cost; that is deliberate and consistent with how metering works elsewhere.
• Generate copy before the Listings stage. Listings lay out copy that exists; with none written there is nothing to check limits against.