Part one described a small habit: the paragraph of set-up you write before you can ask an assistant anything useful about your own manuscript. Small habits are easy to dismiss, and this one gets dismissed on the grounds that it takes ninety seconds. The ninety seconds are the least of it.
What follows is not an argument that you are wasting time. It is an argument that the briefing is a filter, that every filter has a shape, and that the shape of this one has been pressing on your book for as long as you have been using one.
A filter has a shape, and this one has a strong one
You cannot brief a partner on your whole manuscript, so you brief it on a selection. You do not choose that selection at random. Under mild time pressure, at the start of a session, you reliably include the things that are easy to state and omit the things that are hard to state.
Easy to state: names, ages, the premise, what happened in the previous chapter, the genre. Hard to state: tonal range, the reason a particular relationship stays cold, the rhythm of your paragraphs, what you have already tried and thrown away.
So the partner receives a version of your book that is systematically heavier on plot machinery and lighter on everything that makes it a specific book by a specific person. Then it answers well — genuinely well — about the thing it received.
The suggestions are not bad. They are excellent answers to a briefing, and the briefing is the most generic possible account of your novel.
What comes back when the middle of the book is missing
Here is the part that is hard to see from inside a session. A partner briefed on premise and last-chapter-summary will tend to suggest the most probable next move for a book of that description. That is the correct behaviour given what it was told. The trouble is that the most probable next move for a book of that description is exactly the move a reader of that genre has seen before.
Your actual book has spent forty thousand words earning the right to do something less probable. The evidence for that right is in chapters six through fourteen. Chapters six through fourteen were not in the briefing.
So the pressure of the suggestion runs, session after session, in one direction: back toward the centre of the genre. Nobody proposes this. It is not a flaw in the model's taste. It is arithmetic about what it knows.
The corrections you do not count
Ask a writer who works this way what the assistant costs them and they will say a few minutes a day. Then watch what actually happens to a returned suggestion.
You rewrite the voice. Not consciously, most of the time. A phrase comes back a shade brighter or a shade more formal than your book, and you flatten it as you paste it. That is a real edit, performed at speed, on every accepted line.
You check it against a book only you can see. Is this consistent with what she said in eleven? You are the continuity department, because the partner has no access to eleven. This is the specific work you most wanted help with and it has quietly come back to you with an extra step attached.
You throw away the near-misses. The suggestions that would have been right for your book if it had known two more things. You do not log those. They feel like the tool being imperfect rather than the tool being uninformed.
You stop asking the interesting questions. This is the one that matters. After enough sessions you learn, without deciding to, which questions are worth asking. Questions answerable from a briefing — give me three ways to open this scene — stay. Questions that need the whole book — is this chapter doing work I already did in eight? — quietly disappear from your repertoire, because you have learned they come back hollow.
Not in the ninety seconds of briefing. In the shrinking of the question set, which happens slowly enough that you will attribute it to knowing your own book better rather than to having been trained by the limits of the thing you were asking.
Voice is the account that quietly drains
Continuity errors are recoverable. Somebody catches them; you fix the name. Voice is different, because there is no error to catch. Nothing is wrong with a line that is a little more average than your line. It just is not yours, and the reader who feels it will not be able to say why either — they will simply find the book slightly less distinct than the thing that made them pick it up.
Accept enough near-your-voice lines across a draft and the average moves. Not in one place you could point at. Across the whole thing, a few degrees.
This is the cost that is hardest to argue about honestly, so let me be precise about what I am and am not claiming. I am not claiming a measurement of how far any particular draft drifted; I have no such number and would not trust one that came from us. I am claiming something narrower and, I think, unarguable: a partner that does not know your voice cannot check a suggestion against it, and you are therefore the only check that exists. Everything that gets past you goes in.
The honest case for ignoring all this
Some writers genuinely do not have this problem, and it is worth saying who they are.
If you use an assistant for discrete, self-contained jobs — tidy this paragraph, give me a name for a coastal town, what is the word for the smell after rain — the briefing cost is near zero, because those tasks do not need the book. That use is fine and will stay fine. Nothing in this arc argues against it.
If you are two chapters into a first draft, there is barely a manuscript to know yet, and a briefing describes nearly all of it. The problem arrives later, and it arrives in proportion to how much book you have accumulated — which is an uncomfortable property, because it means it is worst exactly when you are closest to finishing and most reluctant to change anything.
What would have to be true instead
State the requirement as a demand rather than a complaint and it becomes surprisingly concrete. You want a partner for which the briefing is unnecessary, because the manuscript is already what it is reading. You want that to persist between sessions without maintenance from you. You want it to hold not only the facts but the decisions, including the ones you never wrote down, and to have learned your voice from the pages you actually wrote rather than from a description of them.
That is four properties, and none of them is about the quality of the prose that comes back. Part three takes them one at a time, says what each looks like in practice, and is honest about which parts are still genuinely hard.
