06 / THE WRITING SHELF

AI for Writers: Brainstorming, Feedback and Use Records

A hand rearranging cream index cards into a developing book sequence beside a blue notebook
3 GUIDES + A PRACTICAL TOOL / ILLUSTRATIVE ART

Work thoughtfully with AI,
one decision at a time.

AI can produce options, questions, and draft material. The useful question is which bounded task you want help with and how you will evaluate the result. Asking for three possible obstacles is different from asking a system to decide the story. Asking where an excerpt becomes confusing is different from accepting every suggested rewrite. Keep enough of your own intention visible to make that distinction.

These guides focus on brainstorming, feedback, and recording how assistance was used. They are practical writing workflows, not legal conclusions or guarantees about a model’s output. Services handle submitted content under their own current policies. Publishers, competitions, and platforms can set different requirements. Review the actual destination before sharing or submitting work, rather than assuming a general label such as “AI assisted” resolves every question.

Find the right lesson
START WITH YOUR QUESTION

Choose assistance with a decision in mind.

AI is optional. Begin with the task you want help thinking about, the material you are comfortable supplying and the judgment you will retain.

  1. I need alternatives.

    Set boundaries, compare possibilities and write a test yourself.

    YOU’LL MAKE A recorded idea decision
  2. I want to inspect a passage.

    Ask one question and check every observation against the text.

    YOU’LL MAKE A checked feedback ledger
  3. I am preparing a submission.

    Audit retained components and check the destination’s current rules.

    YOU’LL MAKE A factual use record

These lessons teach evaluation and record-keeping. The local BookDraftLab tools organize your own notes and do not generate AI text or send a manuscript to an external service.

ASSISTANCE NEEDS A BOUNDARY

Keep the judgment visible.

THE LOOSE VERSION

Rewrite my chapter and make it better.

A CHOICE YOU CAN TEST

Identify two places where the character’s immediate goal is unclear. Quote the relevant sentence, distinguish observation from inference and ask a question. Do not rewrite.

A bounded request makes a suggestion easier to inspect. It cannot make generated claims, citations or judgments reliable by itself.

Your turn: Save the original, choose one suggestion to investigate and record what you actually keep.

An original invented teaching example.
A hand comparing three cards with different pencil gestures
THE WORK, MADE VISIBLECompare before you choose.

Test alternatives against the same intention. Keep responsibility for the choice.

Original editorial illustration
AN INVENTED FEEDBACK EXAMPLE

A useful critique can be partly wrong.

A completed example
The suggestion
“The reader does not know why the visitor needs the key.” Treat this as a hypothesis about a particular passage.
The check
The request may be explicit in the dialogue but obscured by a long memory between the question and answer. Read that sequence before adding an explanation.
The writer’s test
Move the memory after the response and read both versions. Keep the change only if the exchange becomes easier to follow without losing a necessary motive.

An original illustrative specimen. Adapt the method to your own material.

Choose a complete lesson.

The whole library
Two illustrative readers comparing manuscript passages at a bright reading table
THE WORK BETWEEN THE PAGESA response you can investigate.

Locate the passage, separate the reaction from a diagnosis and choose one change worth testing.

Original illustrative photograph, created for BookDraftLab.

Make the request small enough to check.

Supply the context required for the task, identify what must stay unchanged, and ask for a limited number of alternatives. Inspect the reasoning against the actual excerpt or project brief. A model may invent a quotation, misread a narrator, or propose a fact that needs research. Preserve the original and test a change in context. The goal is a decision you understand, not a manuscript made uniform by accepting every confident suggestion.

Separate a local worksheet from an online service.

BookDraftLab’s tools run in your browser and do not perform AI generation. Fyrlo is an optional separate app for AI-supported story work, with its own service and privacy policies. A note saved here is not automatically copied into the app. Keep unpublished or sensitive material out of any service until you have checked whether its handling suits the project. Record meaningful uses as you go so a later disclosure question does not depend on reconstructing months of decisions.

Decide what a useful answer would look like.

Before sending a prompt, describe how you will recognize a helpful response. If you want to test clarity, ask the tool to point to the exact wording that made an interpretation possible. If you want alternatives, say how many and which constraints they must respect. An answer that sounds polished but does not address the task is not a reason to rewrite the whole excerpt. Compare the response with the question you actually asked.

Keep the evaluation separate from another round of generation. A system can produce an explanation for a weak suggestion as fluently as for a strong one. Return to the source passage, the project brief, or an appropriate external source when a factual claim matters. For submission questions, read the destination’s current policy directly.

A useful workflow gives you a record of the task, the material supplied, what was accepted or rejected, and what was checked. That record supports your own decision-making; it does not certify a work’s legal status or make different organizations’ requirements identical.

A closer look