How to Train ChatGPT on a Book Without Retraining the Model

To work with a book in ChatGPT, supply text you are permitted to share, explain your reading goal, and check each important answer against a specific passage. You are giving the assistant reference material, rather than retraining its underlying model. Tools like fenbs can preserve chapter references and open questions between reading sessions.

The useful outcome is a reading companion that can show its evidence. A confident explanation of a book is not enough when you need the author’s actual argument, an exact quotation, or a distinction between two editions.

How do you set up a reliable reading session?

1. Choose one purpose. Ask for help understanding a chapter, comparing two arguments, or preparing discussion questions. Avoid starting with a request to absorb the whole book forever.
2. Identify the source. Record the title, author, edition, chapter, and the page or section numbering your copy uses. Keep those details with every excerpt.
3. Supply permitted material. Attach a supported reference file or paste the relevant passage. Start with the smallest section that can answer your question.
4. Set an evidence rule. Ask ChatGPT to distinguish what the passage states, what it infers, and what the supplied material cannot establish. Require a location for claims about the book.
5. Check a sample before continuing. Ask a question whose answer you already know, verify the location, and correct any misunderstanding before requesting a broader explanation.

For example, say you are reading a nonfiction chapter about workplace incentives. Begin with one question: does the author argue that rewards always reduce motivation, or only under certain conditions? A focused question makes it easier to notice when the assistant broadens a limited claim.

Which version of the text should you provide?

Use a readable source that preserves the structure you need to cite. Keep chapter headings, section titles, and useful location markers. If the document contains scanned pages, check that the text you need can actually be selected and copied accurately before relying on its extraction.

OpenAI’s file upload guide↗ (opens in a new tab) describes supported document types and current limits. It also distinguishes text extraction from support for images inside documents. A successful attachment does not prove that every diagram, footnote, or page was understood.

Use material you have permission to share with the service. For a restricted manuscript or licensed ebook, check the relevant permissions and terms before uploading. When unsure, work from your own reading notes and permitted excerpts rather than assuming that access to a copy permits every use.

Keep the original nearby. If you prepare a shorter reference file, mark omissions and preserve location labels. Otherwise, a sentence from the end of one section may appear to belong to the next, changing the argument the assistant sees.

What instructions make the answers useful?

Separate the reading question from the evidence instructions. The question tells ChatGPT what to investigate. The instructions tell it how to handle gaps, quotations, and outside knowledge. You can reuse the following prompt with different passages.

Use the supplied passage as the source for claims about this book.
Source: [title, author, edition, chapter, location].
Question: [one reading question].
Separate: what the text states; your interpretation; unresolved questions.
Give a location for each central claim.
Do not invent page numbers or quotations. If evidence is missing, say so.
Use outside knowledge only in a clearly labeled separate section.

If you use a ChatGPT project, OpenAI documents how projects hold files and instructions↗ (opens in a new tab). Keep the source rule there and attach the relevant reference material. Still check the answer: organizing context does not certify that the assistant used the right passage.

Do not treat a summary as a substitute for the source on the next turn. Summaries compress detail and may omit exceptions. When a follow-up depends on a narrow phrase, provide that passage again or ask the assistant to locate it in the available material.

How do you check passages and quotations?

Run three checks: location, wording, and meaning. First, find the cited section in your copy. Second, compare any quotation character by character. Third, read the surrounding paragraph to see whether the explanation preserves the author’s qualification, example, or disagreement with another writer.

• For a summary, ask which passage supports each major point and which parts of the source were left out.
• For a quotation, request a short passage with its supplied location, then verify it yourself before using it elsewhere.
• For an interpretation, ask for another plausible reading and the text that would help distinguish the two.
• For a missing answer, accept “not established in the supplied material” instead of asking for an increasingly confident guess.

A useful test question has an answer that is absent from the excerpt. The assistant should identify the gap rather than fill it from a plausible memory of the book. If it invents evidence, narrow the reference material and restart the check before using the output for an assignment or published work.

What should you keep between reading sessions?

Keep a compact trail: the edition, the source location, your question, the checked conclusion, and what remains uncertain. Label your own interpretation separately from the author’s statement. If your view changes after a later chapter, update the conclusion and preserve why it changed.

Keep the book in its appropriate reading or document tool. fenbs can hold source links and work context, including the reference you need next, rather than acting as an ebook reader. This fits the broader practice covered in our context engineering guides.

If you move the discussion to another assistant, our Claude vs ChatGPT comparison covers broader tool differences. For a workplace reading project, ChatGPT for business addresses the surrounding work setting. Supplying a codebase has a different evidence loop, covered in training ChatGPT for coding.

How to do this with fenbs

1. Create a free account, then add a project for the book or reading assignment.
2. Add a task for the reading question. Include the edition, chapter locations, and the link to the source where you are allowed to keep it.
3. Record your evidence rules in Decisions and rules, such as verifying quotations and separating interpretation from the author’s claims.
4. Connect a supported assistant through Integrations. Ask it to read the project context before discussing the next passage.
5. Before stopping, save the checked findings and Where we left off. State which passages remain unchecked and the next question to investigate.

fenbs is useful when reading produces ongoing questions and work. It does not train a model, read your ChatGPT history, or replace a book reader. If you only want to understand one paragraph, the passage and a careful conversation may be all you need.