Interactive eBook BuilderInteractive eBook Builder
$59 · unlimited access, unlimited eBooksGet it for $59

· By · 13 min read

Chain-of-Thought Prompting Examples for eBook Writers

promptsaiwritingebook

Chain-of-thought prompting asks the model to work through intermediate reasoning steps before it gives the final answer. For eBook work, that means planning, checking constraints or structuring an exercise in the open, then writing the deliverable. In Chain-of-Thought Prompting Elicits Reasoning in Large Language Models, Wei and colleagues at Google Research described a chain of thought as a series of intermediate natural language reasoning steps that lead to the final output. They showed that providing a few exemplars in that form improved performance on arithmetic, commonsense and symbolic reasoning tasks. With PaLM 540B and eight chain-of-thought exemplars, they reported state-of-the-art accuracy on the GSM8K math word-problem benchmark. They also found that the gains showed up in sufficiently large models and were not a reliable win for much smaller ones.

For eBook and workbook projects, Interactive eBook Builder is a $59 one-time Notion AI template with Spec Builder, Drafter, Polisher and SEO BOT already set up, so you can put the reasoning in a planning stage and keep the draft itself focused on the page the reader will see. Notion AI access and usage are separate from the purchase. The examples below are teaching prompts written for this article. They are not paid Builder prompts, and they are not recorded model outputs. This is also not the same technique as prompt chaining, which splits a project into separate requests you can inspect between stages. Chain-of-thought happens inside one request; chaining happens across several.

What does chain-of-thought prompting look like?

In the Wei et al. paper, standard few-shot prompting showed a question and a final answer. Chain-of-thought prompting showed a question, the intermediate steps, and then the answer. The model was expected to produce the same kind of steps on a new problem. The authors noted several useful properties: multi-step problems can be broken into intermediate steps; the steps give you a window into how the model arrived at an answer; and the method is, in principle, applicable to tasks people solve with language. They also stated an important limit: there is no guarantee that the reasoning path is correct, and an incorrect chain can still lead to a correct or incorrect answer.

A later paper, Large Language Models are Zero-Shot Reasoners by Kojima and colleagues, showed that you can often elicit a similar step-by-step process without hand-crafted examples, by appending a cue such as “Let’s think step by step.” Using the large InstructGPT model text-davinci-002, they reported that this zero-shot method raised accuracy on MultiArith from 17.7% to 78.7% and on GSM8K from 10.4% to 40.7%, with similar gains on a 540B-parameter PaLM model. Those figures are from the paper’s experiments on reasoning benchmarks, not from eBook writing tests. They are a reason to try a step-by-step brief on planning tasks, not a promise about your chapter quality.

Vendor guidance has since added a nuance. OpenAI’s reasoning-model best practices say those models already perform reasoning internally, and that prompts such as “think step by step” or “explain your reasoning” may not help and can sometimes hinder performance. For non-reasoning GPT models, OpenAI’s broader prompt engineering guide still treats explicit planning as useful on complex tasks. Anthropic’s prompting best practices describe manual chain-of-thought as a fallback when thinking is off: ask the model to work through the problem before it answers, and put the final answer in tags so you can extract it. On current Claude models with thinking enabled, Anthropic recommends relying on that thinking rather than forcing a written chain.

The practical rule for writers is simple. If you are using a model that already thinks privately, a clear brief may be enough. If you are using a general chat model, or you want to see the steps yourself, ask for the reasoning first and the deliverable second.

When does this help an eBook, and when does it get in the way?

Chain-of-thought is most useful when the model has to make a decision before it writes: what belongs in a chapter, whether a claim is supported by your notes, how an exercise should work, or which of two outlines better serves the reader. It is less useful when you already know the shape of the page and only need prose. Asking the model to reason out loud through every paragraph can add length, delay, and a false sense of rigor.

TaskAsk for steps first?Why
Choosing a chapter angle from messy notesYesThe model has to resolve conflicts before outlining
Checking a draft against supplied factsYesYou want the mismatches listed before a rewrite
Designing a workbook page from a chapterYesThe activity has to be completable, not just related
Polishing a paragraph for clarityUsually noThe job is already defined; extra reasoning can wander
Drafting a section from an approved outlineUsually noThe decisions have already been made

Wei et al. also observed that the method helped more on harder, multi-step problems than on easy one-step items. That maps reasonably onto book work: use it where there is a genuine decision, not as decoration.

Example 1: plan a chapter from conflicting notes

A vague prompt is:

Here's a pile of notes. Write the chapter on pricing.

A better zero-shot brief already helps. A chain-of-thought version adds an explicit planning pass:

You are helping me plan one chapter of a short eBook for new freelance
designers. The chapter is about naming a floor rate.

Notes (some of these may conflict):
- The reader currently charges favors or student-level prices.
- I want them to set a floor before the next inquiry, not a full pricing menu.
- I mentioned a $75 example in one note and a $120 example in another.
- I do not have survey data on local rates, and I do not want invented
  "industry averages."
- The chapter should end with a fill-in line for the reader's floor rate.

Before you write an outline, think step by step:
1. What is the reader's job after this chapter?
2. Which notes support that job, and which conflict?
3. Which conflicts need a decision from me instead of a guess?
4. What will the outline include, and what will it leave out?

Then return:
- A short "Decisions and gaps" list, with any "Need from me" items
- A five-heading outline that uses only the settled notes
- One workbook line the reader will complete

Do not invent a recommended rate. If the two dollar amounts conflict, flag
that instead of averaging them.

A possible target—not a recorded answer—is a decisions list that flags the $75 versus $120 conflict, an outline that teaches how to set a floor rather than announcing one, and a workbook line that leaves the number to the reader. If the model “resolves” the conflict by inventing a market rate, the chain failed the most important test, even if the outline looks tidy. That is the limit Wei et al. pointed out: the steps can be fluent and still be wrong. Read them.

Example 2: check a draft against your notes

Writers often ask a model to “improve” a section and then cannot tell what changed. A chain-of-thought check makes the comparison visible first.

Compare the draft below with the notes. Think step by step through each
claim, number and piece of advice in the draft. For each one, say whether
it is supported by the notes, contradicted by the notes, or missing from
the notes.

Then list:
- Supported points to keep
- Unsupported points to remove or mark as "Need from me"
- Any reader-job the draft failed to serve

Do not rewrite the section until that list is complete. Do not add new
advice to fill a gap.

Notes:
- The reader has 45 minutes before the first client call, three days a week.
- The three actions are: review today's calendar, write the first task in
  one sentence, and prepare the workspace for the first call.
- No supplements, no 5 a.m. schedule, no new apps.

Draft:
[paste the section here]

This is closer to a reasoning task than to a writing task, which is why a step-by-step pass helps. You can also ask the model to quote the relevant note before it judges a sentence, a pattern Anthropic recommends for long-document work: ask it to quote first, then carry out the task, so it stays attached to the source.

Example 3: design a workbook exercise before writing it

A vague request produces a generic worksheet. A chain-of-thought brief forces the model to prove the exercise is doable:

Design one five-minute workbook exercise for the chapter on stating a
revision limit before work starts.

Think step by step before you write the page:
1. What should the reader be able to do after completing the page?
2. What information do they already have from the chapter?
3. What would make the page impossible to finish in five minutes?
4. What format leaves writing space and avoids scoring or labels?

Then output:
- A one-paragraph rationale
- The exercise itself: one-sentence instruction, three fill-in lines, and
  one backup question
- A completion test: "A reader can finish this in five minutes if..."

Use only the chapter notes below. Do not add personality quizzes, point
scores or legal language.

Notes:
- The reader is a freelance designer who says yes to extra revisions.
- The chapter teaches them to state a revision limit in the kickoff email.
- They need space to write a limit and a one-line email sentence.

The rationale is there for you to inspect. If it says the page will “transform their mindset,” but the lines only ask for a number and a sentence, you can see the mismatch before you keep the page. If you later want the model to draft the surrounding teaching text, do that in a second prompt, or in a chained step, using the approved exercise as input.

How should you write the cue?

You have three honest options:

  1. A zero-shot cue, in the spirit of Kojima et al.: after the brief, add “Think step by step before you answer. Then give the deliverable.”
  2. A structured list of steps, as in the examples above, when you know the questions that should be asked.
  3. A few-shot chain, in the spirit of Wei et al.: show one or two worked examples that include the reasoning and the final output.

For book work, option 2 is often the most practical. You already know the quality questions—reader job, conflicts, unsupported claims, completion tests—and writing them out is more precise than hoping the model invents a useful sequence. Use option 3 when a specific reasoning style matters, such as always flagging dollar amounts that were never in the notes. Use option 1 when you want a quick planning pass and have not yet settled the questions.

If you are using a reasoning model, follow the vendor’s current guidance: a clear goal and constraints may be enough, and an extra “think step by step” may not help. Check the model’s documentation, because this is an area that changes.

How does this fit Spec Builder, Drafter and Polisher?

A book-length project benefits from putting the reasoning in the right stage. Spec Builder is the natural place for the “decisions and gaps” pass. Drafter can then write from an approved outline without re-litigating every conflict. Polisher can run a notes-versus-draft check before it changes wording. That is closer to chain-of-thought used on purpose than to one prompt that tries to think, write and edit at once.

SEO BOT can later take approved chapter topics and reader questions and publish supporting articles to the Auto-Blogger blog, each linking back to your eBook, site and brand. Those articles still need a human pass, especially when a chain of thought has made a draft look more confident than the sources support.

FAQ

Is chain-of-thought the same as prompt chaining?

No. Chain-of-thought is a way of asking for intermediate steps inside one prompt. Prompt chaining is a sequence of separate prompts, where you inspect the output of one step before starting the next. You can use both: a planning prompt that reasons in the open, then a drafting prompt that uses the approved plan.

Does “Let’s think step by step” always improve writing?

No. Kojima et al. reported large gains on reasoning benchmarks with that cue, not on eBook drafts. It can help when the model needs to make a decision first. It can also add padding, and on some reasoning models OpenAI advises against it. Test it on one planning task and read the steps.

Should I show the reasoning to the reader?

Usually not. The chain is a working aid for you. The chapter should still read like a chapter. Ask the model to put the reasoning in a separate labeled block and the reader-facing text in another, then keep only the latter in the book.

Why did the model give the right outline for the wrong reasons?

Wei et al. noted that a chain of thought can be incorrect even when the final answer happens to look right. That is why you inspect the steps, especially wherever numbers, claims or conflicts appear. If the reasoning is wrong, do not keep the outline just because the headings are tidy.

Before you hit send

Ship niche eBooks with agents already built.

Get the Interactive eBook Builder for $59