Gains:
- Ability to clarify the premise, target audience and main claim of a non-fiction text (article, essay, book chapter, guide) and use artificial intelligence with the right context to produce a content plan (outline).
- Ability to request skeleton and expansion from artificial intelligence in chapter-by-chapter draft production and control each paragraph with its own knowledge
- Ability to preserve that the draft produced by artificial intelligence is a starting point and that the original argument and evidence belong to the author
Non-fiction text is writing based on fact: article, essay, review, guide, book chapter, report, whitepaper. The most difficult moment with such texts is often the first: a topic in your head, a few scattered ideas and a blank page. "Where do I start?" Emotion is the biggest enemy of productivity. Artificial intelligence (AI) is powerful in filling this exact gap; but on one condition: you decide what the text will say, AI only provides a sketch of how it will be arranged. In this unit, we will cover the path from idea to publication in three stops: premise and claim, content plan (outline), chapter by chapter draft.
Step 1: Premise, reader, and main claim
Every good nonfiction text is built on a main claim (thesis): the one-sentence essence that the text attempts to persuade or demonstrate to the reader. "An article about artificial intelligence" is a cloud, not a topic. “AI reduces the nonfiction writer's fear of the first draft but increases the burden of fact-checking” is one claim. As the claim becomes clearer, the text also becomes clearer.
Second constant: target reads. The same topic is written completely differently for a college magazine reader and a weekly newspaper reader. Defining the reader determines the tone, sample density, and assumed knowledge level.
Third: scope and length. An 800-word article and a 4,000-word section require different frameworks. Going to AI without determining these three constants is the shortest way to have it produce a generally valid text.
Tip: Before you start writing, write your main claim in one sentence on a piece of paper. This sentence is the compass of the text; I end each paragraph with “does this serve my argument?” You can test it by saying. Give this sentence to the AI so that it can build the skeleton around it.
Step 2: Producing a content plan (outline)
The content plan is the skeleton that shows the sections of the text and what will be explained in each section. AI generates multiple skeleton options within seconds given a good context. The trick here is not to ask for a single plan, but to ask for several alternatives and choose from them and blend them. This way, you are not stuck with AI's first idea, but you can build your own structure.
The following table shows the difference between a weak and strong plan request:
item
weak desire
strong desire
Subject
"Healthy eating"
"15-minute meal planning for busy workers"
claim
None
"Not time, but planning determines healthy eating"
reader
uncertain
"Adults who work in the office and have limited time"
output
single plan
2 alternative skeletons + justification
Conclusion
general-pass
It's yours, applicable
Here's a template you can use for a content plan:
Your role: content plan assistant assisting a nonfiction writer.Topic: [topic]. My main claim: [single sentence]. Target reads: [reads]. Length: [word].Tone: [tone]. Format: [article / guide / chapter].Task: Propose 2 different section skeletons that support this claim. For each section, write a 1-sentence purpose and the type of evidence (example, data, anecdote) you will suggest. DO NOT FAKE THE EVIDENCE; just tick "what type of evidence is needed". At the end, briefly compare the two skeletons.
Once you have the plan, it's your turn: which part will stay, which sequence flows better, what evidence can you provide? I asked each title the AI suggested “does this serve my argument?” Go through the question. Remove what does not serve; Add what you see missing.
Step 3: First draft, chapter by chapter
Once the skeleton is ready, it is better to proceed section by section rather than printing the entire text at once. The reason is simple: with a long output, the AI gets off topic, becomes repetitive, and you cannot control it. When you ask for each section separately, reminding you of the purpose of that section, both focus and control increase.
Template for a chapter outline:
Context: I am writing the 2nd part of the framework below. Purpose of the part: [purpose]. 3 points to go through in this section: [n1, n2, n3].The previous section ended with: [last sentence].Task: Write a draft paragraph of 250 words for this section.Tone: [tone]. Be concrete and simple; Avoid flowery generalizations. DO NOT make up facts or numbers; Mark the place where evidence is required as [SOURCE REQUIRED].
Once the draft arrives, comes the most critical step: checking each paragraph with your own knowledge and rewriting it in your own voice. The outline of the AI is a block of clay; The statue leaves your hands. Fill in [SOURCE REQUIRED] signs with your own research, replace generalizations with concrete examples, trim repetition.
Caution: The first draft the AI produces looks "finished" because it is fluid. This is an illusion. The first draft is where thinking begins, not where it ends. Every sentence without evidence, every generalization that does not belong to you is a debt waiting to be corrected.
From skeleton to text: the expansion trap
Once the skeleton is ready, an attractive shortcut appears: "let the AI expand each header and I'll assemble it." This is the most common trap. When expanding a title, AI falls into repetition because it does not fully see the context and the previous section, repeats the same idea with different words, and the connection between sections weakens. The result is a text that is "full" but does not flow and repeats each other. The correct method is to view the expansion as a material: read the draft produced by the AI and choose the best sentences, trim the repetition, build the bridges between the chapters yourself. The transformation of a structure into a living text arises not from the AI's filling, but from your labor of selection and binding.
Tip: After printing each chapter, ask "what idea is repeated between the previous chapter and this chapter?" ask yourself. Repetitions are the most distinctive fingerprint of AI-augmented texts; pruning them instantly professionalizes the text.
three mini cases
Case 1 — Plan acceleration. A technical writer would normally spend 2 hours creating the framework for a 3,000-word user guide. He gave AI the features of the product and the target user, took two skeletons, blended them in 25 minutes and added his own titles. Save: 30 minutes instead of 2 hours, plus a more compact structure.
Case 2 — Made-up statistics. A writer asked AI to draft a health article; AI produced an unsourced number saying "research shows 73 percent..." The author noticed this, [SOURCE REQUIRED] was initially overlooked due to his lack of discipline; When he looked for real data, he saw that there was no such ratio and removed the sentence. Lesson: mark each number in the draft and verify it later.
Case 3 — General validity trap. One essayist published the AI manuscript as is; The text was full of empty generalizations such as "in today's world, technology is in every aspect of our lives." The editor said the article contained no original observations. The text came to life when the author rewrote the draft, replacing each generalization with his own concrete experience. Lesson learned: AI gives skeleton, original observation comes from the author.
Weak prompt / Strong prompt
Weak prompt:
Write an article about remote work.
This claim is unpretentious, illegible and without scope. AI lists generalizations that everyone knows.
Powerful prompt:
Your role: assistant to a writer writing for the business reader. Topic: the impact of remote working on building trust in small teams. My main claim: remote working does not reduce trust, it forces trust rituals to be visible. Target reader: Managers of teams of 5-15 people. Length: 1,000 words. Tone: practical, evidence-based.Task: give a 4-part skeleton that supports this claim; State what type of evidence (case study, data, expert opinion) is needed in each section, but DO NOT make up the evidence.
The difference is clear: the claim, readership, scope and expectation of evidence are defined; The AI works in the service of your argument.
Common mistakes
- Asking for a plan without setting a claim. The unambitious skeleton leads to a text without direction.
- Printing the entire text in one long printout. Focus disperses, increases again, control is lost.
- Bypassing [SOURCE REQUIRED] discipline. Unmarked cases are overlooked later.
- Leaving the generalizations alone. Without concrete examples and original observations, the text remains without identity.
- Thinking the first draft was finished. The first draft is the beginning of thinking; The real work is in the correction.
In summary
In nonfiction, AI overcomes the fear of the blank page and reduces hours of skeletal work into minutes. But the main claim, the backbone of the text, is the target reader and the evidence comes from you. The right way: clarify the claim first, request and collate multiple skeletons, draft chapter by chapter, mark and verify each fact, and rewrite each paragraph in your own voice. AI builds the framework; You erect the building.
Application task
Choose a non-fiction topic and write your main claim in one sentence. With the content plan template in this unit, ask AI for two skeletons, blend them and add your own headings. Then have a draft of a section drafted by [citation needed], fill in the marked areas with your own research, and rewrite the paragraph in your own voice. Note how much of the resulting text comes from "you" and how much comes from "the AI."
checklist
- [ ] Have I clarified my main claim in one sentence?
- [ ] Have I defined the target reader, length, and tone?
- [ ] Did I request and blend multiple skeletons instead of a single plan?
- [ ] Have I marked and verified every fact/number in the draft with [SOURCE REQUIRED]?
- [ ] Did I rewrite each paragraph in my own voice?
- [ ] Have I replaced generalizations with concrete examples and specific observations?