Outlining a Thesis Chapter With AI: Structure First, Prose Second

A thesis chapter fails most often at the structural level: the argument appears in the conclusion, sections repeat each other, and the literature review lists papers instead of establishing a position. AI can help build and test structure, which is exactly the part that is hard to hold in working memory. It cannot supply the argument, and a model-generated outline that looks tidy is often an outline of the topic rather than of your claim.

State the claim before outlining

Write one sentence: what this chapter argues and why a reader should accept it. Every section must serve that sentence. If a section cannot be tied to it, it belongs in another chapter or in an appendix.

Feed the model the claim, the chapter's source list, and the word budget. Ask for a section structure where each section has a one-sentence function. Reject any section whose function is "provide background" without saying what background the argument needs.

Map sources to sections

Create a table with sources as rows and sections as columns, marking where each source contributes evidence, method, or counterargument. Empty sections with no sources need either data or deletion.

The table also exposes over-reliance: a section citing five sources for one uncontroversial point is padding, while a contested claim resting on one source is a vulnerability a supervisor will spot.

Build the argument chain

Each section should end with a sentence that sets up the next. Write those transition sentences before drafting any prose. If a transition is difficult to write, the section order is wrong.

A common structure for an empirical chapter: question, prior work and its limits, method, findings, interpretation, threats to validity. A theory chapter: construct definition, existing accounts, points of disagreement, proposed resolution, implications.

Ask the model to critique, and align with your supervisor

The more valuable prompt is adversarial: "Here is the outline and the claim. Identify sections that do not support the claim, repeated content, and missing counterarguments." Models are better at finding structural inconsistencies than at inventing structure.

Run the critique against the source table as context. Without it, the model will suggest generic additions that do not exist in your material.

Supervisors have expectations about chapter order and length that are rarely written down. Before investing in drafting, send the one-page claim and outline. Getting a structural correction at this stage costs an email; getting it after 8,000 words costs a month.

From outline to draft

Draft in the order of the argument, not the order of the sections. Write the findings section first if it is the most concrete, then the interpretation, then the introduction that promises what the chapter delivers. Introductions written last are shorter and more accurate.

Keep the outline as a living document with word counts per section. A chapter that is 40% literature review usually means the argument was unclear at the start.

Word budgets, and mapping claims to evidence

A 15,000-word empirical chapter typically allocates 2,000 to the introduction and question, 3,500 to prior work, 3,000 to method, 4,000 to findings, and 2,500 to discussion. These are starting points, not rules, but writing them into the outline prevents the common failure of a 6,000-word literature review and a 1,500-word discussion.

Adjust for the discipline. A theory chapter in the humanities allocates differently from a lab-based science chapter; ask the supervisor for a recently accepted example from the department.

For each section, list the claims it makes and the evidence behind each: data, citation, or argument. A claim with no evidence is a note to yourself, not a section.

Reviewers of the chapter—supervisors, examiners—will test the chain from claim to evidence. A table makes weak links visible before the writing begins.

Handling contradictory sources, and figures as outline elements

Most chapters need a position on conflicting literature. Outline the conflict explicitly: source A finds X, source B finds Y, the difference is explained by method Z, and this chapter does W. That structure prevents the literature review from becoming an annotated bibliography.

Decide the figures before writing the prose. Each figure should carry one claim. A chapter whose figures appear as an afterthought usually has findings that were not planned.

Write the figure captions during outlining; a caption that states the claim reveals whether the figure supports it.

Transitions, supervisors, and what examiners expect

Read only the last sentence of each section and the first sentence of the next. If the chain does not read as an argument, the order is wrong. This test takes five minutes and catches structural problems that would otherwise surface after 10,000 words.

Send the claim, outline, word budget, and figure list as a two-page document. Supervisors give better feedback on structure than on prose, and structural feedback at this stage saves weeks.

When each section has a claim, evidence, and a transition, and the figure list is stable. Further outlining becomes procrastination; the remaining uncertainty resolves through writing.

Ask for a recently passed thesis in the department and read its chapter structure. Expectations about length and theory coverage vary more than guidelines admit.

If a chapter has no figure or table after outlining, ask whether it is making an empirical claim. Theory chapters may legitimately have none, but empirical ones rarely do.

Draft the findings first, then the discussion, then the introduction. The introduction written last describes what the chapter actually does.

Give each chapter a target range before drafting. A methods chapter at 4,000 words and a findings chapter at 900 signals a structural problem that no amount of editing fixes.

Compare your budget to the department's completed theses. The average is a better guide than the maximum.

A chapter that summarizes papers one by one is a bibliography with paragraphs. Group by debate or method, and let each section end on what the grouping implies for your question.

Aim for one argument per section. If the outline has five sections and no argument, the chapter will read as a list no matter how well it is written.