How do I organize a literature review by argument instead of by paper?

The reason paper-led chapters happen is that the papers are the objects you can hold. You have a folder of them, each with a name, and the natural unit of writing feels like the natural unit of reading. But a reader does not want to know what twenty papers said. They want to know what is known, how confidently, and where the field is unsure.

So the reorganization is mechanical. Go through your notes and write out every claim you might use, one per line, each carrying the source and the page. Aim for the level of specificity you would actually put in a sentence. "Retention improves with mentoring" is usable. "Fenton looks at mentoring" is not.

Then spread the lines out and group them. Use a spreadsheet, a wall, or index cards, whatever lets you move things without committing. The groups that form are your sections, and they will not match the sections you had planned, which is the point of doing it.

Name each group with the question it answers, not the topic it covers. Lingard calls the literature review a map of a meaningful white space rather than a laundry list of known facts, and recommends moving from selected knowledge claims toward a characterized gap, which matters most in a well-studied field where you owe the reader an explicit case for why another study is needed (Lingard, 2018, Part I, pp. 47-48). Pautasso sets the same bar from the other direction: a review should identify the major achievements, the areas of debate, the methodological problems, and the outstanding questions, rather than summarize publications one at a time (Pautasso, 2013, p. 3).

The tool that makes this practical is a claim-and-evidence table instead of paper-by-paper notes. One claim per row, then columns for which studies support it, which qualify it, which challenge it, and how the methods differ. Thomas and Harden built their synthesis this way, comparing codes across studies and grouping similarities and differences into a hierarchical tree that produced 12 descriptive themes (Thomas and Harden, 2008, p. 6). Their coding of children's food preferences sat under both children's understandings of healthy eating and influences on food choice, which is the concrete case for letting one study feed more than one section when its findings warrant it.

Two things usually surface at this stage. The first is a set of claims that appear in six papers and were never worth six paragraphs. The second is a claim that only one paper supports, which is either a gap or a weakness, and either way you now know it.

How do I find the themes when every paper feels different?

Look for the questions the papers answer rather than the topics they cover. Two studies on different populations using different methods are on the same theme if they are arguing about the same thing. Topic similarity scatters your chapter, question similarity organizes it.

A useful test is to write the implied question above each claim. "Does the effect persist?" "Does it hold outside the original setting?" "Is the measure capturing what it claims to?" Papers that share an implied question belong together even when their surface subject matter looks unrelated.

Grouping is only the first half. Thomas and Harden separate descriptive themes, which stay close to what the primary studies reported, from analytical themes, which answer the review's own question by inferring barriers, facilitators, and implications, and they reached the second kind by interpreting the descriptive themes independently, discussing them together, and checking each emerging abstraction back against the descriptive themes (Thomas and Harden, 2008, pp. 7-8). That distinction exists because Dixon-Woods and colleagues had warned that thematic work becomes theoretically weak and insufficiently transparent when it only restates the themes the primary studies already named. It is the difference between a section that says several papers mention X and a section that argues X operates as a barrier under specific conditions.

Weight the groups by study quality when you write them up, not just by how many papers landed in each. In their synthesis the poorer-quality studies contributed relatively little and held few unique themes, while the better-quality studies carried more developed analyses and drove most of the final recommendations (Thomas and Harden, 2008, p. 8). They also qualify how far this goes: their review produced propositions absent from the primary studies because the review question and an external framework allowed new concepts, whereas a literature whose studies already address your question directly may need translation rather than invention (Thomas and Harden, 2008, p. 9). Lingard's version of the same caution is to characterize the gap and then align it with the method you have chosen to fill it (Lingard, 2018, Part I, p. 49).

If a paper still refuses to join a group, that is information. It either belongs in a different chapter, or it is answering a question your review has not yet acknowledged, in which case the missing section just announced itself.

Should the chapter be chronological or thematic?

Thematic, with chronology used inside a theme when the field genuinely moved. A purely chronological review implies that later work supersedes earlier work, which is often false, and it forces the reader to hold the argument together themselves. Use dates to explain a shift, not to structure the chapter.

Chronology earns its place when something specific changed: a method became available, a definition was revised, a major dataset was released, a foundational result failed to replicate. Then the sequence is the explanation and saying so is genuinely informative.

Outside those cases, a run of paragraphs beginning "In 2014", "In 2017", "More recently" is chronology used as a substitute for structure. The dates are doing the work your argument should be doing.