A literature review outline is not a table of contents for the papers you read. It is a decision about the order of your claims. Once that order exists, the papers stop being the subject and become evidence that sits inside the claims.
The difference shows up in the first draft. An outline built from papers produces headings like “Smith (2019)”, “Statistical approaches”, “Recent work” — and a draft that reads as a list. An outline built from claims produces headings that each assert something, and the papers arrive where they are needed.
This article is about the outline itself: how to divide the body, what belongs in each section, a skeleton to copy, and how to tell a broken outline from a working one. It assumes you have already collected the papers and taken notes; outlining before you have the notes just gives you a shape you then hunt for evidence to confirm.
Start from the sentence the review argues
Before any headings, write the one sentence the whole review exists to support. This is not the question you started with. It is the answer you can now defend.
Protein-structure models are often validated on structures released before their training cutoff, so published accuracy figures are an upper bound.
That is a sentence an outline can be built from. It already tells you the body has to establish three things: how validation is normally done, why the cutoff matters, and what happens to the numbers when you control for it.
If you cannot write that sentence yet, an outline will not rescue you: the sections you invent will be topic buckets, and you find that out at the writing stage.
Pick one organizing principle for the body
Writing guides commonly describe four. Choose one and use it at the top level of the body. Mixing them at the same level is a common structural failure.
| Principle | Sections are | Use it when | It fails when |
|---|---|---|---|
| Thematic | concepts or sub-questions | the default choice; the field disagrees about what matters | each “theme” turns out to be one paper |
| Chronological | periods or generations | the change over time is the argument | nothing decisive changed, so you write a timeline instead of a claim |
| Methodological | approaches or study designs | the disagreement in the field tracks which method was used | the methods differ but the results agree, so the split explains nothing |
| Theoretical | schools or framings | the field splits into camps with incompatible assumptions | the camps are your invention, not something the authors would recognize |
Nesting is fine and often correct: thematic at the top level, chronological inside one theme where the story really is “this got fixed in 2022”. What does not work is alternating — two thematic sections and then a chronological one — because the reader can no longer predict what the next heading is about.
The test for whether you picked the right principle: it should make the disagreements in the field visible. If two papers contradict each other and your structure files them into different sections where they never meet, the structure is wrong, however tidy it looks.
What belongs in each part
- Introduction. Scope (what is in, what is out, and why), the question, and one roadmap sentence naming the body sections in order. Reviewers often read this part to decide whether the rest is worth their time.
- Background or definitions. Include this only if the field uses a term in a contested way. Otherwise cut it — it is an easy section to pad.
- Each body section. The heading carries one claim. Under it: what is agreed, what is contested and by whom, how good the evidence on each side is, and one sentence on what this leaves unresolved for your question.
- Synthesis. Where those “unresolved” sentences are collected and put together. This is where the gap gets named, and it should read like arithmetic on what came before rather than a new idea.
- Conclusion. Answer the question at the strength the evidence actually supports. If the honest answer is “the studies are too heterogeneous to compare”, that is the answer.
One warning about gaps. Many templates give “Gaps in the literature” its own section near the end. It works badly: a gap that arrives with no build-up reads as an assertion. Let each body section end by leaving something open, and the synthesis has material to assemble.
A skeleton you can copy
Thematic, three themes. Replace the bracketed parts.
Title: [the claim sentence, shortened]
1. Introduction
1.1 Why this question now — [what changed that makes it askable]
1.2 Scope — includes [X, Y]; excludes [Z], because [reason]
1.3 The question — [one sentence]
1.4 Roadmap — "Section 2 establishes ..., Section 3 ..., Section 4 ..."
2. Theme A — [the claim, e.g. "Validation sets overlap with training data"]
2.1 What is agreed — [consensus + the 2-4 papers everyone cites for it]
2.2 What is contested — [the disagreement, and who is on each side]
2.3 Quality of the evidence — [designs, sample sizes, what would settle it]
2.4 Left open — [one sentence]
3. Theme B — [the claim]
3.1-3.4 the same four slots
4. Theme C — [the claim]
4.1-4.4 the same four slots
5. Synthesis
5.1 Where the themes agree
5.2 Where they conflict, and why — [method? population? definition?]
5.3 The gap — [assembled from the "left open" sentences above]
6. Conclusion — [the answer, at the strength the evidence supports]
The four repeated slots — agreed, contested, evidence, left open — are what makes this an outline rather than a list. They also make section length predictable, which matters when you have a word limit.
For a chronological body, swap sections 2-4 for periods and change the slots to “what this period could newly do” and “what it still could not”. For a methodological body, swap them for method families and add a slot for “what this method can and cannot detect”. The introduction, synthesis, and conclusion do not change.
Three tests before you write
Run these on the outline. Each takes a minute, and each catches a failure that otherwise costs you days.
- The headings-only test. Read your headings, in order, and nothing else. Do they make an argument, or list topics? If a stranger cannot tell what you are claiming from the headings alone, you have topics.
- The one-paper test. Any section resting on a single paper is not a theme, it is a paper. Merge it into a neighbour, or go find its counterparts.
- The deletion test. Delete a section. Does the conclusion still hold? If it does, that section is decoration — cut it, or make it load-bearing.
A fourth, if you have time: for each section, answer “what does this let me say that the previous section did not?” If you cannot, those two sections are one.
Getting from the outline to a draft
Once the outline exists, the remaining work is mechanical: put each paper under the section that will use it. A paper that fits nowhere usually belongs to a review you are not writing; one that fits in three places usually means your theme boundaries are wrong.
Litlas is where we do this part. Papers go into a board for the review and each one carries your own note; because the library’s search covers note text as well as titles, writing the section into the note — “3.2 contested, small-sample failure” — turns every outline section into a one-query list. The review itself can be drafted in the built-in editor (Markdown, LaTeX, or rich text). In the LaTeX and rich-text editors, citations are inserted from your library and the reference list is generated in APA, Chicago author-date, IEEE, MLA, or Vancouver; the Markdown editor has no citation picker. The free plan covers 3 boards and 100 saved items, plus 5 searches, 3 ✨ hand-offs, and 5 graph views a day, which is enough for one review.
Two honest limits. Litlas will not produce the outline for you. It runs no language model of its own; the ✨ button hands a PDF or a draft off to Gemini, ChatGPT, Claude, or Gemini Notebook, and what comes back is their output, reached through Litlas rather than generated by it. And there is no systematic-review machinery here: no screening workflow, no PRISMA diagram, no inclusion and exclusion tracking. If your review has to be reproducible in that sense, you need a dedicated tool for that part.
The failure worth avoiding
The outline that fails is rarely the one that was too thin. It is the one whose headings name the things you read: “Deep learning approaches”, “Statistical approaches”, “Other”. They look like structure, but they commit you to nothing, so nothing in the outline can be wrong — and you find that out at the writing stage, when the sections refuse to add up to a conclusion.
Headings that make claims can be wrong. That is what makes them useful, and being wrong at the outline stage costs an afternoon instead of a fortnight.