A literature review is an
argument, not a reading list.

Read in passes, not cover to cover: First pass is title, abstract, and conclusion, which is enough to decide whether the paper is in or out. Second pass is figures, tables, and the method, which tells you what kind of evidence it is. Only the papers that survive both get a full read. Most reviews stall because the reader treats all sixty sources as equally deserving of three hours.

Record the method and sample beside the finding: A claim without its method and sample cannot be synthesised, only repeated. Put the design, the population, the sample size, and the context on the same card as the finding. Five studies agreeing means something very different when four of them used the same undergraduate cohort.

Capture the quote you will actually use: Pull the one sentence you would cite, with its page number, at the moment you read it. Reopening a PDF three months later to hunt for a half-remembered line is the single most avoidable time sink in the whole process.

Write the inclusion and exclusion criteria down: Which databases, which search strings, which date range, which languages, and the reason a paper was excluded. Keeping this on the board makes the review reproducible, which matters for a systematic review and is a fair question for any review.

Cluster until the themes name themselves: Move cards near the cards they resemble and stop when the groups feel stable. Then name each group. A theme arrived at this way survives contact with your data, because it came from the sources rather than from a heading you wrote before reading them.

Separate consensus from contested ground: Some claims are settled and need one sentence and five citations. Others are genuinely disputed and deserve a paragraph each side. Marking which is which stops you spending equal words on a footnote and a live controversy.

Draw the contradictions and ask why: When two studies disagree, the useful question is what differs: the sample, the measure, the decade, the definition of the construct. Put that reason on the connection between the cards. Explaining a disagreement is a contribution in itself.

Spot the schools of thought: Fields cluster around theoretical camps that cite within themselves and rarely across. When you lay the citation relationships out, those camps show up as dense regions with thin links between them, which is a structure worth naming in the chapter.

Chase citations backwards and forwards: Backward chasing means mining the reference list of a key paper for what it was built on. Forward chasing means finding everything that has cited it since, which is how you reach the recent work no keyword search surfaced. Two or three rounds of both usually beats another afternoon of database queries.

Watch for saturation: When new papers stop introducing new themes and only add citations to clusters you already have, your search has reached saturation for that question. On a canvas that moment is visible: every new card lands in an existing pile.

Look for the column that does not exist: The gap is rarely a topic nobody wrote about. It is usually a population nobody sampled, a method nobody applied to a well-studied question, a context outside the countries everyone studied, or two literatures nobody has connected. All four look like an empty region on a laid-out board.

Point your own question at the empty space: Once the gap is visible, place your research question in it and connect it back to the clusters that make it necessary. That connection is the paragraph the whole review exists to earn.

It clusters and surfaces themes from what you put there: Because the AI has your whole canvas as context, you can ask it to group the cards you have written into candidate themes, to suggest which papers look like they belong together, or to point out two cards whose claims appear to conflict. It works from your notes on the board, which means the quality of what it gives back depends on the quality of what you captured.

It does not write your review, and you should say when you used it: The argument about a field has to be yours, because the argument is the assessed part. Storyflow is a planning canvas rather than a prose generator. Most universities now require disclosure of AI use in submitted work, and policies differ sharply by institution and by department, so check yours and declare what you used.

Verify every citation yourself: Never let a language model supply a reference. Fabricated citations that look completely plausible are a well-documented failure mode, and an invented source in a review is an academic integrity problem regardless of intent. Every entry in your bibliography should come from a paper you have actually opened.

Re-cluster without losing your notes: Move a card into a different theme and everything you wrote on it moves too, including the method, the quote, and the attached PDF. Trying a different structure for the chapter costs a drag rather than a rewrite.