Teaching · Practical guide
Build a Literature Review from Verified Evidence
Use a review matrix to compare studies, preserve disagreement and draft a synthesis that can be traced back to the original sources.
Define the review before collecting summaries
A literature review answers a question by comparing relevant work. It is not a list of polished summaries. Write the question, scope and selection criteria before asking AI to organise material. Decide which contexts, methods, dates and publication types matter, and explain any exclusions.
A short teaching exercise is not automatically a systematic review. Do not label it systematic unless its search, selection and reporting meet the relevant methodological requirements. This lesson develops a transparent small-scale review workflow. AI can assist organisation, but the reviewer remains responsible for what is included, omitted and concluded.
Build a matrix from sources you have inspected
Create columns for source identifier, question, context, method, outcome, limitations and relevant passage location. Enter only what the source reports. Distinguish a study’s finding from an author’s interpretation and from your own inference. Record “not reported” when the material does not specify a detail.
Use approved excerpts or your verified notes with an AI assistant, respecting access and sharing conditions. Ask it to suggest themes across the matrix rather than invent missing cells. A theme is a proposal to test against the sources. Similar vocabulary does not establish that two studies measured the same outcome.
Worked example: synthesise without inventing evidence
The following three source cards are fictional practice material, not published studies. Card A: adult learners report that short feedback is easier to read; no learning test is included. Card B: a small class improves on a second calculation task after receiving a worked example; there is no comparison group. Card C: two groups complete different tasks, and the group receiving hints finishes faster; task difficulty is not controlled.
An unsupported synthesis is “Research proves AI feedback improves achievement.” None of these cards establishes that claim, and the cards do not even specify AI. A defensible synthesis is “The practice cards describe readability preferences, improvement across repeated tasks and completion time. Their different outcomes and design limitations prevent a direct conclusion about which feedback approach causes better learning.”
Notice the change: the revised paragraph compares evidence rather than counting positive-sounding statements. It preserves what each card can contribute while refusing to merge unlike outcomes into a single conclusion.
Organise by questions and themes
A useful structure might separate feedback clarity, independent performance and implementation conditions. Within each theme, compare methods and contexts before discussing agreement. Two papers can appear to disagree because one measures satisfaction and the other measures delayed performance. Explain that difference instead of forcing a winner.
Avoid “most studies show” unless you have a defined set and have actually counted the relevant findings using a defensible rule. A set of sources chosen because they support your position is not a neutral sample. Keep an excluded-source record for decisions that materially affect the review’s scope.
Audit every synthesis sentence
For each sentence in the draft, ask which matrix rows support it and whether the wording is stronger than the evidence. “Was associated with” and “caused” are different claims. “In this setting” and “for all learners” are different scopes. A generated rewrite may silently strengthen both.
Verify citations against the original records and passages. Retain disagreement and uncertainty rather than treating them as writing defects. ACRL’s information-literacy framework is relevant to evaluating authority and research processes; it is not a substitute for the disciplinary methods required by your actual review. Seek a librarian or supervisor’s guidance where appropriate.
Practice: a 20-minute mini-synthesis
Spend five minutes entering the three fictional cards above into a matrix. Spend five minutes selecting two themes. Use five minutes to draft a paragraph under 120 words, then five minutes to map each sentence to its supporting card and revise overclaims.
Example answer: “The cards address different aspects of feedback. A concerns readability, B concerns performance across repeated tasks, and C concerns completion time. None provides a controlled comparison of the same learning outcome. Together they suggest questions for further investigation, including how feedback affects independent performance when task difficulty and prior knowledge are considered.” This is a synthesis of the supplied exercise, not a research finding.
Self-review: are all source details traceable? Does the paragraph compare evidence rather than merely listing titles? Have you preserved limitations and avoided calling the exercise a systematic review? Save the matrix alongside the draft so later corrections can be made at the evidence level.
Reusable prompt
Select and copy this template, or customise your own using the free builder.
Using only this verified review matrix [matrix], propose themes and a cautious synthesis. Distinguish context, method, outcome and limitations. Link each proposed claim to matrix rows, flag disagreements and missing information, and do not invent citations or counts. Do not call the review systematic or infer causation unless the supplied evidence justifies that wording.Open the prompt builder →
Pause · Think · Check
Test your understanding
Choose one answer per question, then check your reasoning. This is practice; scores are not saved.