Plan the review

Define the purpose, boundaries, review type, research question, audience, and expected contribution before asking AI to synthesize anything.

Organize evidence

Build a literature matrix so that sources can be compared systematically rather than summarized loosely.

Use AI critically

Ask focused questions, test classifications, challenge overclaims, and revise section by section.

Produce stronger synthesis

Move from summary to argument by identifying tensions, gaps, disagreements, assumptions, and implications.

Rule 1

Do not start with writing

Start with the purpose of the review, the type of review, the key concepts, the boundaries of the field, and your own position.

Rule 2

Do not trust fluency

AI-generated text can sound confident even when papers are misplaced, claims are exaggerated, or important distinctions are missing.

Rule 3

Keep a human audit trail

Record why papers are included, how they are classified, which claims they support, and where your interpretation differs from the AI output.

Step 1

Define the review task before uploading documents

Do not begin by asking the AI to write a review. Begin by defining the task. A literature review for a book chapter, a systematic review, a scoping review, a doctoral thesis, or an evaluation report will require different levels of evidence, structure, and argument.

Example prompt:

I am preparing a literature review on [topic]. The purpose is [purpose]. The audience is [audience]. My current argument is [argument]. Before writing anything, help me turn this into a review brief with clear boundaries, key concepts, inclusion criteria, exclusion criteria, and possible review streams.

Do not move on until you can explain what the review is for and what it is not for.

Step 2

Prepare your document collection

AI tools work better when the source material is organized. Before asking for synthesis, clean and structure your document collection. This reduces confusion, duplication, and weak retrieval.

Example prompt:

I will provide a list of papers. Help me create a document organization plan. Separate the papers into core sources, supporting sources, background sources, and sources that may need careful checking. Do not synthesize yet.

Step 3

Build a literature matrix before asking for synthesis

A literature matrix is the bridge between reading and synthesis. It helps you and the AI compare papers across the same dimensions instead of relying on a vague summary of each article.

Example prompt:

For each paper I provide, extract the following fields: citation, type of paper, method, role of AI, main claim, evidence strength, possible stream, limitations, and relevance to my review argument. Use cautious language and mark uncertainty clearly.

Step 4

Create provisional review streams

Review streams are not just categories. They are interpretive choices. The AI can help propose streams, but you should decide whether the categories make sense for your argument.

Example prompt:

Based on this literature matrix, propose 4 to 7 possible review streams. For each stream, define the inclusion criteria, exclusion criteria, borderline cases, and what kind of argument the stream could support. Do not force every paper into a stream.

Practical check: if a paper can only fit by stretching the category, mark it as borderline rather than forcing it.

Step 5

Ask the AI to classify papers, then challenge the classification

Classification is where many AI-supported reviews go wrong. A paper may mention dialogue but still be coding-centred. A paper may discuss AI collaboration but offer only a speculative argument. A paper may use qualitative data without being methodologically strong as qualitative research.

Example prompt:

Classify these papers into the proposed streams. For each paper, give: primary stream, possible secondary stream, confidence level, evidence for the classification, reason it might be misplaced, and whether it is a strong or weak example of the stream.

Follow-up prompt:

Now challenge your own classification. Which papers are borderline? Which ones have been classified mainly because of vocabulary rather than method? Which streams are too broad and need to be split?

Step 6

Move from summary to synthesis

AI tools often produce summaries when what you need is synthesis. Summary tells the reader what each paper says. Synthesis tells the reader how the literature fits together, where it disagrees, what it assumes, and what it still cannot explain.

Example prompt:

This section is too descriptive. Rewrite it as synthesis. Organize the paragraph around tensions, differences, assumptions, and implications across the literature. Do not list papers one by one unless necessary.

Step 7

Draft one section at a time

Do not ask AI to write the entire literature review at once. Work section by section. For each section, give the AI the relevant papers, your intended argument, the role of the section, and the claims that must be handled carefully.

Example prompt:

Draft a section on [stream name]. Use only the papers listed below. The purpose of the section is to show [purpose]. My argument is [argument]. Do not overstate the evidence. Distinguish empirical findings from conceptual claims. End by explaining how this stream leads to the next section.

Step 8

Review the AI draft like a critical reader

The most important work begins after the AI produces a draft. Read it slowly. Look for places where the prose is smooth but the argument is weak. Mark claims that are too broad, references that are misplaced, and sections where your own voice disappears.

Example prompt:

Read this draft section as a skeptical reviewer. Identify misplaced papers, unsupported claims, overstatements, missing distinctions, weak transitions, and places where my own position is absent. Then suggest specific revisions before rewriting anything.

Step 9

Make your own position visible

Literature reviews are not neutral catalogues. They are arguments. The reader should understand how you interpret the field, what you see as the main problem, and why your contribution matters.

Example prompt:

Revise this section so that my position is clearer. The position is: [insert position]. Keep the literature central, but make the argument more visible. Avoid sounding as if the review is only reporting what others have said.

Step 10

Run a final evidence audit

Before you treat the review as finished, audit every major claim. Ask whether the claim is supported, whether the evidence is strong enough, and whether the wording is appropriately cautious.

Example prompt:

Audit this section. Create a table with each major claim, the papers used to support it, the type of evidence, the strength of the evidence, possible counterarguments, and whether the wording should be softened.

Weak AI-style paragraph

Several studies show that AI can be used in qualitative research. Some authors use AI for coding, while others use it for analysis. Overall, the literature suggests that AI has many benefits for qualitative researchers, including speed, efficiency, and improved insight.

Stronger synthesis

The literature does not present a single view of AI in qualitative research. Coding-centred studies tend to frame AI as a tool for efficiency and pattern recognition, while more dialogic approaches treat AI as part of an interpretive exchange. This distinction matters because the methodological question is not only whether AI speeds up analysis, but how it changes the researcher’s role in interpretation, reflexivity, and judgment.

1. Review brief prompt

Help me create a review brief for a literature review on [topic]. Include purpose, audience, review type, research question, key concepts, inclusion criteria, exclusion criteria, likely streams, and risks of misclassification.

2. Literature matrix prompt

Extract a literature matrix from the following papers. Use the columns: citation, paper type, method, role of AI, main claim, evidence strength, limitations, possible stream, and relevance to my argument.

3. Stream-building prompt

Propose possible streams for this literature. For each stream, define what belongs, what does not belong, which papers are strong examples, and which papers are borderline.

4. Classification challenge prompt

Challenge your own classification. Which papers might be misplaced? Which categories are too broad? Which papers are being grouped together because of shared vocabulary rather than shared methodology?

5. Synthesis prompt

Turn this descriptive section into synthesis. Organize the discussion around tensions, differences, assumptions, and implications. Avoid listing papers one by one. Make the argument clearer.

6. Evidence audit prompt

Audit this section. For each major claim, identify the supporting source, evidence type, evidence strength, possible counterargument, and whether the wording should be softened.

NotebookLM for researchers

NotebookLM for researchers can be useful for source-grounded summaries, extracting key claims, asking questions across documents, and checking where information appears in the source material.

Literature review with ChatGPT and Claude

A literature review with ChatGPT or a Claude literature review workflow can be useful for developing review frames, comparing interpretations, drafting sections, challenging categories, and improving synthesis.

Literature matrix tutorial workflow

A literature matrix tutorial workflow helps researchers code papers, track decisions, record evidence strength, and keep an audit trail before moving into synthesis.

Reference managers

Useful for organizing citations, checking bibliographic details, managing PDFs, and keeping the review traceable.

Avoid this

• Uploading 100 PDFs and asking for a full review in one prompt.
• Assuming the AI has attended equally to every document.
• Accepting fluent prose as evidence of good synthesis.
• Letting the AI decide the conceptual frame alone.
• Using AI output without checking the sources.
• Treating the literature review as a writing task only.

Do this instead

• Create a review brief before drafting.
• Build a literature matrix.
• Use focused prompts for specific analytical tasks.
• Ask the AI to show uncertainty and borderline cases.
• Revise section by section.
• Use your expertise to evaluate, correct, and redirect the output.

Can AI write a literature review for me?

AI can help draft, organize, summarize, compare, and revise. But it cannot replace the researcher’s expertise, judgment, or responsibility for the final argument.

Is uploading many PDFs enough?

No. Giving an AI tool access to many documents does not guarantee equal or accurate attention to all of them. A structured matrix and focused prompts are safer.

What is the biggest risk?

The biggest risk is accepting a fluent but conceptually weak review. AI can make uncertain classifications sound confident and can smooth over important tensions.

What is the best way to use AI for literature reviews?

Use AI iteratively: define the review purpose, build a literature matrix, classify cautiously, challenge classifications, draft section by section, and audit every major claim.