Conducting a Literature Review with AI

A literature review is not a list of summaries. It is a synthesis that maps a debate, identifies tensions, and positions your own work within an ongoing conversation. The goal of this lesson is to walk you through a complete, AI-assisted literature review workflow — one you can run for an undergraduate research paper, a master's project, or the foundations chapter of a thesis.

Crucially, the AI does none of the synthesis work. It speeds up search, helps with summarization, and supports your thinking. You build the argument.

What You'll Learn

• The four phases of a literature review and which AI tools support each
• A start-to-finish workflow you can copy for any review
• How to spot weak literature review writing (so yours is not weak)
• How to synthesize across papers without plagiarizing

The Four Phases

Every literature review, big or small, has the same four phases.

1. Scoping. Define your question and the boundary of the literature. What time period? What disciplines? What population or context?

2. Searching. Find the relevant papers. AI tools shine here.

3. Reading and extracting. Read each paper closely. Pull out the question, method, sample, finding, and limitation. AI tools help with first-pass triage.

4. Synthesizing. This is where you actually write the review. AI cannot do this for you — but it can help you stress-test your synthesis.

We will walk through each phase below.

Phase 1: Scoping

Before opening any tool, write down on paper (or in a Google Doc) the answers to these questions:

• What is my research question, in one sentence?
• What discipline am I in (so I know what counts as a "good" journal)?
• What is the relevant time range? (For most fields, "the last 10 years" is a safe default for recent work; for foundational papers, longer.)
• What population / context am I interested in?
• What kinds of papers count? Peer-reviewed journal articles only? Or also reviews, book chapters, conference papers, preprints?
• How many papers do I aim to include? (For an undergrad paper, often 8–15; for a master's dissertation, 30–60; for a PhD lit review, 100+.)

This scoping document is the constitution of your review. Without it, you will drift.

You can stress-test your scoping with AI:

Act as a senior researcher in [discipline]. Here is my research question and scope: [paste]. Is the scope too broad, too narrow, or appropriate for a [word count] literature review at [level]? Suggest one way to narrow it and one way to broaden it, with the trade-offs of each.

Phase 2: Searching

This is where you use the tools from the previous two lessons.

1. Start with Elicit or Consensus. Type your question. Get a synthesis and 8–15 candidate papers.
2. Cross-check with Perplexity in Academic mode. Catch papers Elicit missed.
3. Pick one or two strong "seed papers" from those. Open them in Semantic Scholar.
4. Generate a Connected Papers graph. Identify foundational, recent, and review papers in the neighborhood.
5. Save everything to Zotero. Tag each paper with the search query that surfaced it (useful for documenting your methodology later).

At the end of this phase you should have 20–40 candidate papers. You will not include all of them. Most reviews end up using one-third to one-half of what you initially gathered.

Phase 3: Reading and Extracting

For each candidate paper, do not read every word in order. Do this instead:

1. Read the abstract.
2. Read the introduction (especially the last paragraph, where the contribution is stated).
3. Read the conclusion.
4. Skim the methods and results.
5. Read the limitations section.

This takes about 15 minutes per paper. After this, you decide whether to invest another hour reading the paper closely.

For papers worth deeper reading, use NotebookLM or Claude to help extract a structured summary. Try this prompt with a PDF uploaded:

Read this paper. Produce a structured summary with these fields, each in 1–2 sentences and quoting one specific sentence from the paper:

• Research question
• Theoretical framing
• Method
• Sample
• Main finding
• Most important limitation
• One quote that captures the authors' central claim

Do not invent anything. If the paper does not address a field, say "not specified in paper."

Save these structured summaries in a spreadsheet — one row per paper, one column per field. This is your synthesis matrix. It is the single most valuable artifact of any literature review.

Phase 4: Synthesizing

This is the part the AI cannot do for you. Synthesis means identifying patterns, agreements, disagreements, and gaps across the papers — and constructing an argument about the state of the field.

Open your synthesis matrix. Look across rows. Ask:

• Which papers agree with each other? Group them.
• Which papers disagree? On what specifically — the finding, the method, the interpretation?
• What populations / contexts are over-studied? Which are under-studied?
• What methods dominate the field? Are there methodological monocultures?
• What gap can you legitimately claim to address?

From these patterns, draft an outline:

1. Introduction: state the question and why it matters.
2. [Theme 1]: e.g., the dominant finding and the papers that support it.
3. [Theme 2]: contested findings, divergent results, or methodological disputes.
4. [Theme 3]: under-explored questions or populations.
5. Synthesis: what we can conclude, and what remains unknown.
6. Positioning: how your study fits into this landscape.

Now write the prose yourself. Each paragraph weaves together two to four papers, comparing and contrasting. Use AI to stress-test the prose, not to generate it:

Below is a paragraph from my literature review. Identify any claims that are unsupported by a citation, any over-generalizations, and any places where I have summarized a paper without engaging with it critically. Suggest concrete improvements but do not rewrite. [paste paragraph]

This kind of prompt makes your writing stronger without writing it for you.

Spotting Weak Literature Review Writing

A literature review is weak when it reads like a list of summaries: "Smith (2020) found X. Patel (2021) argued Y. Chen (2022) showed Z."

A literature review is strong when it weaves: "Two competing accounts of X have emerged. The first, advanced most clearly by Smith (2020) and replicated by Patel (2021), argues that X is the result of Y. A second tradition, beginning with Chen (2022), challenges this on methodological grounds, noting that Smith's sample was drawn from..."

Read the difference. The weak version is a series of standalone bullets. The strong version is an argument.

This is what makes synthesis hard — and why it is the part of research that AI cannot easily replace.

Avoiding Plagiarism in Synthesis

Even with AI helping you summarize, there is a real risk of inadvertent plagiarism if you paste AI summaries directly into your draft. Two rules:

1. Always rewrite in your own voice. AI summaries often track the original paper's phrasing closely. Your synthesis must use your phrasing, not the AI's reformulation of theirs.
2. Cite every claim. Even when paraphrasing, cite the source. "As Smith (2020) argues..." or "(Smith, 2020)" depending on style.

A literature review with too few citations is suspicious. A literature review with too many citations and not enough synthesis is shallow. Aim for two to four citations per paragraph, weaving them into an argument.

A Quick Exercise

For a current paper or project, scope your question on paper (use the questions in Phase 1). Then run one Elicit search. Pull the top three papers. Read each abstract. Use your AI tool to extract the structured summary fields. Put them in a three-row spreadsheet.

You have just done the first hour of a literature review. Notice how much faster, and yet how much real reading and thinking still lies ahead.

Key Takeaways

• A literature review has four phases: scoping, searching, reading/extracting, and synthesizing.
• AI tools accelerate the first three phases dramatically; the fourth — synthesis — is yours alone.
• Build a synthesis matrix (spreadsheet of papers vs key fields) early. It is the most valuable artifact in your review.
• Weak reviews read like a list of summaries; strong reviews weave papers into an argument. Aim for the latter.
• Always rewrite AI-extracted summaries in your own voice and cite every claim.