Literature Review Examples: 5 Worked Cases + 7 Format Samples (2026)

Quick reference

A literature review synthesizes existing research on a topic. The format depends on your goal: standalone literature review article, dissertation chapter, systematic review, or grant proposal background. Below are seven examples from real published research, with sample structure, length expectations, and the citation visualization tools that produced their reference networks.

What you'll find on this page

Five detailed worked examples (one per field), seven format samples (one per review type), a side-by-side field-differences table, common mistakes, an FAQ, and a step-by-step workflow for using citation maps to build the seed-author shortlist your review will be organised around. Skim, jump, or read top to bottom.

A5 worked examples by field
BWhat differs by field
1Standalone Literature Review (journal article)
2PhD Dissertation Chapter
3Systematic Review (PRISMA)
4Grant Proposal Background Section
5Master's Thesis Chapter
6Conference Paper Related Work
7Annotated Bibliography

Five worked literature review examples by field

Each of these is a sketch of how a strong literature review on a real question in that field would be organised, what the synthesis claim could look like, where the gap typically lives, and which theoretical frame anchors the argument. Paper titles and venues below are real and verifiable. The structures are illustrative — actual published reviews vary — but they reflect the conventions reviewers and committees in each field expect.

Worked example · Machine Learning (computer science)

How self-supervised pretraining changed representation learning (2018–2025).

Anchor papers (illustrative): BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding — Devlin et al., NAACL 2019.
A Simple Framework for Contrastive Learning of Visual Representations (SimCLR) — Chen et al., ICML 2020.
Masked Autoencoders Are Scalable Vision Learners — He et al., CVPR 2022.
LLaMA: Open and Efficient Foundation Language Models — Touvron et al., arXiv 2023.
Sample structure: 1. The supervised-only era (one paragraph, ~5 cites)Brief recap of why ImageNet-style supervised pretraining dominated 2012–2017, and the labelling-cost ceiling researchers ran into.
2. Two parallel threads emerge (3 sub-themes, ~15 cites)Masked-language-modeling (BERT family) in NLP; contrastive learning (SimCLR, MoCo, BYOL) in vision. Discuss each thread's loss formulation and why the field treated them as separate problems.
3. The convergence (~10 cites)Masked autoencoders (MAE), BEiT, and the realization that masking is contrastive learning's twin. Note that this synthesis happened in print only after 2022 — a discoverable history.
4. Scaling laws and foundation models (~10 cites)Chinchilla, GPT-3, LLaMA. Position pretraining as compute-bound rather than method-bound by 2024.
5. Open problems (~5 cites)Sample efficiency on long-tail classes, modality alignment, evaluation under distribution shift.
Synthesis claim: A strong synthesis here doesn't just list models in chronological order — it argues that vision and language solved the same problem in parallel and only later admitted as much, and that the unresolved question by 2025 is no longer 'how do we pretrain' but 'what counts as a sufficient pretraining signal for embodied or multimodal agents'. Reviewers in this field reject reviews that read like a leaderboard.
Where the gap typically lives: Most ML self-supervised reviews under-cite work outside the FAANG / top-10 university orbit. A geographic citation map reveals strong contributing groups in Tsinghua, KAIST, EPFL, INRIA, and Israeli universities that English-language conference proceedings cover unevenly.
Theoretical framing: The theoretical frame is information-theoretic: pretraining as compression of high-mutual-information statistics. Citing Tishby's Information Bottleneck papers and the Kolmogorov-complexity-style framings (Hinton, Sutskever) anchors the review to a conceptual lineage rather than a benchmark race.
Where a citation map adds value: Use a citation map for one or two seed authors per thread (e.g. Hinton for contrastive, Devlin for masked LM) and compare the geographic footprints — the overlap reveals which institutions bridge the two threads, which is exactly the synthesis story your review wants to tell.

Worked example · Biology (CRISPR / gene editing)

Off-target effects in CRISPR-Cas9 therapeutic editing (2014–2025).

Anchor papers (illustrative): A Programmable Dual-RNA-Guided DNA Endonuclease in Adaptive Bacterial Immunity — Jinek, Chylinski, Doudna, Charpentier et al., Science 2012.
Genome-wide specificities of CRISPR-Cas Cpf1 nucleases in human cells — Kim et al., Nature Biotechnology 2016.
Search-and-replace genome editing without double-strand breaks or donor DNA (Prime Editing) — Anzalone, Liu et al., Nature 2019.
Casgevy (exa-cel) BLA approval for sickle cell disease — Frangoul et al. (the underlying CLIMB SCD-121 trial), NEJM 2021.
Sample structure: 1. Foundations and mechanism (~10 cites)The 2012 Jinek/Doudna and Cong/Zhang papers establishing programmable editing; Mojica's earlier microbiology work that pre-dated the therapeutic interest.
2. The off-target problem comes into focus (3 sub-themes, ~25 cites)GUIDE-seq and CIRCLE-seq for unbiased off-target detection; computational off-target predictors (CRISPOR, Cas-OFFinder); chromatin and accessibility effects.
3. Engineering safer enzymes (~15 cites)High-fidelity SpCas9 variants (eSpCas9, SpCas9-HF1, HiFi), Cas12a/Cpf1, base editors, prime editors. Position each as a different trade-off between PAM flexibility, edit type, and off-target rate.
4. Translation to the clinic (~10 cites)ex vivo edits (Casgevy/CTX001 for sickle cell), in vivo LNP delivery (intellia NTLA-2001 for ATTR amyloidosis). Discuss how off-target tolerance shifts when the edit is in millions of cells in a patient rather than a single cell line.
5. Open regulatory and ethical questions (~5 cites)FDA and EMA guidance; germline editing moratoria; the He Jiankui case as a cautionary anchor.
Synthesis claim: The synthesis argument: off-target risk is not one problem but four — sequence-similar mismatches, chromosomal rearrangements, large deletions, and bystander edits — each requiring different detection assays. Reviews that treat 'off-target' as a single metric miss this. The clinical translation has succeeded so far because hematopoietic editing tolerates a higher mutation load than germline or somatic editing in solid tissue.
Where the gap typically lives: Reviews dominated by Broad / UC Berkeley / Boston-area labs. Strong work on alternative Cas enzymes from China (notably Yang Hui's group at CAS) and Korea (Jin-Soo Kim) is consistently under-cited. A citation map for Doudna or Zhang side-by-side with a citation map for Jin-Soo Kim immediately exposes which clusters cite outward and which don't.
Theoretical framing: Theoretical framing draws on adaptive immunity (CRISPR's biological origin), thermodynamic specificity (mismatch tolerance as a function of guide-RNA binding energy), and translational science framing (analogy to early ASO and AAV timelines). Citing Mojica's 2005 paper and Marraffini's later mechanistic work signals you understand the basic-science lineage.
Where a citation map adds value: Geographic mapping is unusually valuable here because regulatory regimes shape which edits get tested where. Mapping citing institutions for the first 20 in-human CRISPR trials surfaces a US/UK clinical concentration with mostly-Asian basic-science feeders — that bifurcation is itself a finding worth a paragraph.

Worked example · History (humanities)

The historiography of the British East India Company's transition from trade to sovereignty, 1757–1857.

Anchor papers (illustrative): The New Cambridge History of India series (multiple volumes, 1987–present) — Various; Bayly, Marshall, Washbrook eds., Cambridge University Press.
The Anarchy: The Relentless Rise of the East India Company — William Dalrymple, Bloomsbury 2019.
Empire, Incorporated: The Corporations That Built British Colonialism — Philip J. Stern, Harvard University Press 2023.
Sample structure: 1. Periodization and stakes (~5 cites)Why 1757 (Plassey) and 1857 (the Rebellion / dissolution of Company rule) bracket the question. Briefly note the older 1600 charter as the deeper origin.
2. The 'imperial historiography' tradition (~10 cites)Vincent Smith, the early Cambridge history. Read sympathetically as a primary source for what 19th- and early-20th-century British historians took for granted, not as authoritative analysis.
3. The Cambridge School and the 'collaboration' thesis (~10 cites)Bayly, Washbrook, Robinson. The argument that colonial rule depended on indigenous intermediaries — moneylenders, sepoys, scribes — and that the Company was structurally weaker than older accounts assumed.
4. Subaltern Studies and the response (~10 cites)Guha, Chakrabarty, Spivak. Recovering peasant and tribal agency, and the methodological critique that elite archives systematically erase subaltern voices.
5. The corporate / financial turn (~10 cites)Recent work (Stern, Erikson, Dalrymple's popular synthesis) reframing the Company as a chartered corporation and asking what corporate-form analysis adds to imperial history.
6. Open questions (~5 cites)Climate, ecology, and the role of South Asian credit networks. The post-2015 literature that takes Mughal-era fiscal continuity seriously.
Synthesis claim: Humanities reviews are thematic and historiographic, not chronological. The synthesis here is that each generation of historians has answered 'what was the Company?' differently — moral-imperial, structural-collaborative, subaltern-resistant, corporate — and the most defensible position by 2025 is that the Company was simultaneously all four, and the historian's job is to choose which lens is sharpest for which sub-question. That meta-claim is the contribution.
Where the gap typically lives: Anglophone scholarship still dominates this field; serious work in Bengali, Marathi, Persian, and Urdu archives is under-translated and under-cited. A literature review that admits this and gestures toward (e.g.) Rajat Kanta Ray, Tirthankar Roy, or Sumit Sarkar's work in vernacular venues is more honest than one that doesn't.
Theoretical framing: Theoretical frames are postcolonial theory (Said, Chakrabarty), Marxist economic history (Hobsbawm, Bayly), and corporate-law history (Stern). State which frame your review adopts; treat the others sympathetically as competing readings, not as wrong answers.
Where a citation map adds value: For humanities, citation maps are most useful as a check on linguistic and institutional reach: are you citing UK Russell Group, Indian Institutes of Advanced Study, US South Asian Studies departments, and continental European area-studies centers, or only the first? Geographic visualization makes that audit five-second fast.

Worked example · Economics (social sciences)

The minimum-wage employment debate after the Card-Krueger 1994 paper.

Anchor papers (illustrative): Minimum Wages and Employment: A Case Study of the Fast-Food Industry in New Jersey and Pennsylvania — Card and Krueger, American Economic Review 1994.
Myth and Measurement: The New Economics of the Minimum Wage — Card and Krueger, Princeton University Press 1995.
Minimum Wage Effects Across State Borders: Estimates Using Contiguous Counties — Dube, Lester, Reich, Review of Economics and Statistics 2010.
The Effect of Minimum Wages on Low-Wage Jobs — Cengiz, Dube, Lindner, Zipperer, Quarterly Journal of Economics 2019.
Sample structure: 1. Theoretical priors (~5 cites)The textbook competitive-market prediction: a binding minimum wage reduces employment. Stigler 1946 as the canonical statement.
2. The Card-Krueger natural experiment and the immediate response (~10 cites)The 1994 paper, the Neumark-Wascher 2000 reanalysis using payroll data, and the methodological debate about telephone-survey vs administrative data.
3. The 'border discontinuity' generation (~10 cites)Dube-Lester-Reich's contiguous-county design, and the broader rise of differences-in-differences as a credibility-revolution method (Angrist, Pischke).
4. The bunching estimator and the modern consensus (~10 cites)Cengiz-Dube-Lindner-Zipperer 2019 reframing the question from 'employment elasticity' to 'where in the wage distribution does the mass shift?' — and finding small disemployment effects concentrated in narrow sub-groups.
5. Heterogeneity and the mechanism debate (~10 cites)Monopsony explanations (Manning, Naidu); price pass-through; firm-entry margins. Discuss why a single elasticity number is no longer the field's target.
6. Policy translation and open questions (~5 cites)$15 federal minimum wage proposals; subnational experimentation; spillover into informal labor markets in developing countries.
Synthesis claim: The synthesis: the question 'do minimum wages reduce employment?' has been productively replaced by 'under what labor-market structure, at what bite, and on what time horizon?' A defensible 2025 review argues that monopsony is the dominant structural assumption supported by current evidence, that effects are heterogeneous and small in aggregate at recent US bites, and that the policy frontier is now low- and middle-income countries where the canonical evidence base is much thinner.
Where the gap typically lives: Most cited evidence is from the US and Western Europe. Africa, South Asia, and Latin America have very different labor-market structures and a much sparser literature. A review that flags this geographic gap explicitly — and cites e.g. World Bank or IZA discussion papers from those regions — distinguishes itself.
Theoretical framing: Theoretical framing: monopsony / search-and-matching labor markets (Manning, Card-Cardoso-Heining-Kline); the credibility revolution methodology (Angrist-Pischke); welfare analysis with heterogeneous workers (Saez-Schoefer-Seim).
Where a citation map adds value: An economics literature review benefits especially from the geographic check — labor-market evidence is jurisdiction-specific. A citation map of the top 5 minimum-wage papers shows the US dominates, with secondary clusters in the UK and Germany, and a striking absence of Latin American and African empirical work in top journals despite the policy interest there.

Worked example · Education (social sciences)

Cognitive load theory in classroom instructional design (1988–present).

Anchor papers (illustrative): Cognitive load during problem solving: Effects on learning — John Sweller, Cognitive Science 1988.
Cognitive architecture and instructional design — Sweller, van Merriënboer, Paas, Educational Psychology Review 1998.
Cognitive Load Theory: A Handbook for Teachers (and the wider 'evidence-informed teaching' movement) — Lovell (2020) and the broader CLT-applied literature, John Catt Educational.
Sample structure: 1. Theoretical foundation (~5 cites)Sweller 1988; the working-memory model (Baddeley 1986/2000) on which CLT rests.
2. The three loads and their evolution (~10 cites)Intrinsic, extraneous, germane load. Discuss the 2010s reconsideration of 'germane load' as a separable construct (Kalyuga, Sweller's own retrospectives).
3. Empirical instructional effects (~15 cites)Worked-example effect, expertise reversal, redundancy effect, split-attention effect, modality effect. Group these as predictions of the theory and review the meta-analytic evidence (de Jong 2010; Hattie's syntheses).
4. Translation into classroom practice (~10 cites)The post-2015 'cognitive science in the classroom' movement (Willingham, Christodoulou, the EEF guidance reports). Position this as a translation literature, distinct from primary CLT research.
5. Critiques and limits (~5 cites)The constructivist critique (Kirschner-Sweller-Clark 2006 vs. responses); ecological-validity concerns; cultural and developmental boundary conditions.
Synthesis claim: The synthesis argument: CLT has moved from a laboratory theory of working memory to a generative framework for classroom guidance, but the field has not yet adequately addressed how its predictions interact with motivation, prior knowledge, and cultural learning norms. Reviews that treat CLT as a closed paradigm (or as a settled debate against constructivism) misread the current state.
Where the gap typically lives: Most CLT empirical work is from Australian, Dutch, US, and UK research groups. There is a fast-growing East Asian literature (mainland China, Hong Kong, Singapore, Taiwan) that tests CLT predictions in mathematics-heavy curricula, and the Anglophone reviews under-engage with it. A review that surfaces this is unusually defensible.
Theoretical framing: Theoretical framing: cognitive architecture (working memory + long-term memory schemas), evolutionary educational psychology (Geary's primary/secondary knowledge distinction), and instructional design as applied science. Naming all three frames signals you can navigate the theoretical field.
Where a citation map adds value: A citation map of Sweller's profile alongside Kirschner's and Paas's quickly shows the Sydney–Eindhoven–North-American axis that defines mainstream CLT, and where you'd need to go (East Asia) to find the strongest current empirical extension.

What differs by field — STEM vs social sciences vs humanities

The biggest mistake new reviewers make is importing the conventions of one field into another. A computer-science Related Work section written like a humanities historiographic chapter reads as unfocused; a humanities review written like a CS Related Work reads as superficial. The table below names six dimensions where the conventions actually diverge.

Dimension	STEM	Social sciences	Humanities
Organizing principle	Methodological synthesis — group papers by technique or theoretical advance, often with a partial chronological backbone.	Theoretical framework first — name the conceptual lens (e.g. monopsony, cognitive load) and review empirical work as tests of the frame.	Thematic and historiographic — group by interpretive school of thought, not by date or method.
Typical length	Conference Related Work: 500–1,200 words. Journal review: 6,000–10,000 words. Methods reviews longer.	Journal review: 8,000–12,000 words. Dissertation chapter: 10,000–15,000.	Journal review: 8,000–15,000 words. Monograph chapter: 15,000–25,000. Length signals seriousness.
Reference count	30–150, heavily concentrated in past 5 years.	60–200, balanced across past 20 years; foundational theoretical citations expected.	100–400, deep historical reach common; primary sources cited alongside secondary literature.
Recency expectation	60–80% of citations from past 5 years; missing recent work is a fast reject.	Roughly 50/30/20 split across past 5, 6–15, and 16+ years.	No strong recency expectation; 20-year-old work is current if no rebuttal exists.
Voice and synthesis	Neutral, often passive; synthesis is methodological progress.	First-person 'we argue' acceptable; synthesis is theoretical refinement.	Strong authorial voice; synthesis is interpretive position-taking.
Gap statement	Empirical or methodological gap (a measurement that has not been made, a benchmark that has not been beaten).	Theoretical or contextual gap (a population, mechanism, or moderator that has not been studied).	Interpretive gap (a reading that has not been advanced, an archive that has not been engaged).

These are conventions, not laws. A landmark interdisciplinary review often breaks one column intentionally — but you need to know the convention before you can break it.

Seven format samples (by review type)

The next seven cards cover format expectations — length, structure, and citation strategy — for each common review type. Click into the one that matches your assignment.

Sample 1

Standalone Literature Review (journal article)

What it looks like: A standalone review is a self-contained journal article — usually 6,000 to 10,000 words — that surveys an entire subfield. The skeleton is almost always: abstract, introduction (the question and why it matters), methods (how you searched and screened), thematically organized findings (3–6 H2 sections each grouping related work), a synthesis or conceptual framework, gaps, and a research agenda. Headings are thematic, not chronological — readers skim by theme.
Real example: Annual Review series articles (e.g. Annual Review of Psychology, Annual Review of Public Health) are the canonical model. Most are open access via the Annual Reviews "Charting a Path" initiative or paywalled abstracts with extensive open metadata. Browse annualreviews.org.
Citation strategy: Aim for 80–150 references with depth over breadth. Cite seminal foundations (one or two from before 2000), bridge papers (5–10 from the past 10 years that connect subfields), and a heavy concentration in the past 5 years (60% or more). Citing fewer well-chosen papers and discussing each is preferred to a long, ungrouped reference dump.
How a citation map helps: A geographic map of your top 10 most-cited seed authors immediately reveals which countries dominate the conversation — and which research traditions you may have missed by reading only English-language top venues.

See a citation map for Geoffrey Hinton (ML/CS)

Sample 2

PhD Dissertation Chapter

What it looks like: Chapter 2 of nearly every PhD thesis. Length is typically 8,000 to 15,000 words and the structure mirrors the dissertation's argument: the chapter ends by motivating your specific research question. Common shape: introduction → theoretical foundations (key concepts and theories) → empirical literature organized by sub-question → methodological literature (how others have studied this) → gap analysis → chapter summary. Defended chapters are visible in institutional repositories.
Real example: Most universities publish defended dissertations openly. Search ProQuest's open-access dissertations or your institution's repository; representative examples appear in MIT DSpace, Stanford's Searchworks, and the British Library EThOS service.
Citation strategy: Expect 100–250 references — broader than a journal review because the chapter must establish your authority across the field. Balance breadth (showing you know the landscape) with depth (3–8 papers you discuss closely as direct precursors to your work). Always cite the exact paper your method or framework is adapted from, with page numbers.
How a citation map helps: Mapping the citation footprints of your top 5 seed authors helps you defend the choice of "who counts" in your subfield — committees often probe whether you've considered work outside your advisor's network. A geographic visualization is a quick honesty check.

See a citation map for Fei-Fei Li (computer vision)

Sample 3

Systematic Review (PRISMA)

What it looks like: A systematic review follows the PRISMA 2020 reporting guideline. Sections are prescriptive: title, structured abstract, introduction, methods (eligibility, info sources, search strategy, selection process, data extraction, risk of bias, synthesis methods), results (with the PRISMA flow diagram showing identified/screened/included counts), discussion, and registration in PROSPERO. Length is 5,000 to 12,000 words. Tone is neutral and reproducible — another team should be able to redo your search.
Real example: The Cochrane Library hosts thousands of open-protocol systematic reviews; PubMed Central also indexes BMJ Open and BMC systematic reviews under permissive licenses. Search pubmed.ncbi.nlm.nih.gov.
Citation strategy: Citation count is determined by the search, not by you. Typical PRISMA reviews include 20–80 studies after screening from an initial pool of hundreds or thousands. Every included study is cited; every excluded study has a recorded reason. Forward citation chasing on the included set is mandatory — you must rule out missing recent work.
How a citation map helps: Geographic maps surface region-specific publication patterns (for example, Asian-language clinical trials underrepresented in English databases). PRISMA recommends documenting language and database limitations; a citation map quantifies the geographic blind spot.

See a citation map for Jennifer Doudna (CRISPR/biology)

Sample 4

Grant Proposal Background Section

What it looks like: The literature review section of an NIH R01 "Significance" or NSF "Project Description" — usually compressed to 800 to 1,500 words. The job is to argue significance and innovation, not to survey. Structure: the unsolved problem, what is currently known (3–5 named groups and their contributions), what is missing, and how your proposal fills the gap. Reviewers spend less than 10 minutes per proposal — every sentence must justify funding.
Real example: NIH RePORTER and NSF Award Search publish funded abstracts and, for many awards, the full Project Description. Browse the NIH RePORTER abstracts to see how successful PIs structure their significance sections.
Citation strategy: 20–40 references, hand-picked for impact, not for breadth. Lead with a high-citation seminal paper, name 3–5 competing or collaborating labs by PI name, cite at least one work from a reviewer you suspect will see the proposal. Recency bias is strong — most citations should be from the past 3–5 years.
How a citation map helps: A geographic map for your closest competitors instantly answers "is the field clustered or distributed?" — a key signal reviewers use to gauge whether your niche is realistic. It also helps you spot unrelated regional groups whose collaboration could strengthen the proposal.

Find seed authors for your proposal

Sample 5

Master's Thesis Chapter

What it looks like: Lighter than a PhD chapter — typically 4,000 to 8,000 words. Often combines literature review and theoretical framework into a single chapter. Standard shape: definitional groundwork (key terms), thematic review of empirical work organized into 3 to 5 sub-sections, identification of one or two gaps, and a clear hand-off to your methodology chapter. Master's committees primarily check that you've engaged with the right literature, not that you've surveyed everything.
Real example: An exemplary version of this format appears in master's theses from STEM and social science programs deposited in open institutional repositories (search any university's library catalog for "Master's thesis" + your topic).
Citation strategy: 40–80 references is typical. Quality of synthesis matters more than count. Group references into named themes ("Theme A: cognitive load in early reading", "Theme B: classroom-level interventions") rather than discussing each paper individually — examiners want to see you can categorize, not just summarize.
How a citation map helps: If your master's program emphasizes original contribution, a citation map helps justify why your topic is unsaturated — a sparse geographic distribution is direct evidence that your question is under-studied somewhere.

Map your seed authors

Sample 7

Annotated Bibliography

What it looks like: Often confused with a literature review, but distinct: an annotated bibliography is a list of references where each entry includes a 100–200 word annotation summarizing and evaluating the source. Length is determined by the assignment — 10 to 50 entries is typical. Format: full citation in the required style (APA, MLA, Chicago), then a paragraph covering scope, methodology, key findings, and relevance to your project. No synthesis across entries.
Real example: Most university writing centers publish annotated bibliography exemplars for student reference; Purdue OWL's annotated bibliography guide includes formatted samples in MLA, APA, and Chicago styles. owl.purdue.edu.
Citation strategy: Set by the assignment. The trick is even coverage — instructors expect a mix of foundational, recent, methodological, and counter-argument sources. A common rubric is 10–15 entries spread across 3–4 themes.
How a citation map helps: When you're asked to ensure "breadth" in your bibliography, a citation map of your shortlist quickly proves you haven't accidentally selected 12 papers from the same lab or country.

Find diverse seed papers

How to use citation maps when building your literature review

Citation maps and citation-network tools are most useful in the early and mid-stages of a review — when you're scoping the field, choosing seed authors, and checking whether you've missed a research tradition. Here's the workflow that takes about 20 minutes per topic and surfaces 10–30 papers keyword search alone won't.

1
Pick 3–5 seed authors central to your topic
Choose researchers whose names appear in every key paper you've already read. Three is enough to start; five is better when your topic spans subfields. The seed authors define the citation neighborhood you're about to map.
2
Generate a citation map for each seed
Paste each scholar's Google Scholar profile URL into the search box. The geographic map renders in 2–4 seconds and shows every country and institution that has cited their published work.
3
Identify clusters and dominant labs
Look for dense city-level clusters — these are the schools of thought in your subfield. Boston + Stanford + Toronto for a deep-learning question. Cambridge + Berlin + Tokyo for a particular biology niche. The clusters tell you whose work to read next.
4
Spot the geographic gaps
Empty regions are research opportunities. If a topic has no citations from Sub-Saharan Africa, Latin America, or Southeast Asia, either the topic is genuinely region-blind or non-English work has been overlooked. Either is worth flagging in your gap analysis.
5
Cross-reference with paper-level tools
Use Connected Papers or ResearchRabbit to expand from each seed paper. A geographic map shows where research lives; a paper-similarity graph shows what each cluster is publishing. The two views together cover both axes.

The geographic dimension matters more than most students realize. When you see a topic's research clustering in three universities on the US East Coast, you're seeing a school of thought forming — and you're seeing whose papers will keep appearing in every reference list you read for the next decade. When you see clusters on three different continents that don't cite each other, you've found a real research gap: a synthesis nobody has written. Both insights change the way you frame your gap analysis. For a deeper walk-through of the geographic-map workflow specifically, see our citation map for literature review guide.

Common literature review mistakes

Eight recurring failure modes that show up in reviewer comments, committee feedback, and rejected manuscripts. The first four are the ones examiners most often write in the margin; the last four are the ones that quietly cap your work at "competent but unremarkable".

Treating it as an annotated bibliography. A literature review is an argument, not a list. If your draft reads "Smith (2021) studied X and found Y. Jones (2022) studied Z and found W" for paragraph after paragraph, you've written an annotated bibliography. Group sources under thematic headings and write topic sentences that make claims, not summaries.
Laundry-list summaries. Even with thematic grouping, a paragraph that just summarizes each cited paper in turn isn't synthesis. Synthesis means: across these five papers, what is the consistent finding, what is the contested finding, and what is the question none of them answers? If you can't name those three things in one sentence, the paragraph isn't doing review work.
Missing the gap statement. A literature review must explicitly say what the existing literature has not yet done — and your work will. Vague gap statements ("more research is needed") are a top reject reason in journal peer review. A strong gap statement names a specific population, mechanism, scale, or context that the literature has missed.
Forgetting to position your work. Late in the review, name two or three nearest existing studies, say precisely how your work differs (population, method, scale, context, theoretical lens), and explain why that difference matters. Without this paragraph, your contribution looks incremental even if it isn't.
Citing 50+ sources but only 3 from the past 5 years. The most common reject reason in fast-moving fields. Reviewers read your reference list before they read your abstract — recency imbalance is the first signal of an out-of-touch author.
Missing key authors. Lit review tools surface these — Connected Papers, ResearchRabbit, and CitationMap will all flag a 5-paper cluster you've completely failed to cite. Run at least one before submitting.
US-centric bias. Citation maps reveal this geographic blind spot at a glance. If 90% of your seed authors sit in the same time zone, your review is regional, not global — disclose the limitation or expand the search.
Not using forward citation chasing. The 5 most-recent high-impact papers in your area cite work you haven't found yet. Following their forward citations (Google Scholar "Cited by" or any network tool) routinely surfaces 10+ relevant papers in 30 minutes.

Use a citation map to build your literature review

Citation maps are not a replacement for reading — they are a way to choose what to read next. The workflow that pays for itself in about 30 minutes per review:

Identify three to five seminal papers. Start from a recent review article in your field, or from the two or three papers your advisor has named as canonical. These are your seeds.
Map the citation network for each seed's author. Drop the author's Google Scholar profile URL into CitationMap and the geographic map renders in 2–4 seconds. The map shows every institution that has cited their work and how heavily.
Spot under-cited regions and institutions. Look at the map for empty regions or surprisingly small markers in countries where you'd expect strong work. Those are candidate gaps. They are also candidate reading lists — the next 5–10 papers you should pull are usually in those clusters.
Identify gaps that survive forward chasing. A real gap is one that the recent high-citation papers also don't cover. Walk the "Cited by" chain on each seed paper for 10 minutes; if the geographic gap holds up after that walk, it's defensible to flag in your review.
Write the geographic blindspot into your gap statement. Examiners and reviewers reward specificity. "Most empirical work on X has been conducted in high-income North Atlantic settings; this review extends the question to Y context" is a defensible, audit-able claim.

Concrete starting points: Geoffrey Hinton's citation map (deep learning) or Yann LeCun's citation map (machine learning, NYU/Meta) — both render full geographic footprints you can use as templates while you build your own. For the deep walk-through, see the citation map for literature review guide.

Tools that generate citation networks for literature reviews

Five tools cover almost every literature review use case. Each is free at the tier most students need.

CitationMap — geographic citation map plus author-identity layer. Free first crawl, no signup. Best for showing where research lives globally.
Connected Papers — semantic similarity graph centered on a seed paper. Free with 5 graphs/month limit. Best for scoping a brand-new topic from one or two seed papers.
Litmaps — citation tracking with weekly digest of new citing papers. Freemium ($10/mo Pro). Best for ongoing monitoring of a watched literature set.
ResearchRabbit — paper recommendations with two-way Zotero sync. Genuinely free. Best for reading-list-driven literature exploration.
VOSviewer — desktop network visualization (requires a CSV export from Web of Science, Scopus, or PubMed). Free. Best for publication-quality co-citation network figures.

For a side-by-side ranking with pricing, free-tier limits, and the decision tree, see the full comparison of citation map tools. If you're specifically deciding between a geographic map and a paper-similarity graph, our CitationMap vs Connected Papers head-to-head walks through the trade-offs. New to citation mapping entirely? Start with the how to create a citation map tutorial.

Frequently asked questions

How long should a literature review be?+

It depends on the format. Standalone journal articles: 6,000–10,000 words. PhD dissertation chapter: 8,000–15,000 words. Systematic review (PRISMA): 5,000–12,000 words. Grant proposal background: 800–1,500 words. Master's thesis chapter: 4,000–8,000 words. Conference Related Work: 500–1,200 words. The format dictates the length — when in doubt, count words in 3 published examples in your target venue and aim for the median.

How many sources should a literature review cite?+

Roughly proportional to the format. A PhD chapter cites 100–250 references, a journal review article 80–150, a master's chapter 40–80, a conference Related Work 30–80, an NIH R01 background 20–40. A common mistake is over-citing breadth at the expense of depth — examiners and reviewers prefer 50 well-discussed papers over 200 listed in passing. The right answer is whatever count lets you build a defensible argument.

What's the difference between a literature review and a systematic review?+

A literature review is a narrative synthesis — the author selects, organizes, and interprets prior work to support an argument. A systematic review follows a pre-registered, reproducible protocol (PRISMA 2020) — explicit eligibility criteria, exhaustive search across named databases, dual-screening, risk-of-bias assessment, and a flow diagram of identified/screened/included counts. Systematic reviews aim to remove the author's selection bias; narrative literature reviews assume it. Use a systematic review when you need to answer a focused clinical or policy question; use a narrative review when you need to argue a position or motivate new research.

Should I include older foundational papers or only recent work?+

Both. The best literature reviews lean roughly 60% on the past 5 years, 30% on the past 6–15 years, and 10% on classic foundations. Citing only recent work makes you look unaware of the field's history; citing only classics makes you look out of touch with current debates. Foundational citations should appear in the introduction and theoretical-framework sections; recent work belongs in the empirical synthesis and gap analysis. Reviewers explicitly flag both extremes — "50 sources but only 3 from the past 5 years" is a common reject reason.

How do I find papers I haven't already discovered?+

Three complementary moves. First, forward citation chasing — start from your 5 most-cited seed papers and walk every paper that cites them (Google Scholar's "Cited by" link, or use Connected Papers and ResearchRabbit to expand the network visually). Second, backward chasing — read the reference lists of your seeds and follow the chains. Third, geographic mapping — a citation map exposes regional research traditions you might have missed because you searched only in English. The combination of these three usually surfaces 10–30 high-relevance papers that keyword search alone misses.

Can I use AI tools to write my literature review?+

AI writing tools can draft summaries and find papers, but they routinely hallucinate citations — fabricating papers, authors, or DOIs that look plausible but don't exist. The 2024–2025 wave of "AI lit review" tools has been called out repeatedly in retraction notices and editor blog posts. Use AI to triage abstracts, surface candidate papers, or suggest themes — but verify every citation against the original source before it lands in your bibliography. Tools that cite real, indexed papers (Connected Papers, ResearchRabbit, Litmaps, CitationMap) are safer than open-ended LLM chat for the citation step itself.

What's the difference between an annotated bibliography and a literature review?+

An annotated bibliography is a list — each entry is a 100–200 word standalone summary of one source, with no synthesis across entries. A literature review is an argument — it groups, compares, and synthesizes sources to motivate a research question. The simplest test: if you removed every citation from your text and the remaining prose still made an argument, it's a review. If removing citations leaves you with disconnected paragraphs, it's a disguised annotated bibliography. Examiners flag this constantly in master's and early PhD work.

What is a gap statement and where does it go?+

A gap statement is one or two sentences that tell the reader what the existing literature has not yet done — and what your work will do about it. It typically sits at the end of the literature review section, immediately before your research question or hypothesis. A strong gap statement names the gap precisely ("no published work measures X in low-resource settings") rather than vaguely ("more research is needed"). Vague gap statements are a top reject reason in journal peer review.

How do I position my work against the existing literature?+

Three sentences, late in the review. First, name the closest two or three existing studies. Second, state precisely how your work differs (population, method, scale, context, theoretical lens). Third, explain why that difference matters — what it lets you see that the prior work cannot. Reviewers and committees specifically look for this 'positioning paragraph' before they read your methods. Without it, your contribution looks incremental even if it isn't.

Worked examples by field

Citation maps for representative seed authors in three of the most commonly reviewed fields. Each map renders in 2–4 seconds and shows the global geographic footprint of that scholar's citing institutions.

Geoffrey Hinton — deep learning, machine learning, computer science.
Fei-Fei Li — computer vision, ImageNet, AI for healthcare.
Jennifer Doudna — CRISPR, biochemistry, gene editing.

For more examples, browse the full citation map showcase or start from the homepage with any Google Scholar profile.