# Literature Review Note-Taking Template: How to Structure Notes By the end of this guide you will have three working artifacts: a reusable note template with nine fields per source, a synthesis matrix that turns those notes into a sortable comparison table, and a small project folder with commands that keep both organized. You will also know how to adapt the template to a systematic review, a narrative review, or a policy review, and how to verify that your notes are complete enough to draft from without reopening a single PDF. You need a computer with three free tools: a plain text editor (Notepad on Windows, TextEdit on macOS set to plain text mode, or Visual Studio Code), spreadsheet software (LibreOffice Calc or Google Sheets), and a terminal for the optional command-line steps. A reference manager such as Zotero helps with citations but is not required. The commands below are Unix commands, so on Windows run them in Git Bash or the Windows Subsystem for Linux. On macOS the default shell is zsh, and on Linux it is usually bash. Both work. If you searched for a free literature review note taking template, the Markdown block in the next section is yours to copy. ## What a Note-Taking Template Does for a Literature Review A literature review is a structured reading of published work on a defined question, followed by a synthesis of what that work shows. The note-taking template has one job: move each paper's usable content into your notes exactly once, in a form you can compare across papers. Notes that fail at this force you to reread every study during drafting, which is where most reviews lose weeks. Ad hoc notes fail in predictable ways. Highlights and margin comments stay tied to the original PDF, keep the author's phrasing, and cannot be sorted or counted. A template fixes capture at the level published reviewers actually use. Published reviews show what that level is. A systematic review of physical activity (PA) interventions for post-treatment breast cancer survivors pre-specified the descriptors its authors would extract from every study before reading any of them: study descriptors, participant descriptors, intervention descriptors, and outcome descriptors [1]. A review of ambient AI scribes in healthcare mapped its themes onto the domains of the NASSS framework (Non-Adoption, Abandonment, Scale-Up, Spread, and Sustainability), a checklist of factors that determine whether health technology actually gets used in practice [2]. A narrative review of telehealth in radiation oncology sorted its findings into SWOT buckets (strengths, weaknesses, opportunities, threats) [3]. In each case the framework is the note-taking template. Every source is read against the same fixed set of buckets, which is what makes sources comparable. The nine fields below cover the descriptors most life science reviews need. Treat them as a starting research template and prune or extend them per project. Before your first note, write your review questions as a numbered list (Q1, Q2, Q3). The relevance field points each paper at one of them. ## The Core Literature Review Note-Taking Template: Nine Fields per Source | Field | What to record | |--|--| | 1. Citation | Authors, year, journal, DOI or PMID | | 2. Provenance | Database, search string, date captured | | 3. Design | Study type, e.g., randomized controlled trial, systematic review, narrative review | | 4. Population | Who was studied, sample size, setting | | 5. Intervention or exposure | What was tested, duration or dose, comparator | | 6. Outcomes and numbers | Primary outcome, effect size with confidence interval, direction of effect | | 7. Quality | Risk of bias assessment, funding, limitations the authors name | | 8. Quotes | Verbatim sentences you may quote, each with a page number | | 9. Relevance | A tag tying the paper to one of your numbered review questions | Three field names need definition. An effect size is a standardized measure of how large a finding is, such as Cohen's d, the difference between two group means expressed in standard deviation units. A confidence interval (CI) is the range of values consistent with the data, conventionally reported at 95 percent. Risk of bias is a structured judgment of how likely systematic error is in a study, and ROB 2 is the revised Cochrane tool for assessing it in randomized trials. Here is the template as a Markdown file. A PMID is the accession number a paper carries in PubMed, and a DOI (digital object identifier) is its permanent web address. ```markdown --- pmid: doi: first_author: year: journal: added: found_via: search_date: tags: [] --- # {first_author} {year}: {short title} ## One-sentence takeaway ## Design ## Population and setting ## Intervention or exposure ## Outcomes and numbers ## Limitations and risk of bias ## Quotes worth keeping (with page numbers) ## Relevance to my question ``` The block between the two triple-dash lines is YAML, a simple key-value format that many editors and tools parse. Keep values free of raw colons, or wrap them in double quotes. Below the YAML, each heading marked with two hash marks is one field of the template. ## A Filled-In Literature Review Note-Taking Template Example The following note describes a real published review, so you can see the detail level that makes later synthesis possible. The workflow is abstract and methods first, fill every field you can, then complete the quotes field during full-text reading. ```markdown --- pmid: 41518388 doi: first_author: Leitzelar year: 2026 journal: Breast Cancer Res Treat added: 2026-11-11 found_via: PubMed search_date: 2026-11-11 tags: [systematic-review, meta-analysis, physical-activity, oncology] --- # Leitzelar 2026: Physical activity interventions in post-treatment # breast cancer survivors ## One-sentence takeaway Home-based, theory-based physical activity counseling modestly increases activity in early post-treatment breast cancer survivors. ## Design Systematic review and meta-analysis (statistical pooling of results across studies) of RCTs published 2014 to 2025. Random effects model (a pooling model that assumes true effects vary between studies) on post-intervention Cohen's d standardized mean differences. Risk of bias assessed with ROB 2. ## Population and setting Breast cancer survivors within 5 years of treatment. 22 RCTs, 2,390 participants total (mean 109 per trial, range 26 to 692). Mean age 57, all female, more than 60 percent non-Hispanic White, mostly mixed stage and treatment type. ## Intervention or exposure Behavioral PA interventions. All but one partially or fully home-based. All counseling theory-based. Duration 6 to 104 weeks. ## Outcomes and numbers PA behavior at post-intervention. Overall SMD d = 0.36 (95% CI 0.22 to 0.50) favoring intervention. ## Limitations and risk of bias ROB 2 applied across trials. Predominantly non-Hispanic White, female samples limit [generalizability](/blog/data-analysis/generalizability-in-research) to other populations. ## Quotes worth keeping (with page numbers) (added during full-text reading, verbatim, each with its page) ## Relevance to my question Answers Q1 (does PA counseling work after treatment). Feeds the effect-size column of my matrix. ``` Every number in the note traces to the source, the provenance fields record how the paper was found, and the relevance line names the question it answers. When you draft, the sentence carrying the effect size is already written, sourced, and traceable. ## Matching the Template to Your Review Type ![Comparison of systematic, narrative, framework-anchored and policy reviews showing the extra field each one needs and what its notes capture.](/article-images/illustrations/literature-review-note-taking-template-review-type-fields.webp) Review types differ in what each note must capture. Adjust the nine fields accordingly. **Systematic reviews** test a pre-specified question with fixed eligibility criteria, a documented search, a formal risk-of-bias assessment, and often a meta-analysis. The worked example above follows that pattern. Keep all nine fields and add outcome subfields your evidence table needs. **Narrative reviews** synthesize a broader body of work with a flexible structure, so their notes trade fixed effect sizes for context. A narrative review of Canadian hypertension guidelines searched PubMed, Cochrane, and Google Scholar and traced how recommendations evolved from 2020 to 2025 [4]. If your review tracks guideline evolution, add a guideline-year field. A narrative review of pharmacological treatment for neuropathic pain in diabetes deliberately limited its search to the previous ten years in PubMed and ScienceDirect [5]. Record that kind of window in the provenance field, because readers will ask how current your evidence is. Narrative rows in your matrix can also carry qualitative entries instead of effect sizes, for example the observation that neuropathic pain affects up to roughly half of people with diabetes, which that review reports in its background [5]. **Framework-anchored reviews** attach every source to a theory. The ambient scribe review mapped 27 articles from Embase, PubMed, and Scopus onto NASSS domains [2]. If your review uses a framework, add a framework-domain field and tag each note against it. The telehealth review gathered its data following PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses, a reporting checklist) and organized the discussion by SWOT [3], which hands you four ready-made synthesis tags. **Mixed reviews** combine published work with new data. A literature overview of proctored online intelligence testing paired its review with a new empirical experiment [6]. If yours does the same, add a field that marks each entry as published data or your own measurements, so the two never blur in the draft. **Policy reviews** organize sources around problems, barriers, and recommendations. A policy review of digital health and artificial intelligence in African oncology structured its analysis around potential gains in access, efficiency, and equity, weighed against adoption barriers such as infrastructure gaps and limited digital literacy [7]. Add barrier and recommendation fields if that is your genre. **Taxonomy-driven reviews** live and die by their category system. A review of information technology tools against substandard and falsified medicines grouped its sources by tool type, including mobile authentication apps, 2D barcoding, radiofrequency identification tags, and machine-learning-based field testing kits [8]. Give your template a controlled tag list, a fixed vocabulary you never deviate from, so categories stay countable. ## Choosing a Note Format: Files, Spreadsheets, and PDF Export | Format | Best for | Strengths | Watch out for | |--|--|--|--| | One Markdown file per paper | Most projects | Searchable with grep, works with version control, pastes into drafts | File sprawl without a naming convention | | Single running Markdown file | Small reviews under about 30 papers | Everything in one scroll | Hard to sort or count later | | CSV synthesis matrix | Comparison across many studies | Sorts, filters, feeds tables directly | Breaks when fields contain unquoted commas | | Reference manager notes | Citation-heavy theses | Notes travel with the citation and PDF | Locked to one tool unless you export | | Hybrid (Markdown notes plus CSV matrix) | Systematic-style reviews | Detail and comparison in their best forms | Two files to keep in sync | CSV means comma-separated values, a plain text table format that every spreadsheet program opens. Every format above has a free option, and the Markdown master also exports to PDF, which matters if you searched for a [lab notebook](/knowledge/molecular-biology/lab-notebook) template pdf and want a printable form. Keep the Markdown master and print from it with pandoc, a command-line document converter: ```bash pandoc lit-review/templates/paper-note.md -o paper-note.pdf ``` If no LaTeX toolchain (a typesetting system pandoc uses for PDF output) is installed, pandoc exits with an error telling you to choose a different pdf engine. Install a LaTeX distribution such as TeX Live, select another installed engine such as wkhtmltopdf, or open the Markdown preview in your editor and print that to PDF. ## Setting Up the Template: A Numbered Walkthrough ![Six-step flow: create folders, save template, create matrix header, copy template for first note, append matrix row, run duplicate and section checks.](/article-images/illustrations/literature-review-note-taking-template-setup-walkthrough.webp) ### Step 1: Create the project folder ```bash mkdir -p lit-review/templates lit-review/notes lit-review/matrix ``` The -p flag creates parent folders as needed and stays silent on success. Verify: ```bash ls lit-review ``` Expected output: ``` matrix notes templates ``` ### Step 2: Save the note template ```bash touch lit-review/templates/paper-note.md ``` Open the empty file in your editor and paste the Markdown template from the nine-fields section. Save it. Verify the top of the file: ```bash head -12 lit-review/templates/paper-note.md ``` Expected output: ``` --- pmid: doi: first_author: year: journal: added: found_via: search_date: tags: [] --- ``` ### Step 3: Create the synthesis matrix ```bash cat > lit-review/matrix/synthesis.csv << 'EOF' pmid,first_author,year,design,population,intervention,outcomes,effect_size,risk_of_bias,relevance EOF ``` The here-document (the text between the EOF markers) writes the header row into the file. The single greater-than sign creates the file, and a later mistake with a second one would overwrite it, so keep them straight. Verify: ```bash cat lit-review/matrix/synthesis.csv ``` Expected output: the ten-column header row and nothing else. ### Step 4: Take your first note ```bash cp lit-review/templates/paper-note.md lit-review/notes/41518388-leitzelar-2026.md ``` The naming convention is PMID, first author, year. It sorts chronologically, stays unique, and makes the note findable from the matrix row. Open the copy in your editor and fill it in. The result should look like the filled-in example above. Verify: ```bash ls lit-review/notes ``` Expected output: ``` 41518388-leitzelar-2026.md ``` ### Step 5: Add the matching matrix row ```bash echo '41518388,Leitzelar,2026,systematic review + meta-analysis,"breast cancer survivors within 5 years of treatment","physical activity counseling, home-based, theory-based",PA behavior,"d = 0.36 (95% CI 0.22 to 0.50)",ROB 2,Q1' >> lit-review/matrix/synthesis.csv ``` The double greater-than sign appends. Any field containing a comma sits inside double quotes. Verify the last line: ```bash tail -1 lit-review/matrix/synthesis.csv ``` Expected output: the row you just appended. Open synthesis.csv in LibreOffice Calc or Google Sheets. You should see two rows, the header and the Leitzelar entry, with ten columns each. If the effect size spilled into the risk-of-bias column, a comma escaped its quotes. That is error 3 below. ### Step 6: Run the setup checks ```bash cut -d, -f1 lit-review/matrix/synthesis.csv | sort | uniq -d ``` This splits each line on commas, takes field 1 (the PMID), sorts it, and prints only duplicated values. Expected output: nothing. Any PMID that prints appears in two rows. ```bash grep -c "## One-sentence takeaway" lit-review/notes/*.md ``` Expected output: ``` lit-review/notes/41518388-leitzelar-2026.md:1 ``` A count of 0 means a note is missing that section. ## Common Errors and Exact Fixes 1. **Command typo.** You type `mkdir` instead of `mkdir` and the shell answers: ``` zsh: command not found: mkdir ``` On bash the message reads `mkdir: command not found`. Fix: retype the command and use tab completion, which fills folder names and prevents most typos. 2. **YAML colon error.** You paste a journal name containing a colon into the front matter, and a tool that parses YAML (a static site generator, an Obsidian export, a Python script) fails with: ``` yaml.scanner.ScannerError: could not find expected ':' ``` Some tools print a longer version with the line and column number. Fix: wrap any value containing a colon in double quotes, for example `journal: "JAMA: The Journal of the American Medical Association"`. 3. **CSV column shift.** You type an effect size with an unquoted comma, and the spreadsheet pushes everything after it one column right, so `0.50)` lands in risk_of_bias and relevance ends up empty. Fix: wrap every field containing a comma in double quotes, or write ranges with the word to instead of a comma. 4. **Wrong path.** You mistype a folder name and grep answers: ``` grep: lit-review/nodes: No such file or directory ``` Fix: run `ls lit-review` to confirm the folder names, then retype the path. 5. **Duplicate notes for one paper.** You find the same paper through PubMed and again through Embase and note it twice. The duplicate check from Step 6 prints the PMID instead of returning nothing. Fix: merge the two notes into one, keep both databases in the found_via field, and delete the extra matrix row. ## How to Verify Your Notes Are Complete Rerun the two Step 6 checks after every batch of notes. The duplicate check should print nothing, and no note should show a count of 0 for a template section. Then run three tests that commands cannot run for you. The column test: sort the matrix by the relevance column in your spreadsheet, and empty cells surface at the top. The traceability test: pick three numbers from your draft, locate each one in a note, then locate the page-anchored quote behind it. The reread test: draft one synthesis paragraph with every PDF closed. If you must reopen a paper, the note is missing a field. Add the field to the template itself, not just to that one note, so every future paper captures it. ## Common Mistakes and Limitations 1. **Pasting the abstract into the takeaway field.** This creates two problems. Verbatim text becomes a plagiarism risk the moment it reaches your draft, and abstracts describe papers persuasively rather than accurately. Paraphrase in one sentence. 2. **Quotes without page numbers.** An unanchored quote is unfindable six months later and uncitable in the draft. Page number or nothing. 3. **Highlighting instead of noting.** Highlighting leaves the words in the author's phrasing and outside your system, which means they cannot be sorted, counted, or combined. 4. **Skipping provenance.** Months later you cannot reconstruct the search or judge coverage gaps. Record database, search string, and date at capture time, not retrospectively. 5. **Over-built templates.** Thirty fields per paper and note-taking takes longer than reading. Nine to twelve fields is the working range. Prune anything you never fill. 6. **Theme-only notes.** Notes organized by theme before per-paper capture make cherry-picking easy, because you end up recording only what fits the argument you already have. 7. **What a template cannot do.** A template standardizes capture, not judgment. It will not decide whether a study is trustworthy or relevant, which stays with your risk-of-bias reading and your review question. Templates also decay, so revisit the field list after every ten papers. If you are preparing a thesis or a journal submission, follow the reporting standard your program or target journal requires rather than the defaults in this article. ## Frequently Asked Questions ### What is a literature review note-taking template? A reusable form with fixed fields, including citation, provenance, design, population, intervention, outcomes, quality, quotes, and relevance, that you fill in once per source. Fixed fields make sources comparable, so synthesis happens in your notes instead of during a second read of every paper. ### What does a good literature review note taking template format look like? Two linked artifacts. One structured note per paper, usually a Markdown file with a labeled heading per field, plus a synthesis matrix, which is a spreadsheet with one row per study and one column per field. The note holds the detail and the matrix holds the comparison. ### Is there a free literature review note taking template I can copy? Yes. The Markdown template in this article is free to copy, and every tool it uses has a no-cost option, including Visual Studio Code, LibreOffice Calc, and Google Sheets. Duplicate the template file once per paper and add one matching row to the matrix. ### Should I write one note per paper or one note per theme? Both, in that order. Write one structured note per paper first, then theme-level synthesis notes that cite the paper notes. Theme-only notes invite cherry-picking, because you record only what fits the argument you already have. ### How do I record where each paper came from? Fill the provenance field at capture time with the database, the search string, and the date. Published reviews report exactly this information in their methods sections, and reconstructing it months later is close to impossible. ### Can AI tools fill in my literature notes? They can draft them. Large language models can summarize lengthy texts and automate parts of note-taking [9], but research on AI documentation tools reports hallucinations and omissions that require human editorial work [2]. Treat any AI summary as a draft and verify every number against the paper itself. ### How is a literature note template different from a lab notebook template? A lab notebook records what you did and observed, chronologically, one entry per experiment. A literature note records what others did and found, one entry per source, structured for side-by-side comparison. The habits transfer, including dated entries and export to PDF. ### How many fields should my template have? Nine to twelve. Below nine you lose the comparison points synthesis needs, and above twelve note-taking takes longer than reading. Start with the nine fields in this article, drop any field you never fill, and add a field only when drafting reveals a need. ## Sources 1. [Updated systematic review and meta-analysis: taking the next step in physical activity behavioral interventions for post-treatment breast cancer survivors.](https://pubmed.ncbi.nlm.nih.gov/41518388/) 2. [Beyond productivity: a NASSS-informed review of implementation risks and research priorities for ambient AI scribes in healthcare.](https://pubmed.ncbi.nlm.nih.gov/42877726/) 3. [The evolving role of telehealth in interprofessional radiation oncology service delivery: A narrative review.](https://pubmed.ncbi.nlm.nih.gov/41539064/) 4. [The Evolution of Hypertension Management in Canada: A Review of the Current Guidelines.](https://pubmed.ncbi.nlm.nih.gov/41527632/) 5. [Neuropathic pain in diabetic patients: a review of pharmacological treatment selection according to psychopathological and somatic factors.](https://pubmed.ncbi.nlm.nih.gov/41669622/) 6. [Effects of Proctoring on Online Intelligence Measurement: A Literature Overview and an Empirical Study.](https://pubmed.ncbi.nlm.nih.gov/41003250/) 7. [Digital health and artificial intelligence innovations for oncology in sub-Saharan Africa.](https://pubmed.ncbi.nlm.nih.gov/41038203/) 8. [Future Information Technology Tools for Fighting Substandard and Falsified Medicines in Low- and Middle-Income Countries.](https://pubmed.ncbi.nlm.nih.gov/30233373/) 9. [The Role of Large Language Models in Ophthalmology: A Review of Current Applications, Performance, and Future Directions.](https://pubmed.ncbi.nlm.nih.gov/41431521/)