
Data Extraction for a Literature Review: Tools and Template
By Jessica McLaren 15 min read Updated
Data extraction is the step where you pull the same set of facts from every included study (citation, design, sample, setting, measures, results and your appraisal) into one table, with one row per study and one column per fact. In a scoping review the same step is called charting the data, and in both cases the finished table is what lets you compare studies instead of summarising them one at a time.
Definition: data extraction. Data extraction is the recording of predefined facts from each study in a review into fixed fields of a table or form, so that every study is described in the same way and the studies can be compared, counted or pooled.
What is data extraction in a literature review?
Data extraction in a literature review means deciding in advance which facts you need from each study and recording those facts in the same fields for every study. The result is a grid: rows are studies, columns are the questions you ask of every study. Highlighting a PDF or writing a paragraph of notes per paper is not data extraction, because neither can be sorted or compared across studies.
Data extraction sits after screening and appraisal and before synthesis. The Cochrane Handbook calls the step “collecting data” and makes a piloted data collection form mandatory for a Cochrane systematic review (Cochrane Handbook version 6.5, 2024, chapter 5, MECIR standard C43). For a narrative review or a taught master's literature review chapter, nobody will ask to see the extraction table, but building one is still the most reliable way to write a chapter that compares studies rather than listing them.
What should a data extraction table include?
A data extraction table should include, at minimum, the citation, study design, sample, setting and country, the key variables or concepts, the outcome and result, the effect size or themes, an appraisal note, the page where each fact was found, and a note in your own words. The table below is a default set for a student review. Trim the fields to your research question rather than adopting all of them.
| Field | What to record | Why the field earns its place |
|---|---|---|
| Citation | Author, year and an identifier you can search on (DOI) | The row label, and the thing you paste into the chapter |
| Study design | Randomised trial, cohort, cross-sectional survey, interview study, case study | Decides how much weight the result carries and lets you group like with like |
| Sample | How many participants and who they were | Differences in sample often explain why two studies disagree |
| Setting and country | Where and when the study was done | Produces the geographic and time-period gaps in your review |
| Key variables or concepts | What was measured or explored, and how it was measured | Catches two studies using the same word for different things |
| Outcome and result | What was found, in the direction the authors report it, with the number where there is one | The finding you will compare across studies |
| Effect size or themes | The magnitude for quantitative studies; the named themes for qualitative studies | Lets you compare size of effect, not just count studies for and against |
| Appraisal note | Your judgement of how far to trust the study, and the main limitation behind that judgement | Stops a weak study and a strong study looking equally true |
| Page or location | The page, table or section each fact came from | Lets you check one number later without rereading the paper |
| Your note | One line, in your own words, on how this study relates to the others | The first comparative sentence of your review, kept separate from what the authors said |
The Cochrane Handbook's own checklist of data to collect (Table 5.3.a) groups the same ground into eight categories: information about the extraction itself, eligibility, study methods, participants, intervention, outcomes, results, and miscellaneous items such as the study authors' conclusions and funding. A health sciences systematic review should follow the Cochrane categories; a social sciences or education review can use the student default above.
Write “not reported” in a cell when a study does not give the information, rather than leaving the cell blank. The Cochrane Handbook recommends including “not applicable”, “not reported” and “cannot tell” options on a data collection form, because a blank cell cannot be told apart from a field you forgot to fill.
What does a data extraction table look like?
A data extraction table looks like a spreadsheet with one row per study and one column per field, and each cell holds a short, specific value with a page reference. The row below is a format illustration only: the study is not real, and the bracketed placeholders show what kind of value goes in each cell, not a result.
| Field | Recorded value (illustration) |
|---|---|
| Citation | (Author, Year) |
| Study design | Prospective cohort, no randomisation, [duration] |
| Sample | [number] first-year undergraduates at one university |
| Setting and country | [country], [academic years] |
| Intervention or exposure | Weekly structured peer mentoring by trained senior students |
| Outcome measure | Retention into second year, from registry records (p. [x]) |
| Result | [percentage] retained vs [percentage] in the comparison group (p. [x]) |
| Effect size | [difference] as reported by the authors; confidence interval: not reported |
| Appraisal note | Participation was voluntary, so students who opted in may differ from those who did not; no adjustment for prior academic performance |
| Your note | Same direction as (Author, Year) and (Author, Year), but the only study of the three using registry data rather than self-report |
Compare that row with a typical first attempt: “(Author, Year) found that peer mentoring was quite effective and improved retention significantly. Good study.” Every word of the first attempt may be true, but it cannot be sorted, it hides the size of the effect, and it loses the self-selection problem. The table row keeps the number, the page and the weakness visible, so the result cannot quietly turn causal in your chapter.
How do you do data extraction, step by step?
Data extraction works best in five steps: set the fields from your research question, pilot the form on a few contrasting studies, extract every study, check a sample of your own rows, and record what you did for your methods section. Setting the fields first matters most. If you let columns emerge as you go, by the tenth paper you will have invented new fields and have to reopen every earlier paper.
- Set the fields from your research question. Every field should answer one of two questions: what do I need to compare these studies fairly, or what will I need to quote to defend a claim?
- Pilot the form. Extract two or three studies that differ from each other as much as your included set allows. Usually one field turns out to mean different things in different papers, and one fact you keep wanting to record has no field. Fix both before you go on. The Cochrane Handbook makes a piloted data collection form mandatory for Cochrane reviews (MECIR standard C43).
- Extract every included study. Use one row per study, never one row per finding; put several findings on separate lines inside one cell. Copy numbers exactly as printed with their units, and do not round or convert while you extract.
- Check a sample of your own rows. If you have no second extractor, re-extract a few rows several days later without looking at your first attempt, and compare. Disagreements show you which fields are ambiguous.
- Record the process. Note who extracted, how the form was piloted and how disagreements were resolved, because the methods section of a systematic or scoping review must report it.
Steal this methodology sentence, and edit it so every part is true of your review:
Data were extracted into a piloted spreadsheet capturing study design, sample characteristics, setting, measures, reported outcomes and appraisal notes. The form was piloted on three studies and refined before full extraction. Where a value was not reported, this was recorded as “not reported” rather than left blank.
How is data extraction done in a systematic review?
In a systematic review, data extraction is done on a piloted form by at least two people working independently, with a set rule for resolving disagreements. The Cochrane Handbook (version 6.5, 2024, chapter 5) makes two independent extractors mandatory for outcome data (MECIR C46) and highly desirable for study characteristics (MECIR C45), and says the protocol should state how disagreements will be resolved.
The reason is error. Buscemi and colleagues (Journal of Clinical Epidemiology, 2006) found that single data extraction produced more errors than double extraction, a relative difference of 21.7%, while double extraction took longer, a relative difference of 36.1%. The Cochrane Handbook also cites Jones and colleagues (2005), who found data extraction errors in 20 of 34 reviews.
| Review type | Guideline and item | What you must report |
|---|---|---|
| Systematic review | PRISMA 2020, item 9 (data collection process) | The methods used to collect data from reports, including how many reviewers collected data from each report |
| Systematic review | PRISMA 2020, items 10a and 10b (data items) | Every outcome and every other variable for which data were sought, defined, plus any assumptions about missing or unclear information |
| Scoping review | PRISMA-ScR, item 10 (data charting process) | How data were charted, for example forms tested by the team before use, whether charting was done independently or in duplicate, and how data were obtained or confirmed from investigators |
| Scoping review | PRISMA-ScR, item 11 (data items) | Every variable for which data were sought, defined, with any assumptions and simplifications |
A student writing a systematic review alone cannot meet the two-extractor standard. Say so in the methods, describe the substitute you used (such as re-extracting a sample of rows after a gap), and report it as a limitation. A trimmed version of the extraction table usually becomes the study characteristics table in the results, and the study counts feed your PRISMA flow diagram.
Is charting the same as data extraction?
Charting is the scoping review name for data extraction: the physical act is the same, but the purpose of the table differs. The PRISMA-ScR checklist (Tricco and colleagues, Annals of Internal Medicine, 2018) notes that the Arksey and O'Malley framework, the Levac and colleagues framework and the JBI guidance all call data extraction in a scoping review “data charting”.
A data extraction table in a systematic review aims at a finding, so it goes deep on outcomes and effect sizes that can be compared or pooled. A charting table in a scoping review aims at a map of the field, so it usually has more columns about the studies (design, country, population, year) and fewer about results, because the shape and the gaps of the literature are the finding. If you are not sure which review you are writing, settle that first with scoping vs systematic vs narrative review, because the review type changes the columns.
What do you extract from qualitative studies?
From a qualitative study you extract the themes exactly as the authors named them, one short illustrative quote per theme with its page number, the analytic approach, and the participants and setting. From a quantitative study, by contrast, you extract values: the measure, the comparison, the direction, the magnitude and the precision where reported.
Keep the qualitative authors' themes in their words in one column and your own interpretation in a separate column, so you can always see which is which. Note the analytic approach (thematic analysis, grounded theory, interpretative phenomenological analysis), because each produces a different kind of claim. If your review includes both qualitative and quantitative studies, use two tabs with shared descriptive columns and different outcome columns, rather than one table full of empty cells, and bring the two together at the synthesis stage.
What tools can you use for data extraction?
The tools for data extraction in a literature review range from a plain spreadsheet, which suits most student reviews, to dedicated systematic review software for team reviews. The table below compares the main options by what each is good for. Product details are as of November 2026; check each tool's current documentation before choosing.
| Tool | Good for | Limits |
|---|---|---|
| Spreadsheet (Excel, Google Sheets) | Most student reviews: sorting and filtering by any column, freezing the citation column, hiding everything except design and outcome | Long qualitative text is awkward in cells; no built-in comparison between two extractors |
| Word or Google Docs table | Small narrative reviews and qualitative studies with long quotes | Limited sorting and no filtering; keep it in its own file, separate from the chapter draft |
| Literature Review Matrix (Litrevu free tool) | Starting a table: free Excel, Word and CSV templates, or add up to three of your own PDFs to fill the cells, each filled cell quoting its source passage | Three PDFs per matrix; for a full review, download the template and continue in a spreadsheet |
| Data systems named in the Cochrane Handbook (Covidence, EPPI-Reviewer, SRDR, DistillerSR) | Team systematic reviews; the Cochrane Handbook (Table 5.4.a) lists them as data systems for collecting data from study reports | Usually more than a solo student review needs; check whether your university library provides access |
| Elicit | Finding papers across a corpus of more than 138 million papers, systematic review screening, and structured data extraction into tables, including from PDFs you upload | Output is reports, tables and structured comparisons rather than written prose; as with any extraction tool, check values against the paper before they go into your review |
| Litrevu | Checking your extraction: comparing a handful of your own papers side by side in a cited table, and asking questions answered only from your papers | Not an extraction workflow and no screening at scale; for large extraction tables, Elicit or a dedicated data system is the better fit |
A reference manager such as Zotero, Mendeley or EndNote is not a data extraction tool. A reference manager stores citations, PDFs and notes, and you should keep using one alongside the extraction table; the comparison is in Zotero vs Mendeley vs EndNote. For a broader comparison of AI tools across screening, extraction and writing, see the best AI tool for a systematic review.
Can AI extract data from studies, and how do you check it?
AI tools can extract data from studies, but every value an AI tool extracts needs checking against the passage in the paper before it goes into your review. A living systematic review of data extraction automation (Schmidt and colleagues, F1000Research, version 3, 2025) found large language models used for data extraction in 17 of its included publications, and also reported a trend of decreasing quality in how results such as recall are reported.
Check AI extraction the way a second human extractor would: take a sample of rows, open the paper, and confirm each value against the exact page, table or passage. Treat a value you cannot trace to a passage as unverified. Check your university's AI policy before using AI at any stage, and declare how you used it; the AI disclosure statement generator drafts a declaration you can edit.
How do you turn a data extraction table into your literature review?
You turn a data extraction table into a literature review by reading down the columns instead of across the rows, and writing one sentence for each pattern you find. Reading across a row gives you one study, which you already know. Reading down the outcome column across every study gives you the pattern, and the pattern is what your review has to say.
Sort the extraction table by design and see whether the stronger studies agree with the weaker ones. Sort by country and see whether the finding travels. Sort by year and see whether the field changed its mind. Each sort produces a synthesis sentence, for example: “Most studies report a positive effect, but the studies using registry data rather than self-report show a smaller one.” Groups of related synthesis sentences become the themes that structuring a literature review by theme needs, and how to synthesize sources covers turning several rows into one argued paragraph.
What are the most common data extraction mistakes?
The most common data extraction mistakes are extracting into prose, letting field definitions drift, collecting too many fields, leaving out an appraisal column, and assuming a single extractor never slips. Each one makes the table look finished while leaving the studies impossible to compare.
- Extracting into prose. Twenty-two paragraphs of notes, one per paper, cannot be sorted. That is summarising, not extraction.
- Fields that drift. “Sample size” means participants in one row and schools in another. Write a one-line definition for each field at the top of the sheet and keep it visible.
- Too many fields. A field you never sort by is a field you did not need. Trace every field to your research question or to a comparison you plan to make.
- No appraisal column. Without one, a well-controlled trial and a small convenience survey sit side by side looking equally reliable. Add appraisal fields from the appraisal checklist and appraise in the same pass, since you are already reading the methods closely.
- Trusting a single pass. Buscemi and colleagues (2006) found 21.7% more errors with single extraction than with double extraction. If you work alone, re-extract a sample of rows later and compare.
One more check belongs in the same pass: confirm each paper has not been retracted before you extract from it.
How do you check a data extraction table with an AI assistant for writing backed by sources?
You check a data extraction table with an AI assistant for writing backed by sources by adding the studies you have already chosen and extracted, asking the assistant to compare them on your own fields, and opening each citation to confirm your table matches the passage. You still decide which studies count and what each field means; the assistant lines up the evidence in your papers so you can check your record against it.
Litrevu is an AI assistant for writing backed by sources. It reads every paper, report and PDF you add, answers your questions with a citation to the exact passage behind each claim, compares sources side by side, and checks your own draft against them.
The Litrevu workflow for checking an extraction table, using only the steps on how Litrevu works:
- Add your included studies. Upload the PDFs, paste a PDF link, a DOI or an arXiv link.
- Compare them on your fields. Ask for a side-by-side comparison on the fields you extracted, for example design, sample, findings and limitations. Each cell of the comparison table carries a citation.
- Open the citations. Each citation opens the source PDF with the cited passage highlighted, so you can check a value in your own table against the page it came from.
- Ask about disagreements. Ask where the studies agree and where they differ on an outcome. Litrevu answers only from your papers and says when your papers do not cover a question.
- Check your synthesis paragraph. Paste a paragraph you wrote from the table, and each claim comes back as Supported, Partly supported, Contradicted or Not in your papers. “Not in your papers” means your library does not cover that claim, not that the claim is wrong or that the literature has a gap.
Litrevu's comparison tables are built in chat for a handful of papers. Litrevu is not an extraction workflow and does not replace your extraction table; Litrevu is a way to check your extraction table against the papers before you write.
Data extraction FAQ
Do I need data extraction for a narrative review?
A narrative review does not formally require data extraction, and most universities will not ask to see an extraction table for a narrative literature review. Building a simple extraction table is still worth it, because sorting studies by design, sample or result is the quickest way to write a narrative review that compares studies instead of summarising them one by one.
How many people should extract the data?
The Cochrane Handbook makes at least two people working independently mandatory for outcome data in a Cochrane systematic review, and highly desirable for study characteristics. A student working alone should say so in the methods, re-extract a sample of rows after a gap to check for errors, and report single extraction as a limitation.
What is the difference between a data extraction form and a data extraction table?
A data extraction form is the template of fields you fill in for each study, and a data extraction table is the collected result, with one row per study. In a spreadsheet the two are usually the same file: the column headings are the form and the rows are the extracted data.
Should I appraise and extract in one pass?
Appraising and extracting in one pass saves opening every paper twice, because filling in the design and sample fields already means reading the methods section closely. Add appraisal columns from a checklist suited to the study design to the extraction table, and record the main weakness of each study as a short sentence rather than a single score.
Can AI extract data from studies, and how do I check it?
AI tools can extract data from studies, but every AI-extracted value has to be checked against the passage in the paper before it goes into a review. Check a sample of rows the way a second extractor would, opening each paper at the cited page, and declare AI use in line with your university's AI policy.
Is a data extraction table the same as a literature review matrix?
A data extraction table and a literature review matrix are the same kind of grid, with one row per study and one column per question. The term data extraction is used for systematic and scoping reviews and usually has more fields, while a literature review matrix is the lighter version used for a narrative review or a thesis chapter.
Build your data extraction table on your own papers
Start with the three most different studies in your set: set your fields, extract those three properly, and fix whichever fields break before you do the rest. Once the extraction table is built, add the same papers to Litrevu to check each row against the passage it came from before you write. Free to start, no credit card required.
Check your table against your papers