Used by 1,000+ researchers and students

Compare what the studies found, then write the literature review

A literature review stalls in the middle: you have twenty PDFs and no way to see where they agree. Grongy puts the comparison in one table, then drafts from the papers you kept.

From search to a written section

The four steps a review actually takes.

  • Search the indexes, not a model's memory

    OpenAlex, Crossref and DOAJ run on every search, plus PubMed and arXiv when your topic is in their fields: more than 320 million records as of October 2026. Free PDFs come from the indexes' open-access records, with Unpaywall checked for the rest.

  • Read and highlight in the browser

    Your PDFs open in the tab, highlights and notes collect in one notebook, and a note keeps the page it came from.

  • Compare them side by side

    A synthesis table holds one row per source: focus, population, sample, method, finding, limitation. Where a study does not state something, the cell says so rather than guessing.

  • Draft from what you kept

    Sections read from your library, topped up from the indexes when it runs short, and a paper the draft cites is added to it. If the sources don't support a sentence, it is written without a citation instead of with an invented one.

Why a table beats a pile of summaries

Summarizing twenty papers one by one gives you twenty paragraphs and no argument. A review needs the opposite: one axis at a time, across all of them.

That is what the table is for. Read down the Method column and you see whether the field has settled on an approach. Read down Population and you see who was never studied — which is usually where your own gap sits.

The cells are filled from each paper's own abstract or indexed text. A cell is left marked rather than filled when the paper does not report that detail; a blank you can see is worth more than a sentence that sounds complete.

Four studies, one table

This is the table for a real question, filled from four published studies. Read it down a column rather than across a row: the Method column tells you whether the field has settled on an approach, and the Population column tells you who was left out.

Research question How does academic self-efficacy relate to student performance?
SourceFocusPopulationSampleMethodFindingLimitationRelevance
Pajares & Miller, 1994Abstract onlyMath self-efficacy and self-concept in problem solvingnot statedN = 350Path analysisSelf-efficacy predicted problem solving better than self-concept, usefulness, prior experience or gendernot statedDirect: self-efficacy as a predictor of math performance
Chemers et al., 2001Abstract onlyAcademic self-efficacy and optimism in first-year adjustmentFirst-year university studentsnot statedLongitudinal; measured in the first quarter and at year endSelf-efficacy and optimism strongly related to performance and adjustment, directly and through expectationsnot statedDirect: links self-efficacy to classroom performance
Usher, 2009Abstract onlyHow students form their math self-efficacyMiddle school students with high or low self-efficacy, their parents and math teacherseight middle school studentsSemistructured interviewsStudents drew on all four hypothesized sources; teaching structures, course placement and self-regulation also matterednot statedPartly: where self-efficacy comes from; performance not measured
Richardson et al., 2012Abstract onlyPsychological correlates of university GPAUniversity students; English-language studies, 1997–2010241 data setsSystematic review; random-effects meta-analysisPerformance self-efficacy was the strongest of 50 correlates of GPAnot statedDirect: self-efficacy's correlation with GPA across studies
  • Cells that say “not stated” are not gaps in our reading — the paper does not report that detail. None of these four abstracts states a limitation, and one does not state its population. A blank you can see is worth more than a sentence that sounds complete.
  • The four studies reach the same subject through four different methods: a path analysis, a longitudinal study, semistructured interviews and a random-effects meta-analysis. That spread is why the agreement between them counts for something.
  • Every row carries the “Abstract only” badge: the cells are filled from each paper's abstract, not from its full text. Open the paper before you cite it.
  1. Chemers, M. M., Hu, L., & Garcia, B. F. (2001). Academic self-efficacy and first year college student performance and adjustment. Journal of Educational Psychology, 93(1), 55-64. https://doi.org/10.1037/0022-0663.93.1.55
  2. Pajares, F., & Miller, M. D. (1994). Role of self-efficacy and self-concept beliefs in mathematical problem solving: A path analysis. Journal of Educational Psychology, 86(2), 193-203. https://doi.org/10.1037/0022-0663.86.2.193
  3. Richardson, M., Abraham, C., & Bond, R. (2012). Psychological correlates of university students' academic performance: A systematic review and meta-analysis. Psychological Bulletin, 138(2), 353-387. https://doi.org/10.1037/a0026838
  4. Usher, E. L. (2009). Sources of middle school students’ self-efficacy in mathematics: A qualitative investigation. American Educational Research Journal, 46(1), 275-314. https://doi.org/10.3102/0002831208324517

How to write the conclusion of a literature review

A conclusion is not a summary. A summary repeats what each study said; a conclusion says what the studies together have settled, where they split, and what nobody has looked at yet. If you filled the synthesis table, you already have the material — read it by column instead of by row.

  1. Name what the field agrees on

    Find the rows where the finding repeats across different populations and methods. Agreement that survives a change of method is the strongest thing a review can report, and it belongs in your first sentences.

  2. Show where the studies split

    Two papers reaching opposite conclusions is not a problem to smooth over; it is a result. Say what differs — the population, the measure, the design — and let the disagreement stand with its reason.

  3. Point at the column that is empty

    Read down Population and Sample. The group nobody studied, the setting nobody tested, the measure nobody repeated — that blank is your gap, and it is defensible precisely because you can show the table it came from.

  4. Turn the gap into your question

    A gap is only useful if it produces a question you can answer with the data you can reach. Write the question, then check it against your own method: if you cannot collect it, the gap belongs to someone else's study.

What a conclusion does not do: introduce a source that appears nowhere above it, restate your whole method, or claim the field is settled when your own table shows it is not.

An example, written from the table above

Across three designs — a path analysis with 350 participants (Pajares & Miller, 1994), a longitudinal study of first-year university students measured twice in one year (Chemers et al., 2001), and a random-effects meta-analysis of 241 data sets (Richardson et al., 2012) — self-efficacy is consistently tied to academic performance in this set. Richardson et al. place performance self-efficacy first among fifty correlates of GPA, and Pajares and Miller find that math self-efficacy predicts problem solving better than self-concept, perceived usefulness, prior experience or gender. The agreement carries weight because the three arrive at it by different routes.

Where the set thins out is the question of origin. Usher (2009) is the only study here that asks where self-efficacy comes from, and it answers with eight middle school students in semistructured interviews: rich on sources, silent on performance. Nothing in this set connects the formation of self-efficacy to a measured outcome.

The blank sits in the Population column. Two studies work with university students, one does not state its population, and the only school-age study (Usher, 2009) is middle school and measures no performance. High school is absent. A study that measured both the sources of self-efficacy and performance in a high school cohort would answer a question none of these four do — and that is a gap you can defend, because you can show the table it came from.

Three paragraphs, four sources, and not one claim that isn't in a cell. Notice what the conclusion does not do: it adds no study that appears nowhere in the table, and it does not call the field settled — the Population column says otherwise.

Judging a source before you cite it

A review is only as good as what you let into it. The source integrity check runs on the records in your library with no AI and no usage credit; these are the four findings that matter most before you cite.

  • Retracted papers

    Retraction status comes from OpenAlex. Papers with a DOI are re-checked every week, and a retracted one is marked on its card; a source without a DOI can't be re-checked.

  • A DOI that is not registered

    The DOI in the reference is looked up. One that resolves nowhere, or resolves to a different publication, is a critical finding.

  • A record that does not match

    Year, authors and title are compared against the index entry, and the finding shows you what the index holds so you can correct your own.

  • Not found in the indexes

    Crossref, Semantic Scholar and arXiv are searched. It is only a minor finding: books, reports and journals published in languages other than English are often simply not held there.

How a review gets written

Four steps. Writing draws on your library and the scholarly indexes; only the chat page searches the open web.

  1. Define the question

    Narrow the topic until you can say what would count as an answer. A review without a question becomes a list.

  2. Search and screen

    Run the searches, keep what fits, and record why you excluded the rest. Screening notes are what makes the review repeatable.

  3. Fill the table

    One row per source. The columns force you to read for the same thing in each paper.

  4. Write by theme

    Draft along the table's axes, not paper by paper. Each theme becomes a section, and the citations come from the rows.

Start free

This is not a systematic review tool

A systematic review has a protocol, a registered search strategy, screening by two people and a reporting standard. Grongy does not run that process and does not claim to.

What it does hold is the part a narrative review needs, and some of it overlaps with the early stages of a systematic one.

What it does

  • Search across the indexes your topic reaches, with free PDFs where an open-access copy exists
  • A synthesis table with one row per included source
  • Citations that resolve to records you can open
  • Checks for retractions, unregistered DOIs and mismatched records

What it does not

  • No PRISMA flow diagram or checklist
  • No protocol registration
  • No dual independent screening
  • No risk-of-bias assessment or meta-analysis

If your program requires a systematic review, treat this as the stage before the protocol, not a replacement for it.

Start free

Search, the library, reading PDFs, the notebook and the map are free; the synthesis table, writing and questions to a PDF draw on your usage allowance.

Compare plans

Free

$0/mo

Start for free
  • Find academic sources and write from them
  • A limited review of your thesis before you hand it in
  • Upload PDFs and ask your papers questions
  • AI help that refills every 5 hours
  • Word, PDF, LaTeX and Markdown export
  • 20 PDF uploads (10 MB, 150 pages each; 50 MB in total)

Plus

$11.99/mo

Get Plus

Everything in Free, plus:

  • More AI use
  • Vision: slides, designs and dashboards
  • A deeper review of argument and structure
  • Automatic reference list in any citation style
  • Advanced models in chat: Claude, GPT, Gemini, Grok
  • 10 GB storage
  • Share drafts with your supervisor

Pro

$19.99/mo

Get Pro

Everything in Plus, plus:

  • Even more AI use
  • 25 GB storage
  • Early access to new AI models

Prices in US dollars, taxes included.

Questions researchers ask

How many sources should a literature review have?

There is no number that transfers between fields. A master's review in a narrow area may hold thirty; a doctoral chapter in a crowded field may hold two hundred. The test is coverage: can you name the main positions and say who holds them?

Can a chat tool write my literature review?

It can group and criticize sources you give it, and it can search the web. The risk sits in citations given from memory: a source that does not exist, or a real paper cited for a claim it never made. In Grongy the citations stay bound to records in your library and can be compared against the source's own text.

Does the table fill itself?

The cells are drafted from the paper's indexed full text where you have it, otherwise from its abstract. They are read-only: when your library changes, you rebuild the table. Where a paper does not report something, the cell is marked instead of filled — that marking is the useful part.

Can I export the table?

Not yet. The table is saved in this browser only, and there is no CSV export. Your reference list exports as BibTeX or RIS on Plus and Pro, and the live BibTeX link to the library is on Plus and Pro too.

Does it exclude preprints?

Search filters by document type, and in chat you can leave preprints out of both web results and your library. The source integrity check flags preprints already in your library rather than removing them, because for some questions a preprint is the only thing published.

Start with the table

Add four papers, fill one row each, and read down the Method column. That is the moment a pile of PDFs becomes a review.