A free systematic review screening tool for title and abstract screening
This systematic review screening tool handles the study-selection stage of a review: title and abstract screening. You import your de-duplicated search results, define your eligibility concepts, and the tool ranks every record by relevance so the studies most likely to meet your criteria appear first. Screening this way is faster than reading records in arbitrary order, and because the tool never hides borderline records, recall is preserved: ranking orders the list, a human reviewer still decides every include and exclude. It is a free, account-free alternative to Covidence and Rayyan for the screening step. By default everything runs locally in your browser, so no record, abstract, or decision is uploaded to a server. When you want a second reviewer, an optional real-time collaboration mode lets you share one invite link and screen together; only then is the project synced, and only between people who hold the link.
The tool reads the formats databases and reference managers actually export, RIS, PubMed .nbib/MEDLINE, CSV, TSV, and Excel, and merges multiple files while removing duplicates by DOI, exact title, and fuzzy near-duplicate matching (Jaccard title similarity) that catches the same study indexed differently across databases. Ranking runs in a background worker and projects persist in your browser's database rather than its size-capped local storage, so reviews with tens of thousands of records stay responsive. As you mark records Include, Maybe, or Exclude, the optional active-learning classifier trains on your own decisions, re-ranks the unscreened pile, and estimates how many relevant records likely remain, giving you a defensible, recall-oriented stopping signal rather than a guess. Structured exclusion reasons (wrong population, intervention, comparator, outcome, or study design) are tallied automatically for your reporting. Because best practice is two independent reviewers, the Reconcile step loads two saved projects, scores agreement with Cohen's kappa, and lists every conflict to resolve, the offline equivalent of dual screening. You can map your question first with the PICO framework builder so the eligibility concepts you screen against are well defined.
Screening sits in the middle of the review pipeline. Before you screen, finalize your search with the search strategy builder and remove duplicates with the reference deduplication tool. Pin down your eligibility rules with the inclusion and exclusion criteria builder. When screening is done, the counts here flow straight into the PRISMA flow diagram generator for a publication-ready figure that meets PRISMA 2020 reporting standards expected by Cochrane, JBI, and major journals. To see how screening fits the wider workflow, follow our step-by-step systematic review guide.
How accurate is this screening tool? Measured, not promised
We benchmarked every automated component of this tool on five published screening corpora spanning four research fields: the van de Schoot 2018 post-traumatic stress disorder review (SYNERGY collection), the Smid 2020 Bayesian statistics review, and the Hall 2012, Wahono 2015, and Radjenovic 2013 software engineering reviews, together more than 28,000 records at the realistic 0.4 to 1.2% relevance prevalence of real reviews. The benchmark is replayable, and we re-run it whenever the screening prompt, ranking, or de-duplication engine changes.
AI screening: 99%+ recall across all five datasets. On ~200-record labeled samples screened against each review's eligibility criteria, the AI kept 229 of 231 relevant records for human review (99.1%, with 100% on three of the five corpora), suggesting include or maybe rather than exclude. False includes averaged 3.5%, and on the PTSD corpus not a single irrelevant record was suggested as an include. Both misses were borderline judgment calls on full abstracts, the same kind human reviewers disagree on. The tool also has two hard safety rules no competitor states: a record without a usable abstract is never suggested for exclusion, and every exclude suggestion is re-read by an independent second model pass that demotes anything plausibly eligible to maybe. Suggestions are never applied automatically; a human confirms every decision.
Relevance ranking: 69 to 93% of screening work saved. In active-learning simulations seeded with one include and one exclude, reaching 95% recall saved between 69% and 93% of screening effort (WSS@95) across the five corpora: 92.7% on Hall 2012, 88.3% on Radjenovic 2013, 83.8% on Wahono 2015, 76.5% on the PTSD corpus, and 69.2% on Smid 2020, beating the best published active-learning results on two of the datasets and approaching them on the rest. The ranker starts learning from your very first included record. Screening just the top 10% of the ranked pile already finds 82 to 99% of relevant records. Records with no abstract are the known weakness of every ranking engine, which is why this tool does two things competitors skip: it can fetch missing abstracts from OpenAlex at import, and it surfaces the remaining abstract-less records in a dedicated No abstract tab so they are screened by hand instead of silently sinking to the bottom.
De-duplication: zero wrong merges. Against 300 distinct same-topic studies, the hardest kind of false-merge bait, the de-duplicator made no incorrect merges (100% precision) while catching 97.4% of synthetic duplicate variants across the classes seen in real database exports: case changes, punctuation, hyphenation differences such as post-traumatic versus posttraumatic, truncated subtitles, and records matched by DOI alone. Anything it merges is reversible and listed for human review.
These numbers come from five public benchmark datasets; your corpus will differ. But they are measured numbers, not marketing copy, and the same recall-first design principle applies everywhere in the tool: nothing is ever hidden or auto-excluded on a model's say-so.
Reporting these results in a manuscript? How to cite this tool in your systematic review, with ready-to-paste methods-section text and the benchmark reference.
A free alternative to Rayyan and Covidence for screening
Most reviewers reach the screening stage already weighing a Rayyan alternative or a Covidence alternative, usually because of cost, a seat limit, or a data-residency policy that forbids uploading unpublished records to a third-party server. The difference here is where the work happens. Rayyan and Covidence are hosted platforms: your library, decisions, and reviewer activity live on their servers behind a subscription or a per-review charge. This tool keeps the same screening workflow, relevance ranking, structured exclusion reasons, two-reviewer reconciliation with Cohen's kappa, and PRISMA counts, but runs in your browser with nothing uploaded by default and no account required to start. There is no record cap and no paywall on the screening step itself.
The same reasoning applies to the wider category of systematic review software: DistillerSR and Covidence are paid hosted platforms, while ASReview and Abstrackr are free machine-learning screeners that still expect you to install software or upload your library. This tool sits alongside them as free, browser-based literature review software for the screening step, with active-learning prioritization built in and nothing to install.
It is not a like-for-like replacement for every team. Covidence bundles full-text review and data extraction, and Rayyan's paid tiers add team management; if you need a hosted shared workspace with managed accounts, those remain reasonable choices. If you would rather not screen at all, our methodologists can run the entire systematic review for you, from search to synthesis. What this tool replaces is the common case: a single reviewer or a small pair who need rigorous, recall-oriented title and abstract screening without a subscription, and who would rather keep unpublished search results on their own machine. When you do want a second reviewer, optional collaboration syncs the project only between people holding the invite link.
How it compares with Covidence and Rayyan, feature by feature
The table below is factual, not promotional: competitor cells reflect what Covidence and Rayyan state on their own pricing and documentation pages, and our cells reflect what ships in this tool. Both competitors have genuine strengths this table also shows, Rayyan has a native mobile app with offline screening, and Covidence has been Cochrane's standard production platform since 2015. Where a vendor publishes no information, the cell says so instead of guessing.
| Capability | Research Gold | Covidence | Rayyan |
|---|---|---|---|
| Price for a small team, per year | Team plan: $39 per month or $390 per year with 5 seats included | Package: $907 per year for up to 3 reviews; a single review is $339 per year | Advanced: $8.33 per seat per month billed annually, or $13.33 per month billed quarterly; no monthly billing |
| Free tier scope | Unlimited local screening plus 3 cloud-saved reviews of 5,000 records each, with de-duplication, relevance ranking, and the PRISMA 2020 flow diagram included | Free trial limited to 1 review capped at 500 records | Free forever: 3 active reviews and 2 free reviewers, but no PRISMA flow diagram and no automatic duplicate resolution |
| AI screening suggestions you can buy self-serve | Yes. Credit-metered suggestions with a reason per record, open to any signed-in account; new accounts get 50 trial credits | No. Covidence publicly chose not to release its language-model screening feature | AI Reviewer exists but is available exclusively on institutional plans with a 5-license minimum |
| Published accuracy with methodology | Yes: 99%+ recall (229 of 231 relevant records kept) measured across five named public benchmark datasets in four research fields, with the methodology on this page | Publishes recall above 99.5% only for a classifier that tags randomized controlled trials; no criteria-based screening accuracy | No accuracy numbers published |
| Active-learning relevance ranking | Yes, free, with a recall-based stopping estimate; starts learning from your first included record | Yes (Most relevant sorting); activates only after at least 25 screened studies and does not run on reviews under 100 records | Yes (5-star relevance predictions); no performance numbers published |
| Data extraction, including AI and dual extraction | Custom forms with prebuilt templates (Pro); AI extraction that cites an evidence quote per field for you to approve; dual extraction with reopenable consensus (Team) | Two incompatible extraction modules; the customizable one cannot export to RevMan; AI suggestions cover only funding, country, and intervention fields | Custom forms; AI Auto-Extract is limited to institutional plans |
| Risk of bias frameworks | RoB 2, ROBINS-I, QUADAS-2, Newcastle-Ottawa Scale, JBI checklists, AMSTAR 2, plus a custom framework builder (Team and above); full signaling questions with suggested judgments, blinded dual assessment with a conflict log, and per-outcome RoB 2 assessments | Default template based on the Cochrane Risk of Bias version 1 tool; RoB 2 is not supported in-app | RoB 2, ROBINS-I, QUADAS-2, Newcastle-Ottawa Scale, JBI tools, AMSTAR 2, and custom frameworks |
| Meta-analysis built in | Yes (Pro): pooled effects, heterogeneity, subgroup analysis, forest and funnel plots | Not offered; analysis happens outside the tool in RevMan or a statistics package | Appears as a single row in its pricing table with no page or detail published |
| PRISMA 2020 flow diagram | Included free | Yes, auto-updating, with a DOCX download | Paid plans only; not included in the free tier |
| Protocol and search strategy builder | Yes: a protocol builder and a search strategy builder are built in | Not offered | Protocol and Search appear as stage names in its pricing table with no documentation published |
| Direct database search and citation chasing | Searches 7 free databases directly (PubMed, Europe PMC, OpenAlex, Semantic Scholar, Crossref, DOAJ, ClinicalTrials.gov) plus backward and forward citation chasing (Institution) | Not offered; import only | None advertised; literature searching happens outside the tool, import only |
| REST API, webhooks, and Model Context Protocol server | Yes (Enterprise): REST API with decision writes, webhooks, a Model Context Protocol server for AI assistants, plus read-only share links | No public API mentioned anywhere on its site | REST API and a Python kit on Enterprise only; a Model Context Protocol server is listed in its pricing table |
| Duplicate detection with measured precision | Measured: zero wrong merges (100% precision) and 97.4% of synthetic duplicate variants caught; every merge is reversible | Deliberately conservative; its own blog acknowledges some duplicates can get missed | Automatic detection with a paid auto-resolver; cites a number one ranking from an independent study it does not name |
| Translation of non-English titles and abstracts | Yes: credit-metered translation into English with a show-original toggle | Not advertised | Language filter facets only; no translation feature advertised |
Competitor details verified from public pricing and documentation pages, July 2026. Vendors change plans and features; check their sites for current terms.
What one systematic review actually costs
Covidence prices by the review: its Single plan is $339 per year and covers exactly one review, valid for 12 months, with a 3-review Package at $907 per year and no monthly billing option. Our Pro plan is $12 per month or $99 per year and covers unlimited reviews. In plain math: the price of one Covidence review for one year funds more than three years of unlimited reviews on Pro. Rayyan sells by the seat with no monthly option either; its cheapest paid entry is a quarterly commitment, while you can pay here month to month and cancel any time.
Writing a thesis or dissertation? Screen free.
The free tier is real screening, not a teaser: unlimited local title and abstract screening in your browser, automatic de-duplication, active-learning relevance ranking with a stopping estimate, PICO highlighting, structured exclusion reasons, a pilot sample for calibration, the PRISMA 2020 flow diagram, and two-reviewer reconciliation with Cohen's kappa. Sign in and you also get up to 3 cloud-saved reviews of 5,000 records each. No credit card is required, and the whole workflow works solo, which is exactly how most theses and dissertations are screened. For contrast, Covidence's own pricing page says it does not offer an additional student discount, so one dissertation review there costs $339 for a year.
Built to be checked, not taken on faith
- Accuracy is measured on a public dataset and published with its full benchmark methodology, not claimed in a slogan.
- Reviewers can grab the recommended citation and copy-paste methods text for their manuscript.
- Documentation is public: step-by-step guides for every feature, from import to PRISMA export.
- No lock-in: decisions export to CSV and whole projects to JSON on every plan including Free, with RIS, BibTeX, EndNote XML, and RevMan formats on Pro.
- No credit card is required for the free tier, and there is no time limit on it.
- Local projects never leave your browser by default; cloud-saved reviews sit behind authenticated accounts with a logged audit trail on paid plans.
Systematic review screening software that runs in your browser
As systematic review screening software, this tool is local-first rather than cloud-first. There is nothing to install and no server round-trip on import or ranking: your file is parsed in the page, projects are stored in your browser's IndexedDB, and the active-learning classifier trains on-device, so a review with tens of thousands of records stays responsive offline. That design is what makes it genuinely private, the records, abstracts, and include/exclude decisions never leave your computer unless you deliberately turn on collaboration.
Because the engine is the same one a hosted platform would run, the outputs are publication-grade: a recall estimate for a defensible stopping point, a kappa score for inter-reviewer agreement, and exact counts that drop into a PRISMA 2020 flow diagram. The trade-off of local-first software is that your projects live in one browser profile, so export a backup if you switch machines, or sign in to sync saved reviews across devices.
How to use the screening tool, from import to PRISMA
The core workflow (import, rank, decide, reconcile, export) is free and runs entirely in your browser. AI screening is metered by credits, and features marked Pro, Team, Institution, or Enterprise describe what the paid plans add on top for full-text review, quality appraisal, and compliance.
- 1
Import your search results
Upload or paste your search export in RIS, PubMed .nbib/MEDLINE, CSV, TSV, or Excel format (BibTeX and EndNote XML on Pro). Multiple files merge and de-duplicate automatically, the duplicate count feeds straight into your PRISMA identification numbers, and records missing an abstract can fetch it from OpenAlex at import. Institution plans can also search seven free databases directly (PubMed, Europe PMC, OpenAlex, Semantic Scholar, Crossref, DOAJ, ClinicalTrials.gov) and run backward and forward citation chasing, while built-in export guides cover subscription databases such as Embase, Scopus, Web of Science, CINAHL, and Google Scholar.
- 2
Set your relevance criteria
Enter inclusion concepts (one per line, commas for synonyms) and any exclude signals, then rank. Likely-relevant records rise to the top and exclude signals are highlighted in red but never auto-removed.
Want your eligibility criteria pressure-tested against your protocol before you screen? Get a quote for screening support from our PhD methodologists.
- 3
Run AI screening if you want a first pass
Signed-in users can run credit-metered AI screening: the model reads every title and abstract against your criteria and suggests include, maybe, or exclude with a reason for each call. Across five published benchmark datasets it kept 99%+ of relevant records (229 of 231). Two recall safeguards run on every batch: a record without an abstract is never suggested for exclusion, and every exclude suggestion is re-read by an independent second pass that demotes plausible candidates to maybe. Suggestions are never applied automatically; a human reviewer confirms every decision.
- 4
Decide and let the tool learn
Mark each record Include, Maybe, or Exclude with a structured reason, using keyboard shortcuts for speed. From your first included record, the active-learning classifier re-ranks the rest, and once you also have an exclude it estimates how many relevant records likely remain so you know when it is safe to stop. Pro plans can also check every record's DOI against Crossref retraction notices, sourced from the Retraction Watch database, so a retracted study never slips into your review.
Screening thousands of records on a deadline? Get a quote for screening support from our PhD methodologists.
- 5
Reconcile two reviewers
Screen as two independent reviewers two ways. Offline: each reviewer saves a project file and you load both into the Reconcile step to score agreement with Cohen's kappa and resolve every conflict. Or opt into live collaboration: share one invite link, screen the same records together in real time, and watch agreement build as you go, the equivalent of dual screening in Covidence or Rayyan. The Team plan turns this into a shared workspace with blinded independent dual screening, a conflict dashboard with third-reviewer tiebreak, roles and permissions, and inter-rater reliability analytics over time.
Need a qualified second independent screener on your review? Get a quote for screening support from our PhD methodologists.
- 6
Screen full texts and extract data (Pro)
Pro unlocks the second selection stage: full-text screening with a built-in PDF viewer, an open-access full-text finder powered by Unpaywall, structured full-text exclusion reasons for PRISMA reporting, and data extraction forms, all backed by an audit trail that records every decision.
- 7
Assess quality and generate your methods section
Team plans add risk-of-bias and critical-appraisal modules (RoB 2, ROBINS-I, QUADAS-2, Newcastle-Ottawa Scale, JBI checklists, AMSTAR 2, plus a builder for custom appraisal frameworks) with publication-ready traffic light and judgment bar figures, and Institution plans add GRADE certainty assessment, so appraisal lives beside your screening decisions. Pro's PRISMA 2020 checklist and methods-section generator then turn your screening log into reporting-ready text for the manuscript.
- 8
Run your meta-analysis without leaving the tool (Pro)
Pro adds a synthesis stage that neither Covidence nor Rayyan offers: enter events and totals (or means, standard deviations and group sizes) per included study for each outcome, or prefill the grid from your extraction data, then pool with Mantel-Haenszel fixed effect or DerSimonian and Laird random effects (risk ratio, odds ratio, risk difference, mean difference, or standardized mean difference with Hedges g). You get Q, I-squared and tau-squared heterogeneity, subgroup analysis with a between-subgroup test, a downloadable forest plot and funnel plot (SVG and PNG), Egger's regression test, a CSV analysis table, and ready-to-adapt methods text.
- 9
Export, chart, and keep the review alive
Download decisions as CSV, save the whole project as JSON to resume later, copy your PRISMA counts, or build a PRISMA 2020 flow diagram and export it as an image; on the free local workflow nothing is uploaded to a server. Sign in to save reviews to your account and sync across devices (three cloud reviews on Free, unlimited on Pro). Pro adds custom exports for RevMan, Cochrane, and citation styles, Team adds living-review update screening for when your search is re-run, and Enterprise adds REST API access plus 21 CFR Part 11 electronic signatures for regulated environments.
Frequently asked questions
What is title and abstract screening?
Title and abstract screening is the first study-selection stage of a systematic review. After de-duplicating your database search results, two reviewers independently read each record's title and abstract and decide whether it could meet the eligibility criteria. The goal is high recall, not precision: a record is only excluded when its title or abstract clearly shows it does not fit the population, intervention, comparator, outcome, or study design (PICOS). Anything borderline is carried forward to full-text screening. This tool ranks records by relevance so likely-includes surface first, but a human reviewer still makes every include or exclude call.
How to do abstract screening?
Import your search export (RIS, PubMed .nbib/MEDLINE, CSV, TSV, or Excel) and de-duplicate. Define your inclusion concepts and exclude signals against your protocol. Then read each title and abstract and mark it Include, Maybe, or Exclude, recording a structured exclusion reason (wrong population, wrong intervention, wrong outcome, wrong study design, and so on). Screen liberally: keep anything you cannot confidently exclude. Best practice is two independent reviewers who screen the same records and then reconcile disagreements. This tool orders records by relevance, lets a classifier learn from your decisions, and exports the counts you need for the PRISMA 2020 flow diagram.
How to do title and abstract screening in Covidence?
In Covidence you import references, the platform removes duplicates, and two reviewers vote Yes / No / Maybe on each title and abstract; conflicts are resolved by a third reviewer or by discussion before full-text review. This browser-based tool follows the same two-reviewer workflow without an account or subscription: each reviewer screens in their own browser and saves a project file, then you load both files into the Reconcile step, which scores agreement with Cohen's kappa and lists every conflict to resolve. It adds relevance ranking and active-learning prioritization on top, and nothing ever leaves your browser.
What is study selection in systematic review?
Study selection is the process of deciding which records identified by your search are eligible for inclusion. It runs in two stages: title-and-abstract screening (a fast first pass to remove clearly irrelevant records) followed by full-text screening (a detailed eligibility check of the remaining reports). Each exclusion at the full-text stage must be recorded with a reason. The numbers from both stages, identified, duplicates removed, screened, excluded, and included, populate the PRISMA 2020 flow diagram and must be reported transparently so the review is reproducible.
What are the 7 steps of a systematic review?
A systematic review typically follows seven steps: (1) formulate a focused, answerable question (often using PICO); (2) write and register a protocol (for example on PROSPERO); (3) run a comprehensive, reproducible search across multiple databases; (4) screen records by title and abstract, then full text, against eligibility criteria; (5) extract data from included studies; (6) assess risk of bias and certainty of evidence (for example with RoB 2 and GRADE); and (7) synthesize the findings, narratively or through meta-analysis, and report following PRISMA 2020. This tool supports step 4, the screening and study-selection stage.
Screening hundreds or thousands of records and want PhD methodologists to run the dual screening, reconcile conflicts, and document it for PRISMA? We can take any or all of the screening workload.
Get a quote for screening support