Guide · Comparison
SyRev vs Covidence vs Rayyan for AI title/abstract screening
How the three tools use AI at the title/abstract stage, what evidence each publishes, and which might suit your review. Checked against public pages on 15 October 2026.
Dated comparison. Details for Covidence and Rayyan come from their public web pages, checked on 15 October 2026, and can change. SyRev is our product, so we have tried to be especially careful about their strengths. If anything here is out of date, tell us and we will correct it.
The short answer
The three tools take different approaches to AI at the title/abstract stage:
- Covidence uses AI to prioritise and filter: active-learning "most relevant" sorting (humans still screen every reference), and an RCT classifier, developed by the EPPI-Centre, that can tag or remove likely non-RCTs in medical and healthcare reviews, with high reported sensitivity [1]. Covidence tested criteria-based LLM auto-exclusion and chose not to release it [2].
- Rayyan offers AI relevance predictions, and its ResearchPilot add-on includes an AI Reviewer that screens all articles against your criteria as a team member, with reasoning notes and blind-mode support, plus an AI Analyzer that recommends decisions [3][4].
- SyRev scores each title/abstract against your criteria and, at its default Careful setting, auto-excludes only records with an estimated chance of inclusion below 0.1%. People screen the rest. Every review gets a validation pack with a blind sample of the exclusions [5].
Side by side (as of 15 October 2026)
| SyRev | Covidence | Rayyan | |
|---|---|---|---|
| AI role at title/abstract | Criteria-based scoring; auto-excludes clear exclusions (Careful default), humans screen the rest; AI can also act as second reviewer | Prioritisation (active learning); RCT classifier can tag or remove likely non-RCTs, medical/healthcare reviews only [1] | AI relevance predictions; ResearchPilot AI Reviewer screens against your criteria as an extra team member [3][4] |
| Does AI remove records without a human? | Yes, only below the Careful threshold; exclusions are sampled for a blind check | Only the non-RCT filter (optional); removed items reviewable via PRISMA diagram [1] | Not stated on the pages we checked; AI Analyzer "assists — never replaces" final decisions [4] |
| Published validation | 96.5% recall at Careful across 310 public review datasets (96.2% if 26 unscored relevant records count as missed); fewer than 95% in 31 datasets; citable report [5][6] | RCT classifier with high reported sensitivity on English-language biomedical records; per-feature validation and reporting templates [1] | Not stated publicly on the pages we checked |
| Per-review check | Validation pack: blind random sample of AI exclusions with Wilson CI, decision log, methods paragraph, AI-use statement | Reporting templates per AI feature [1] | Not stated on the pages we checked |
| Entry price | 2 free title/abstract evaluations; then Replay $149 or Plus $249 per review (academic) [7] | Single-review and multi-review plans, see Covidence’s pricing page; institutional licences on enquiry [8] | Free plan (3 active reviews); paid plans from $8.33/month billed quarterly, lower billed annually [9] |
Where Covidence is strong
A complete, well-established workflow from deduplication to extraction and risk of bias; a cautious, transparent approach to AI with per-feature validation and reporting templates [1][2]; mobile screening; and concessions for teams in low- and lower-middle-income countries [8]. Covidence is run by a not-for-profit [8].
Where Rayyan is strong
A free plan for getting started, a very large user base (Rayyan says 1 million+ researchers) [9], broad import formats, and ResearchPilot, which adds an AI screener and AI-assisted data extraction for teams on eligible plans [3][4].
Where SyRev is strong
Removing the clear exclusions from a large title/abstract pile using your own criteria, with public validation that includes its weakest results, and per-review evidence you can put in your methods. Dual review with AI as second reviewer, import with deduplication, human full-text screening and a PRISMA 2020 diagram that separates AI and human exclusions are on every plan [7][10].
Where SyRev is weaker
SyRev is newer and smaller than both. Data extraction is newly launched and its accuracy validation is still in progress [6]. Recall is not guaranteed: individual reviews can fall below 95%, often because of criteria wording, and diagnostic test accuracy is our weakest benchmark (94.9%) [6].
How to choose
- If your institution already licenses Covidence or Rayyan and your search pool is manageable, their AI prioritisation may be all you need.
- If you are screening RCTs in a medical topic, Covidence's non-RCT filter is a well-validated way to remove a large share of records.
- If the title/abstract pile itself is the bottleneck and you want criteria-based removal with per-review evidence, try SyRev on a review you have already finished and compare.
- Whatever you pick, validate it on your own kind of question first: see how to validate AI screening before using it in your protocol.
For a fuller feature matrix, see Using SyRev with Covidence and SyRev and Rayyan.
Sources (checked 15 October 2026)
- Covidence. Overview of all Automation (AI) features available in Covidence. support.covidence.org
- Covidence. How Covidence decides what AI to release (and what to hold back). covidence.org/blog
- Rayyan. Pricing and feature comparison. rayyan.ai/pricing
- Rayyan. ResearchPilot. rayyan.ai/researchpilot
- SyRev. Validation. syrev.ai/validation
- SyRev Team. SyRev title/abstract screening: validation on 310 public systematic-review datasets (v1.0). Zenodo, 2026. doi:10.5281/zenodo.23399663
- SyRev. Pricing & features. syrev.ai/pricing
- Covidence. Pricing. covidence.org/pricing
- Rayyan. Pricing. rayyan.ai/pricing
- SyRev. Features. syrev.ai/features
Frequently asked questions
Does Covidence auto-exclude records with AI?
Only through its optional RCT classifier, which can remove likely non-RCTs in medical and healthcare reviews. Its active-learning sorting reorders records but humans still screen all of them. Covidence tested criteria-based LLM auto-exclusion and did not release it (as of 15 October 2026).
Does Rayyan have AI screening?
Yes. Rayyan offers AI relevance predictions, and its ResearchPilot add-on includes an AI Reviewer that screens articles against your criteria as a team member and an AI Analyzer that recommends decisions (as of 15 October 2026).
How accurate is SyRev's AI screening?
At its default Careful setting SyRev kept 96.5% of relevant studies across 310 public systematic-review datasets (96.2% if 26 unscored relevant records are counted as missed). In 31 datasets it kept fewer than 95%, so every review includes a blind validation sample.
Can I use SyRev alongside Covidence or Rayyan?
Yes. SyRev imports and exports RIS, so you can remove clear exclusions in SyRev and continue screening or extraction in another tool.
Written by the SyRev Team · 15 October 2026