Academic plagiarism checker.

For most cases, no — if your target journal mandates iThenticate (Elsevier, Springer, Wiley, IEEE workflows), they'll run it at submission regardless. What Noplag replaces is the iterative self-check across drafts before that submission. iThenticate at $125/document gets prohibitive after the 4th revision; flat-rate Pro doesn't. See the full vs-iThenticate comparison at noplag.com/vs-ithenticate.

Correct — that gap is real. ProQuest's dissertation database is proprietary, paywalled, and not in OpenAlex/CORE. We cover the ~700M open-access scholarly works, which is broader than ProQuest for post-Plan-S research (most major journals 2018 onwards); narrower for older paywalled dissertations specifically. If cross-checking against US PhD dissertations is your dominant requirement, iThenticate's the right tool for that one job.

MLA, APA, Chicago / Turabian, Harvard, Vancouver, and IEEE auto-detect from the bibliography format. For less common styles (ACS, AMA, ASA, Bluebook, Oxford), the engine still classifies intervals based on quotation marks + nearby parenthetical patterns, but the style-specific in-text validator is on the v1.2 roadmap. Self-host users can patch in additional styles via the alignment/ module.

Lower than uncited-paraphrase reviewers expect, because the classification model excludes properly-cited intervals from the headline score. We measure ~3.4% false-positive rate on the PAN-PC-11 academic-prose subset (where most intervals are legitimate scholarly quoting). The closed-source competitors who don't classify cited intervals report 12–18% on the same corpus.

Pro tier — add a self-citation allowlist (your ORCID, your past DOIs, your institutional repository URL). Matches against allowlisted documents tag as SELF-CITE and don't count toward the score. Useful for academics whose grant proposals build on their previous published work; self-recycling with citation is expected, not plagiarism.

Detection is calibrated for English, Spanish, Portuguese, Polish, and Ukrainian at launch, with more languages rolling out. Detection quality is benchmarked on PAN-PC-11; per-language evaluation is on the roadmap. Strongest on the European languages with the longest validation history; newer additions ship as their calibration matures.

Your IT / DPO / IRB can clone github.com/NoplagLabs/noplag-engine, run the engine on a sandbox, audit the data flow, verify right-to-erasure works end-to-end, and reproduce our published benchmark numbers against their own test corpus. Without signing anything, without an NDA, without trusting our marketing. For institutions whose procurement process requires this depth of review, that's the unlock.

AI detection is coming soon — it's on the v1.2 roadmap, not live at launch. The approach is Binoculars (Hans et al. 2024), which is model-agnostic: it doesn't train per-LLM, it compares a small reference model's perplexity against the test text, which generalises. Accuracy figures publish when the feature clears evaluation, rather than being quoted ahead of it, on mixed-LLM content (vs ~70% on Claude-3 for GPT-trained detectors). The methodology paper documents the failure modes; we don't hide them.

Every flagged interval ships with the source DOI, the corpus snapshot timestamp, the engine commit, and the classification reason. Re-run the check against that engine commit and the result reproduces exactly. Bring the printed report to the meeting; the citation paths are click-through. We don't have a “trust the score” mode, because that's what black-box checkers have.

Apache 2.0 + Docker — runs entirely inside your network. Pair with PostgreSQL + Redis on your own infra; no traffic leaves the VPC. Useful for clinical drafts, patient case studies, or any pre-publication research that can't go to a third-party cloud. Enterprise customers get the corpus shipped as an indexable bundle.