DISSERTATION50,000-word checks · self-citation aware

Dissertation plagiarism checker.

Check your dissertation before defence. Premium tier handles 50,000 words in one pass. Self-citation across chapters tags as SELF-CITE (legitimate at this length, not a flag). Embargo-aware for cumulative-PhD chapters already submitted as journal papers. EU residency for sensitive primary-source research.

0 / 1,500 words
Up to 50,000 wordsSelf-citation across chaptersEU residency · IRB-reviewable
WHY DISSERTATIONS NEED SPECIAL HANDLING

Four things about a 60,000-word doc that a 1,500-word essay tool gets wrong.

The classifier is tuned for long-form scholarly work: self-citation across chapters is expected, bibliography auto-exclude is critical, and the report has to reproduce months later when your viva voce panel asks.

01 · LENGTH

50,000-word check in one pass

Premium tier handles a typical dissertation (60,000–80,000 words) in chunks of 50,000 per check — usually two passes covers the whole thesis. Chapter-by-chapter is the more common workflow during writing. Free tier (2,500 words) only covers a single section excerpt; useful for a quick sanity check, not whole-thesis review.

02 · SELF-CITATION

Your own prior work is expected

Cumulative PhDs explicitly include your already-published journal articles as chapters. Add your ORCID + DOI list to the allowlist; the engine tags those intervals as SELF-CITE and excludes them from the headline number. The same applies to passages reused from your master's thesis or qualifying papers.

03 · BIBLIOGRAPHY EXCLUSION

100+ references can't inflate the score

Auto-exclude the references list (detected from heading + format). At dissertation length you might have 150–400 sources cited; flagging the references themselves would push the headline to 18–22% before you start. We don't.

04 · MULTI-YEAR TIMELINE

Drafts span 3–5 years of writing

Engine commit + corpus snapshot stamped on every report. Re-run the same chapter against the same commit two years later and the verdict reproduces exactly. Your January 2024 self-check against your viva in October 2026 still works — and the report is auditable months later.

THE DRAFTING TIMELINE

Three milestones. The check fits each one differently.

Most PhDs follow a similar arc: chapter drafting → full-thesis assembly → defence preparation. The check serves a different purpose at each.

01

01 · Chapter drafts (months 1–24)

Self-check each chapter as you finish it. Catch the uncited paraphrase before your supervisor reads chapter 2 for the third time. Use the allowlist (ORCID + co-author DOIs) so legitimate self-citation across your prior papers doesn't inflate the score. Iterate without burning per-document credits — flat-rate Premium means dozens of revisions cost nothing extra.

02

02 · Full-thesis assembly (months 24–32)

Run the full thesis (Premium: 50,000 words at a time; for an 80,000-word dissertation, two passes cover everything). Bibliography auto-excludes. Cross-chapter self-cite catches passages you rewrote in chapter 5 that originally appeared in chapter 2 — sometimes legitimate, sometimes a draft-trail issue you'd want to know about.

03

03 · Defence prep (months 33–36)

Re-run the final version against the engine commit your earlier reports used — reproduces exactly. Print the per-interval report; bring it to your viva. If a panel member asks “this passage looks like your 2022 paper” the report already shows it as SELF-CITE with the DOI cross-link. Reproducible months later for any post-defence review.

SELF-PLAGIARISM HANDLING

Six categories of “recycling your own work” — none of which should fail your viva.

PhD candidates legitimately recycle text from prior publications, conference papers, qualifying exams, and master's theses. The classifier separates the expected categories from the ones that actually need attention.

01 · PRIOR PUBLISHED PAPERS

Your own journal articles (ORCID + DOI allowlist). Cumulative PhDs explicitly include these as chapters — flagged as SELF-CITE, excluded from the headline. The journal's own copyright statement might require attribution; the classifier tells you which sections need a “this chapter is based on...” preamble.

02 · CONFERENCE PAPERS

Add conference-paper DOIs (or ACM/IEEE/ACL anthology URLs) to the allowlist. Common at the PhD level — a conference paper becomes the seed of a thesis chapter. Tag as SELF-CITE-CONF; some institutions explicitly allow this with attribution, others want a substantive rewrite. The report shows where the recycling sits.

03 · MASTER'S THESIS

Add your master's thesis URL to the allowlist (institutional repository, often ProQuest or your university library). Background-chapter literature reviews often share structure across master's → PhD. Flagged as SELF-CITE-MASTERS; some institutions allow, others require explicit disclosure in the preface.

04 · QUALIFYING / COMPS

Some institutions allow incorporating your qualifying exam material or proposal/prospectus into the dissertation; others require it be fully rewritten. The classifier separates these recycled passages and surfaces them; your dean of graduate studies has the policy. The engine doesn't make that call for you.

05 · CO-AUTHORED CHAPTERS

Some PhDs incorporate co-authored papers as chapters. Add your collaborator's ORCID to the allowlist; matched intervals tag as CO-SELF-CITE. The standard requirement is a clear preface noting your specific contribution to each chapter (typically a CRediT statement). The report shows which intervals came from co-authored work.

06 · UNATTRIBUTED RECYCLING

The genuinely-problematic category: passages reused from your own prior work without attribution — including unpublished drafts. Counted at full weight in the headline. Even self-recycling needs declaration; most institutions consider undeclared self-plagiarism an integrity violation even if technically you wrote both versions.

THE CHAPTER REPORT

What a chapter report looks like during drafting.

Per-paragraph classification with source DOIs. The headline number reflects only the unattributed categories. Reproducible against the engine commit months later.

DISSERTATION · CHAPTER 4 · 8,420 words · 64 intervals · allowlist activeChapter 4: Statistical learning on noisy graphs
Headline · 3.2%SELF-CITE · 24%v0.4.2
#INTERVAL · SOURCECLASSIFICATION
08
We define a noisy graph as a tuple (V, E, p) where p assigns each edge an independent corruption probability.Your NeurIPS 2023 paper, §2.1 · ORCID + DOI in allowlist · cumulative-PhD chapter source
SELF-CITE
12
Recent advances in graph neural networks have focused on stability under structural perturbations (Kim et al. 2022).openalex.org/W93b · attribution + (Kim et al. 2022) present
PARA·CITED
14
The Laplacian spectral gap controls the rate of information diffusion across the network.Hammond et al. (2011), p.142 — cited 4 paragraphs ago, classifier recognises ongoing attribution
PARA·CITED
19
We adopt the convention that edge weights are non-negative and the graph is undirected.Boilerplate convention statement · matches 2,300+ graph-theory papers — common phrasing
COMMON-KNOWLEDGE
23
Robustness to label noise is essential for real-world deployment of semi-supervised classifiers.Co-author Zhang's ACL 2023 paper · allowlist via co-author ORCID · co-authored chapter source
CO-SELF-CITE
28
The algorithm converges to a local minimum in O(n log n) steps under our assumptions.Your unpublished tech report (Feb 2024) — not in allowlist, no attribution this section
UNATTRIBUTED
41
Bibliography — Kim, Hammond, Zhang, and 187 others.References section, pp. 240–248
REFS · EXCLUDED
Headline 3.2% = unattributed only. SELF-CITE + CO-SELF-CITE + CITED + COMMON + REFS all excluded.Engine v0.4.2 · Export PDF · Reproducible at viva
50,000words per check (Premium)
700M+scholarly works indexed
6self-citation categories
5detection languages at launch
EUresidency · IRB-reviewable
WHERE WE DON'T MATCH ITHENTICATE

Two honest concessions for doctoral candidates.

01 · PROQUEST GAP

The dissertation database your discipline expects

iThenticate ships with ProQuest's dissertation database (~5M US dissertations); we don't have access to it. If your committee specifically wants cross-checking against prior US PhD dissertations, iThenticate is the right tool for that one job. We cover ~700M open-access scholarly works (broader for journal articles, narrower for the dissertation back-catalogue). Honest concession.

Read the iThenticate comparison
02 · JOURNAL MANDATES

When your target journal requires iThenticate at submission

Some journals (Nature, Cell, NEJM, Elsevier titles) run iThenticate at submission via Crossref Similarity Check. That happens regardless of what you ran first — you can't substitute. What Noplag IS for: the iterative self-check across every draft before that submission, where iThenticate's per-document credits ($125/doc) make the dozens-of-revisions workflow unaffordable. Use both.

Read the workflow guide
DEFENCE-READY

Reproducible months later. Bring the printed report to your viva.

Every report stamps the engine commit (e.g. v0.4.2) and the corpus snapshot timestamp. Re-run the same chapter against that commit two years later and the verdict reproduces exactly — no silent retraining, no “the algorithm changed” surprise mid-defence. The classification of every interval is reviewable: a SELF-CITE tag links to the source DOI; a CITED tag links to the matched paragraph; an UNATTRIBUTED tag links to the original source URL. Apache 2.0 at github.com/NoplagLabs/noplag-engine — your viva voce panel can read the threshold logic, the AI calibration, the dedupe rules. Closed-source checkers fail this test because their report depends on what their algorithm was on the day of the check; if the model has been updated since, the report you brought to your defence might not reproduce. Ours does.

Read the viva-prep guide
DISSERTATION CHECKLIST
ALLOWAdd ORCID + DOI list for your prior journal papers · conference papers · master's thesis · co-author ORCIDs
BIBBibliography auto-detected by heading + format. Excluded entirely. References don't inflate your headline score.
CHAPTRun per-chapter during drafting. Full-thesis pass during assembly. Final pass during defence prep. Same engine, same report shape.
COMMITEvery report stamps engine commit + corpus snapshot. Reproducible at viva voce months or years later.
EUEU residency endpoint for sensitive primary research. Self-host inside your institution if your IRB requires it.
DOIEvery matched interval ships with source DOI · canonical URL · snapshot timestamp. Click-through for any panel review.
FAQ

What doctoral candidates actually ask.

Can I check the whole dissertation in one pass, or do I have to split it?
Premium tier handles up to 50,000 words per check. For a typical 60,000-80,000-word dissertation, that means two passes (chapters 1-4, then chapters 5-8) or four passes if you want chapter-level granularity. Chapter-by-chapter during drafting is the more common workflow; full-pass during the final assembly + defence prep phase.
How does the self-citation allowlist actually work?
Drop your ORCID into the allowlist; the engine pulls your DOI list via the ORCID public API. Add specific DOIs for conference papers / master's thesis / co-author papers manually. Matched intervals against allowlisted sources tag as SELF-CITE, CO-SELF-CITE, or SELF-CITE-MASTERS — excluded from the headline score. Saves you arguing about “isn't this my own work” at the viva.
What about ProQuest dissertations? You don't have those.
Correct, honest concession. ProQuest's dissertation database (~5M US dissertations) is proprietary, paid, and not in OpenAlex/CORE. If cross-checking against US PhD dissertations specifically is your committee's requirement, iThenticate has the dataset we don't. We cover ~700M open-access scholarly works (broader for journal articles, narrower for older dissertations specifically).
Will my draft chapters be stored or used to train AI?
No. Drafts are fingerprinted (winnowing hashes, non-reversible signature) and stored in your tenant-isolated database. The text itself is purged after 30 days unless you opt into retention (most dissertation candidates do). Detection models are pre-trained on public corpora; we don't add candidate manuscripts to training. Explicit in the DPA.
Reproducibility at the viva — what does that actually mean?
Every report stamps engine commit (e.g. v0.4.2) + corpus snapshot timestamp. If you save your January 2024 check and re-run the same chapter against v0.4.2 in October 2026, the verdict reproduces exactly. The classification of every interval is reviewable: SELF-CITE → source DOI, CITED → matched paragraph, UNATTRIBUTED → source URL. Closed-source checkers fail this because their report depends on the algorithm-of-the-day.
My dissertation includes patient data / classified research. Can I self-host?
Yes. Apache 2.0 + Docker, runs entirely inside your institution's network. No traffic leaves the VPC. Useful for clinical research, IRB-protected data, classified material, and any dissertation under protective order. The engine doesn't need internet access for indexed-corpus checks; the live-web layer is optional.
How will the AI detector handle a 60,000-word doc with mixed AI / human paragraphs?
AI detection is coming soon — it's on the roadmap, not live at launch. When it ships, the plan is per-paragraph AI-likelihood (Binoculars, Hans et al. 2024). The verdict will be paragraph-level, not document-level — so a chapter where you used Grammarly for the first draft and rewrote in your own voice gets different verdicts per paragraph. An ESL-adjusted band kicks in when the lexical signature suggests non-native English. The methodology is open; you'll be able to defend it at viva.
What's the deal with cumulative PhDs — are journal-paper chapters allowed?
Many institutions explicitly allow it (Germany, Netherlands, Scandinavia common; UK + US increasingly). The journal's copyright statement often allows reuse with attribution in your dissertation. Add the published DOIs to your allowlist; the engine tags those intervals as SELF-CITE. The institutional policy + your specific journal's reuse policy determines whether you need to rewrite or just disclose.
What if my supervisor or committee disputes a flag?
Every flagged interval ships with source DOI + canonical URL + snapshot timestamp + engine commit. Pull up the original source side-by-side; show the per-interval classification reasoning. The engine is Apache 2.0 at github.com/NoplagLabs/noplag-engine — your committee can read the threshold logic line by line. This is what “defensible” actually means: methodology citation + reproducible verdict + open algorithm.
How much does it cost for the 3-5 year writing period?
Premium $79/mo unlimited checks during active writing; cancel anytime and resume later (your folders persist). Most doctoral candidates run 6-12 months of Premium across the 4-year program, paused during non-writing periods. Flat-rate matters here because per-document credits (iThenticate's $125/doc model) would cost thousands across hundreds of revisions.

Run a chapter. Reproduce the report at your viva.

Drop a draft chapter into Noplag. Add your ORCID. See the SELF-CITE, CITED, UNATTRIBUTED classification. Free tier covers 2,500 words for a section sanity check; Premium $79/mo handles 50,000 words per check for whole-thesis review.

Dissertation plagiarism checker — self-citation aware