Find Source
For every flagged sentence, Find Source surfaces real peer-reviewed papers you can cite to attribute the idea properly.
Audit your draft for academic integrity. Every claim missing a citation gets flagged. Every citation that's there gets resolved. Then the same verifier behind Verify Source checks whether each source actually backs the claim it's attached to. Paywalled reference? Drop in the PDF and we'll check support against your copy.
Drop PDF, TXT, or DOCX here
or
This isn't a text-overlap hunt. It's an audit of your draft's intellectual integrity. Every sentence gets two questions. Is it cited? And if so, does the cited source actually back the claim?
Each sentence gets scanned for an in-text marker. APA-style (Smith et al., 2023). Chicago author-year. Numeric [12]. If a sentence makes a substantive claim with no marker attached, it gets tagged missing citation. That's the most common kind of accidental plagiarism. For each flagged sentence there's a one-click jump to Find Source, which queries Crossref, Semantic Scholar, arXiv, OpenAlex, and PubMed for a paper you could actually cite.
Every cited sentence gets its marker resolved to a real reference through Crossref or Semantic Scholar. Then the same verifier behind Verify Source runs the cited paper's abstract against your sentence. The verdict lands on supported, partial, unsupported, or needs review. Cited-but-unsupported is the one that matters. Real integrity issue. Not a phrasing coincidence.
Not every reference is open access. Numeric markers like [12] won't resolve without a bibliography, and plenty of journals lock their abstracts behind a paywall. Sentences in the needs source PDF state show an inline upload button. Drop your copy of the cited paper in. The verifier reruns against the uploaded text and returns the same verdict.
Your text is processed in memory and dropped the second the result comes back. Nothing stored. Nothing used for training. The privacy policy spells out the specifics.
Turnitin and GPTZero hunt for text overlap. Sentences whose wording matches something in their database. That catches copy-paste plagiarism and misses the deeper failure: a paper full of cited claims where the cited source doesn't actually back the claim. We treat the citation as the unit of integrity. Uncited claims get flagged as missing attribution. Cited ones get verified source-by-source. Run both kinds of tool. They catch different failures.
For each cited claim, we pull the (Author, Year) marker, hit Crossref and Semantic Scholar for the matching paper, and grab its abstract. The abstract plus your sentence go to a verifier model. Four possible verdicts. Supported (the abstract clearly backs the claim). Partial (related but doesn't fully support). Unsupported (off-topic or contradicting). Unknown (abstract too short or missing to judge). Same verifier Verify Source uses when you highlight an in-text citation inside a paper.
If we can locate the paper but its abstract isn't public, the sentence goes into needs source PDF and the detail panel shows an upload button. Drop in your local copy and we rerun the verifier against the uploaded text. 1 credit per upload-verify. Numeric citations like [12] land here too, since they can't be auto-resolved without your bibliography in hand.
Missing citation means the claim has no in-text reference at all. If the idea came from a source, even a paraphrase, it needs attribution. Unsupported means the claim has a citation, but the cited paper doesn't actually back it. Unsupported is the more serious one. It usually means the citation got added without anyone reading the source, or the source got misread.
Short sentences (under six words) and topic-sentence fragments get auto-skipped. Longer non-claim sentences like "In this section we describe our methodology" can still get tagged "missing citation" because they look like a claim. Review and dismiss. The detail panel always shows the exact phrasing and the reason it was flagged.
Input cap is 25,000 characters, roughly 3,500 words. The audit covers the first 30 sentences in detail to keep response time reasonable. For longer drafts, run separate passes on each section.
No. Submitted text lives in memory on the server and is discarded once the result is returned. We don't store audit results. Nothing you submit goes near a training run. Uploaded source PDFs get the same treatment. The privacy policy has the specifics.
For every flagged sentence, Find Source surfaces real peer-reviewed papers you can cite to attribute the idea properly.
Once you know which source to cite, the citation generator formats it in APA, MLA, Chicago, Harvard, IEEE, or any of 9,000+ CSL styles.
Need to rephrase a flagged passage in your own voice? The Writing Assistant rewrites prose while preserving the meaning, then lists every change it made.
AI-powered features require an account. The Citation Generator stays free forever.