Don't Let AI Hallucinations
Ruin Your Research.

Yes. The Free plan requires a login (no anonymous usage) and includes 50 credits every month, no credit card required.

Every plan gives you one shared pool of credits instead of separate verification and finding limits, so you can spend them on whichever you need more of. A citation verification costs 1 credit; finding a new supporting citation costs 4 credits, since it takes more work — searching for and relevance-checking a real source rather than just looking one up. See the Pricing page for per-plan credit totals.

Verify accepts BibTeX, RIS, plain text, or PDF (Pro), and detects citations written in APA, MLA, Chicago, Harvard, IEEE, or Vancouver style.

We parse each reference, cross-check it against the academic databases above, and fuzzy-match the returned metadata against what you provided. If nothing matches closely enough, we flag it as unverified and suggest a real paper to replace it, instead of assuming it's genuine. We also separately flag retracted papers, since a citation can be real and correctly matched and still be one you shouldn't be citing. On our own leak-free benchmark, this catches 99.4% of fabricated citations at 90.1% overall accuracy — see the full benchmark methodology.

We benchmarked it against two alternatives on claims about recent, lightly-cited research — deliberately the hardest case, chosen so no AI model could just recall a famous paper from memory instead of actually searching. Every citation we returned was a real, verified paper — never invented, unlike asking an AI directly with no search, which fabricated an outright non-existent citation 53.1% of the time. Checking each candidate's abstract also gets a precise, on-topic match 56.2% of the time per citation, well ahead of both a naive title match (24.3%) and asking an AI directly (32.1%). See the full benchmark, including what it doesn't cover.

Yes. A real, correctly-matched citation can still be flagged as retracted — we check retraction metadata already published by OpenAlex, Crossref, and PubMed, corroborated across sources the same way the rest of verification is. When finding sources for a claim, a retracted paper is excluded from results entirely rather than being recommended with a warning.

No. Uploaded PDFs are processed in memory and discarded once your verification job completes. We keep the verified citation metadata (title, DOI, etc.) to speed up future searches — see our Privacy Policy for details.

Yes. REST API access is available on every plan, including Free, drawing from your plan's regular credit pool. See the API Docs for endpoints and authentication, or use the MCP server to call it directly from Claude or Cursor.

Yes — the Bulk Verification plan lets you drop up to 30 PDFs into a single scan (handy for a teacher or reviewer checking a stack of student papers at once) and get one report with a per-file breakdown of what verified.

Most tools check a citation against one database and stop at the first match. We query all 6 sources at once and look for independent agreement between them, separately verify that DOIs actually resolve to the right paper, and tell you (with a refund) when we can't find real support for a claim instead of padding results with a loosely related paper.