Loading...
Describe the point you need to cite, in your own words, and get back real papers: retrieved from trusted open scientific databases like Crossref and Semantic Scholar, ranked for how well they serve your argument, and formatted in seven citation styles. The AI here reads your description and ranks records; it is never allowed to write a reference, so it cannot invent one.
Sent to our server only to run this search; we keep no copy.
0 / 50 words




Trusted by researchers publishing in
Write at least 50 words on what you are arguing, studying or introducing: the claim, the population or material, the field's terminology. Add whether you need support, counter-evidence, background or methods, and set the year range and source type.
Your description becomes several database searches, run against trusted open scientific databases covering hundreds of millions of records. The returned papers are ranked against your actual question, each with a one-line note on what it contributes.
Follow the record links to read what you plan to cite, then copy individual references or the whole list in APA, MLA, Chicago, Harvard, IEEE or Vancouver, with italics that survive pasting into Word or Google Docs.
Results are retrieved from trusted open scientific databases like Crossref and Semantic Scholar, never generated. The AI ranks what the databases return and cannot add a paper of its own, so an invented reference has no path into your list.
Tell it whether you need sources that support your point, challenge it, give background or describe methods. Results carry stance labels and a one-line note on what each paper contributes, so triage takes seconds instead of an afternoon.
A 50-word description carries what three keywords cannot: the claim, the population, the kind of evidence you need. The finder searches from several angles and ranks against your question, which is how it surfaces papers a keyword box would miss.
Every result is formatted from the record's own metadata by the same engine as our citation generator: seven styles, correct italics, per-row copy and a copy-all for the full list.
The fastest way to get a reading list in 2026 is to ask a chatbot, and it is also the fastest way to get a reference list with invented entries in it. A language model asked for sources writes text shaped like sources; whether each one exists is something the model itself does not know. Search engines have the opposite problem: every result is real, but a keyword box cannot understand that you need evidence against a claim, or studies on a specific population, or methods you could borrow. Combining the two properly means giving each side only the job it is good at.
That is how this finder is built. The language model reads your description and does two things only: it writes the database search queries a subject librarian might write, and it ranks the records those searches return. Retrieval itself happens in trusted open scientific databases like Crossref, the DOI registry with publisher-deposited metadata for 160+ million works, and Semantic Scholar, the Allen Institute for AI's index of 200+ million papers. Because the model never writes a reference, the hallucination failure is not filtered out after the fact; it is structurally impossible.
The 50-word minimum is the other half of the design. Relevance ranking is only as good as the question, and a full description of your point, in your field's own terminology, lets the finder search from several angles at once: the core claim, the mechanism or population, and, when you ask for counter-evidence, the critiques. The stance labels on each result come from the paper's title and abstract, so treat them as triage, and read anything you intend to cite. A citation is a claim about your reading, not just about the paper's existence.
When your list is assembled, two companion tools close the loop. The AI citation checker verifies any reference list, from any tool, against the same open databases, and the citation generator formats individual sources from a DOI or title lookup. Our hallucinated-citation audit explains the manual checks behind all three.
Chatbots hallucinate references because the model writes them. Here the model is kept away from that job entirely.
The language model in this tool has exactly two jobs: turn your description into database search queries, and rank the records those searches return. It has no way to add a paper of its own, so the hallucinated-citation failure of chatbots is designed out, not filtered out.
Titles, authors, years, venues, DOIs and citation counts come verbatim from trusted open scientific databases like Crossref and Semantic Scholar. If a paper is not in those indexes, it cannot appear here, and nothing in a result is generated text.
Each result links to the record it came from, DOI first, so verifying a source is one click. And when your reference list is final, the AI citation checker on this site verifies every entry against the same databases as a closing integrity pass.
ProofreaderPro proofreads your full manuscript with tracked changes and never touches your citations: grammar, clarity and academic register, reviewable line by line. Built for researchers, free to try.
Proofread my writing free