arXiv Citer
Citation generator - prototype, not live yet
Paste an arXiv ID, get a citation back. APA, MLA, or BibTeX.
A command-line prototype exists and is measured against 100 real papers pulled live from arXiv. There is no in-browser tool to try yet - this page exists to find out whether enough people want one before it gets built.
Nothing here is live to click yet. Leave your email and you’ll get exactly one message: the day the in-browser version ships. No account to make, no mailing list beyond that single notification.
of papers get a fully correct author-name citation
How the number was checked by hand
An automated check first reported 100% - but that check reused the same name-splitting logic the tool itself uses, so it couldn’t catch its own mistakes. A full manual review of all 103 author names flagged as higher-risk for a splitting error (every one of them, not a sample) produced this estimate:
- 93 of 100 papers get every author name split correctly.
- The automated check found no failures on year, title, arXiv ID, or citation-style structure across all 100 papers - that part of the check holds up independently.
- Every error found is concentrated in Spanish, Portuguese, Catalan, and Dutch two-part surnames, and non-Western name order - a real accuracy gap for those naming conventions, not spread evenly.
Full method: 100 papers fetched live from arXiv’s public API across 11 subject categories, 300 citations generated (3 styles each), all 103 at-risk author names reviewed by hand against known naming conventions.
What it produces
A real example, not a mockup: arXiv:2608.16889, formatted by the CLI prototype exactly as shown.
Xu, B., Shang, Y., & Ferrara, E. (2026). Don't Drop the BATON: Long-Horizon Robot Manipulation via Agentic Subtask Exploration and Transition-aware Memory. arXiv. https://arxiv.org/abs/2608.16889
Generated by the local CLI prototype, unedited.
Xu, Bingxin, et al. "Don't Drop the BATON: Long-Horizon Robot Manipulation via Agentic Subtask Exploration and Transition-aware Memory." arXiv, 17 Aug. 2026, https://arxiv.org/abs/2608.16889.
Generated by the local CLI prototype, unedited.
@misc{xu20266889,
title={Don't Drop the BATON: Long-Horizon Robot Manipulation via Agentic Subtask Exploration and Transition-aware Memory},
author={Xu, Bingxin and Shang, Yuzhang and Ferrara, Emilio},
year={2026},
eprint={2608.16889},
archivePrefix={arXiv},
primaryClass={cs.AI},
url={https://arxiv.org/abs/2608.16889}
}
Generated by the local CLI prototype, unedited.
Why not just use Zotero, Mendeley, or Citation Machine
-
One ID, zero setup
Zotero and Mendeley are reference managers built around a library you maintain - the right tool if you’re tracking hundreds of sources, overkill for citing one paper you just read. This does one thing: an arXiv ID in, a formatted citation out, no library or install.
-
The accuracy figure is published, not assumed
Citation Machine and EasyBib don’t publish an error rate for what they generate. This page states one - 93%, with the exact failure mode named - because a citation tool that hides its mistakes is more dangerous than one that admits them.
Not live yet - here’s the plan
The CLI prototype works today from a terminal; it has no web interface. If enough people want the in-browser version - paste a link, get a citation, nothing uploaded or stored - it gets built next. Your email is the only signal that decides that.