Suppose a reviewer asks you to discuss an earlier method. Before adding its citation, open that paper, confirm its identity, and find the passage supporting your comparison. Our suggested starting point is the references you added or changed. That is a way to prioritize the work; it does not establish that the rest of the bibliography is correct.
1. The rebuttal dates, and a disagreement in the official guidance
ICLR’s schedule lists public author–reviewer discussion for 5–18 November, followed by internal reviewer–AC discussion and meta-reviewing from 19 November. All dates here are in 2026.[1] Authors respond through comments on OpenReview.[2]
The revision cutoff needs care. The author guide allows revisions through 18 November.[2] The Area Chair FAQ instead says 27 November, even though that page also lists author discussion as 5–18 November.[3] We cannot reconcile the two from the published guidance. Our recommendation is to plan for 18 November unless the program chairs or instructions on your submission explicitly confirm a later cutoff.
Check the live schedule and your OpenReview submission before relying on a date. This note records what the linked pages said when checked, rather than resolving the discrepancy on the conference’s behalf.
2. Start with what changed
Keep a copy of the submitted PDF and bibliography. In a working copy, mark entries added since submission and entries whose title, authors, identifier, year or venue you have edited. Include references you plan to put directly into a response comment, even if they do not enter the revised PDF.
For each changed entry, keep three things together: the citation you intend to use, the source record identifying the work, and the passage supporting your new sentence. A publisher or proceedings page helps with identity. The paper itself is where you check the argument.
Crossref, for example, describes its API as access to metadata deposited by its members and other trusted sources.[4] Our recommendation to read the relevant passage adds a separate check: an accurate bibliographic record does not by itself show that your sentence represents the paper fairly.
3. A real DOI can identify the wrong paper
Consider this deliberately mismatched entry:
Vaswani et al., 2017
DOI: 10.18653/v1/N19-1423
The NeurIPS proceedings record identifies Attention is All you Need as a 2017 paper by Ashish Vaswani and seven coauthors.[5] But the DOI in this example belongs to BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding, by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova, published in the 2019 NAACL proceedings. The ACL Anthology record lists that DOI explicitly.[6]
The identifier is genuine. The entry pairs it with a different work. Checking only that the DOI opens would miss the disagreement; comparing the title and author list exposes it. To repair this example, first decide which paper the sentence actually needs, then use that paper’s own record. Changing the title to match the DOI could quietly change the cited work.
This construction illustrates a failure mode. It is not a measurement of how often authors make this mistake, an allegation about either paper, or a benchmark of StrictCite’s detection rate.
4. Five checks for each reference you add
The following is our suggested workflow, rather than an additional ICLR requirement:
- Identify the work. Open the publisher, proceedings or repository record. Compare the title and author list with the citation you intend to use. Do not stop at a familiar title fragment.
- Compare the identifier. Follow the DOI or repository link and check the work it identifies. The two primary records above show why a working link alone is insufficient.[5][6]
- Check the version, year and venue together. BERT’s arXiv record dates its first submission to October 2018; its NAACL proceedings record is from 2019.[7][6] A year difference can reflect a preprint and a conference version. Inspect which version you used before treating the difference as an error.
- Read for support. Find the passage, result or limitation your sentence relies on. Ask whether the paper supports the comparison you are making, and whether its assumptions match the way you describe it. A metadata match cannot answer those questions.
- Inspect the rendered citation. Check the response comment and compiled PDF, including links and citation keys. For another submission under review, ICLR specifies anonymous OpenReview BibTeX entries containing title, year and URL, with authors omitted.[2] Use the supplied entry; do not reconstruct an author list to make an anonymous record look complete.
Keep unresolved cases visible. Crossref provides a reporting route for incorrect or partial metadata, so a registry record can itself need correction.[4] If a lookup fails or the evidence is incomplete, investigate the primary source. Do not turn “not found here” into “does not exist.”
5. What the policy says about responsibility
ICLR’s AI policy says authors are responsible for their submission’s contents. It treats substantial LLM-produced falsehoods, plagiarism or misrepresentation as potential ethics violations that might lead to desk rejection.[8] That wording does not make every typo, unresolved lookup or checker flag an automatic rejection.
Our reason for checking additions is practical: you should know what the reference identifies and why you are citing it before asking a reviewer to rely on it. A checker finding is evidence to inspect. Decisions about a paper’s relevance, the strength of your comparison, or the adequacy of a rebuttal remain matters for human judgment.
6. Make the revision easy to inspect
ICLR’s author guide permits 10 pages of main text during rebuttal, excludes references from that limit, and describes a PDF comparison against the original submission. Reviewers and ACs may disregard changes that substantially depart from the original paper.[2]
After checking a new reference, explain the change where you answer the reviewer. Name the work, link its source, and point to the revised section. If you change a scientific claim as well as its citation, describe that change explicitly. A citation added to a paragraph should not obscure a substantive revision.
Before uploading, read the compiled PDF against your response. Confirm that the cited work, the statement it supports and the section you name still agree. For the broader formatting and compliance questions, see our source-by-source ICLR rules note and submission checklist, alongside the current official guidance.
7. Where StrictCite fits, and where it stops
StrictCite compares bibliographic details with scholarly records and lets you inspect the evidence behind a finding; our coverage page explains the sources and limits. The short trial accepts up to five references without an account. Use it on the entries you are adding, then review the records it returns.
That check does not establish that a citation supports your scientific claim, and it is not a ruling on conference compliance. Keep the paper open beside the record. Resolve the bibliographic question, read the relevant passage, and make the revision you can explain.