How to fix common BibTeX errors

BibTeX's error messages are cryptic by design — the engine assumes you'll know what "I was expecting a comma" or "non-existent cross-reference" means in context. After helping researchers debug bibliographies for years, I've found the same six root causes account for nearly every breakage. This guide walks each one with before-and-after examples, and an editor-agnostic debugging strategy when nothing seems to fix it. You can spot-check any of these in seconds with the free BibTeX validator — paste your .bib in, and it surfaces missing fields, duplicate keys, dead DOIs, and AI-hallucinated entries against CrossRef and Semantic Scholar.

1. Missing required fields

Every entry type has a required-field whitelist. Miss one and you'll get either a silent omission in the rendered reference or, with stricter styles, an explicit "missing field" warning at compile time. The required sets are remarkably consistent across BibTeX styles but vary by entry type:

• @article → author, title, journal, year
• @inproceedings, @conference → author, title, booktitle, year
• @book → author or editor, title, publisher, year
• @incollection → author, title, booktitle, publisher, year
• @phdthesis, @mastersthesis → author, title, school, year
• @techreport → author, title, institution, year
• @misc → nothing strictly required, but at minimum title and either author or note

The classic mistake: cite a workshop paper as @article when it's actually @inproceedings. The "journal" field then produces a malformed reference because the renderer expects "booktitle." Fix it by switching the entry type — not by adding a fake journal name.

2. Unbalanced braces

BibTeX uses braces as its primary delimiter. Every { must have a matching }. When you're off by one, the parser doesn't see "missing brace on entry X" — it sees "entry X bleeds into entry X+1," and you get a cascade of mystifying errors on every subsequent entry. The error message itself usually points to a line far past the actual mistake.

Strategy: find the first reported error, then look at the entry just before that line. Most likely the closing brace of an inner field (like a title with embedded math) is missing. A common culprit:

@article{smith2023,
  title = {A study of {nested math \\(x^2 = y\\) groups},   
  author = {Smith, John},
  journal = {Physics Today},
  year = {2023},
}

That single missing brace will cause every entry after smith2023 to be silently reinterpreted. Counting braces line-by-line is tedious; the validator shows the exact entry where the count goes off.

3. Duplicate citation keys

Each entry needs a unique key — the identifier you cite with \cite{key}. Duplicate keys cause BibTeX to use the first occurrence and warn about the second:

@article{smith2023, title={Paper A}, ...}
@article{smith2023, title={Paper B}, ...}   

Common when you merge two .bib files from different projects, or when you import entries from a citation manager that auto-generates keys. Use the validator (which flags duplicates) or run biber --validate-datamodel. A naming convention helps: lastname-yearLetter (e.g., smith2023a, smith2023b) for same-author-same-year clashes.

4. Unescaped special characters

LaTeX reserves several characters for its own syntax. Inside a BibTeX field value, the same rules apply — you must escape them or wrap them in braces. The full list of LaTeX special characters: #, $, %, &, _, {, }, ~, ^, \.

In titles, the two most common offenders are & (in conference names) and % (in chemistry / statistics titles):

@inproceedings{lee2024,
  title = {Compositionality & abstraction in neural nets},
  booktitle = {Proc. NeurIPS},
  year = {2024},
}


@inproceedings{lee2024,
  title = {Compositionality \\& abstraction in neural nets},
  booktitle = {Proc. NeurIPS},
  year = {2024},
}

For underscores in chemical formulas or programming language names, wrap the whole value in extra braces: title = {Python \\_3 deployment} works; title = {Python_3 deployment} doesn't.

5. Missing commas between fields

Fields inside an entry must be comma-separated. The trailing comma after the last field is optional and harmless. But miss one between fields and the parser sees "value of next field is part of previous one":

@article{lee2024,
  author = {Lee, Jane}
  title = {Some Paper}      
  journal = {Nature},
  year = {2024},
}

BibTeX won't always crash on this — sometimes it just renders nonsense. Worse, the error often points at a downstream entry that has nothing wrong with it. Best preventive practice: always end every field with a comma. It's defensive and harmless.

6. Smart quotes and stray Unicode

Pasting bibliographic data from a publisher's website, a PDF, or a citation generator frequently introduces typographic substitutions: curly quotes “ ” instead of straight ", em-dashes — instead of --, non-breaking spaces, and the like. Classic BibTeX (8-bit) chokes on most of these. Symptoms include "I was expecting a `,'" or simply unreadable rendered references.

Two workable fixes:

1. Sanitize to ASCII. Run your .bib through a sed or your editor's find-replace to swap curly quotes back to straight ones, em-dashes to --, and non-breaking spaces to regular spaces.
2. Switch to Biber + BibLaTeX. Biber is the modern replacement for the BibTeX engine and is UTF-8 native. Add \usepackage[backend=biber]{biblatex} to your preamble, change your compile chain (latex → biber → latex → latex), and the same .bib with accented authors and special characters will compile cleanly.

7. The DOI didn't resolve

Less common but increasingly important now that AI-assisted writing tools hallucinate plausible-looking but invalid DOIs. If your reference list has a DOI field with a 404, your reader's reference-management software won't find the paper — and journals are starting to validate DOIs at submission. The validator's "Check DOIs" option resolves each one against doi.org and flags dead ones. Replace the DOI with the correct one (the BibTeX Builder can fetch a fresh entry from CrossRef given a corrected DOI), or remove the field if no DOI exists.

8. The title doesn't match a real paper

This one is new in the LLM era. A typical hallucinated citation has plausible authors, a plausible journal name, and a plausible year — but the title doesn't correspond to any paper in CrossRef or Semantic Scholar. The validator's CrossRef and Semantic Scholar checks compare your title against the matched paper's title and report a similarity score. Below 50% almost always means the citation doesn't exist. If you suspect a hallucination, paste the title into Google Scholar — if no match comes back, the citation is fake and needs to be replaced before submission.

When nothing else works

If you've been through this list and the build still won't finish, the debugging strategy is:

1. Comment everything out. Open the .bib and prefix every @ with a comment marker (most editors will let you block-comment). Build with an empty bibliography — it should compile.
2. Bisect. Uncomment half the entries. Build. Did it break? The bad entry is in that half. Halve again, and so on. Five or six iterations is enough to locate the bad entry in a 500-line file.
3. Inspect the entry. Once isolated, the issue is almost always one of the six categories above. Most often missing braces.

Better: don't get there in the first place. Run your .bib through the BibTeX validator after every meaningful change. It catches all eight categories above in under a second, never uploads your file, and produces an annotated .bib you can paste back into your project with inline warning comments. If you'd rather build the file from a DOI list, the BibTeX Builder goes the other direction.