Why Manual Research Also Needs a Completeness Check
A reference compiled by hand can be just as structurally incomplete as one generated by an AI tool, for entirely different reasons. A source noted during reading and added to the list later from memory can lose a field in the process. A URL copied early in a research project can break by the time the paper is finished, with no record of an alternative identifier. A reference pulled from another paper's bibliography, copied without independently verifying every field, inherits whatever gaps the original had.
These gaps don't carry the same fabrication risk as an AI-hallucinated reference, but they create the same practical problem: a reviewer, editor, or examiner can't verify the source. A structural check catches both categories the same way, because the test is about what's present in the entry, not about how the entry was created.
What Gets Tested, the Same Way Regardless of Source
Every entry is checked for the same five fields: an author name in the position expected for the detected citation style, a four-digit publication year in a plausible range, a title fragment long enough to be a real title, a publisher or journal name, and an identifier — a DOI matching the standard format or a working URL. A reference passes when all five are present; it's flagged when one or more — most significantly the identifier — is missing.
This is deliberately the same test applied across every page on this site, whether the stated concern is a ChatGPT-drafted literature review, a student dissertation, or — as on this page — a general bibliography with no assumption about how it was compiled.
When a Mixed Reference List Is the Most Common Case
Most real reference lists today aren't purely manual or purely AI-assisted — they're a mix, where some entries were found through traditional research and others were suggested or drafted with help from ChatGPT, Gemini, or Claude during outlining or revision. A structural checker doesn't need to know which entries fall into which category; it tests all of them the same way and returns a report showing exactly which specific entries need a closer look, regardless of where they came from.