Short answer: Claude generally produces fewer hallucinated references and more traceable citation outputs for long PDF methods sections. ChatGPT can match this if you supply exact page-extracted text and force verbatim quoting.

To minimize hallucinations: extract text by page into the prompt; require page numbers + verbatim quote + DOI/URL or explicitly "not found"; ask the model to cite the original PDF page for each reference and include the quoted snippet for verification.

Compare notes: Compare Claude and ChatGPT