Recommendation
Use a two-part prompt that (A) forces machine-readable output (BibTeX/CSL-JSON) and (B) returns human-formatted citations in your chosen style, plus a short verification report per item (DOI resolves, publisher page found, PDF available). If you need live-checks, run the prompt in a model or environment with web access (ChatGPT with Browse or an LLM that can query CrossRef/PMC).

Decision criteria (pick before you prompt)
- Citation style needed: APA / MLA / Chicago / Vancouver / BibTeX/CSL-JSON
- Output use: human read (formatted list) vs import to reference manager (BibTeX/CSL-JSON)
- Verification level: quick (DOI resolves) vs thorough (publisher page + PDF + license)
- Tools available: local scripts + CrossRef API vs LLM with browsing capability
- Team & workflow: single-researcher = manual checks OK; team/shared bibliography = enforce machine format + checks

Repeatable prompt template (copy + paste and replace placeholders)
System message (set model role):
"You are an academic reference assistant. Always output two clearly labeled sections: (1) Formatted citations in [STYLE] (e.g., APA 7) numbered; (2) Machine-readable entries as CSL-JSON and BibTeX. For every item include metadata fields: authors, year, title, journal (or book), volume, issue, pages, DOI (as https://doi.org/...), URL, access_date, and a verification block with: DOI_resolves (yes/no), publisher_page_found (yes/no + URL), pdf_available (yes/no + URL). If you cannot verify, state which checks you could not run. Return strictly parseable JSON at the end in key 'bibliographyJSON'."

User message (example):
"Please process these items and return (A) formatted citations in APA 7, (B) BibTeX and CSL-JSON, and (C) a short verification report for each. Limit summaries to 20 words each. Output final machine JSON exactly under 'bibliographyJSON'.

Items:
1) DOI: 10.1038/s41586-020-2649-2
2) Title: 'Deep learning for genomics' — if multiple matches, pick the most-cited one and give the DOI.
Style: APA 7
Return only the requested sections."

Example small output snippet (what to expect)
- Formatted: 1. Smith, J., & Lee, A. (2020). Deep learning for genomics. Nature, 580(7804), 123–130. https://doi.org/10.1038/s41586-020-2649-2
- BibTeX: @article{smith2020deep,...}
- Verification: DOI_resolves: yes (https://doi.org/10.1038/...), publisher_page_found: yes (https://www.nature.com/...), pdf_available: no (paywalled)
- bibliographyJSON: [{...}, {...}]

Verification tips (practical & repeatable)
- Automated quick checks: curl https://doi.org/ (HTTP 200/301 means resolves). Use CrossRef REST: https://api.crossref.org/works/
- Google Scholar / PubMed / CrossRef to confirm titles/authors match exactly.
- Check publisher landing page for official PDF/open-access flag; use Unpaywall API for OA status.
- For arXiv/PubMed Central, prefer canonical URLs (arxiv.org/abs/..., ncbi.nlm.nih.gov/pmc/...).
- Save the publisher URL, DOI URL, and access_date for reproducibility.

Practical checklist to validate a citation before using it in a literature review
1. Confirm title + author list match across CrossRef/Google Scholar. 2. Check DOI resolves (curl or browser). 3. Open publisher landing page and record URL. 4. Note PDF availability (download or Unpaywall). 5. Export BibTeX/CSL-JSON and import to your reference manager. 6. Record access date and confidence (0–100).

Best-for / Avoid-if
- Best for: fast creation of formatted refs + machine export; initial lit-scan and shared bibliographies.
- Avoid if: you require legally guaranteed accuracy (final thesis submission) or the model has no web access — always run the verification checklist yourself.

When the right answer depends
- Budget/skill: If you can pay for a browser-enabled LLM session or use CrossRef API scripts, you’ll automate checks. If not, plan for manual verification with Google Scholar and your library.
- Team size/workflow: For teams, enforce machine-readable exports (CSL-JSON/BibTeX) and a required verification step in PRs.

If you want, I can give a ready-to-paste system+user prompt tuned for APA+BibTeX that outputs strict JSON you can pipe into a script. (I recommend running it in ChatGPT with browsing enabled.)

Compare ChatGPT and Gemini