--- title: ChatGPT Reference Checker - Verify AI Citations Are Real description: Paste the references ChatGPT, Claude, or Gemini gave you and check whether they are real. Each identifier is resolved against the scholarly registries and compared to the claimed title, catching the dominant AI fabrication pattern (real DOI + invented title - Topaz et al., Lancet 2026). Free, no signup. doc_version: "2026-07-16" last_updated: "2026-07-16" --- # ChatGPT Reference Checker *Paste the references an AI chatbot gave you. The checker resolves each identifier and compares the resolved record to the claimed title, flagging the ones that are fabricated.* ## What this tool does ChatGPT, Claude, Gemini, and Copilot all produce references that read like the real thing - real author names, real journals - but with invented titles or identifiers. This checker verifies each reference against the scholarly registries (Crossref, PubMed, arXiv, and more) instead of trusting how it looks. It runs the same engine as the general [Citation Verifier](/tools/citation-verifier); this page is the workflow for AI-generated reference lists. ## Why AI references look real The dominant fabrication pattern, documented by [Topaz et al. (Lancet 2026)](https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(26)00603-3/fulltext) across 2.5 million biomedical papers, is a **real, resolvable DOI paired with an invented title**. The DOI opens a genuine paper - just not the one the citation claims - so clicking the link does not catch it. By early 2026 roughly 1 in 277 biomedical papers contained at least one fabricated reference. The only reliable check is comparing the claimed title against the record the identifier actually resolves to, which is what this tool does. ## How to check a ChatGPT reference list 1. **Get the list in a structured format.** Ask the chatbot to *output the references as BibTeX, including DOIs* (RIS or CSL-JSON also work). 2. **Paste it into the batch checker.** Up to 10 entries are verified in one pass, each against the appropriate registry: Crossref for DOIs (doi.org and DataCite as fallbacks), PubMed for PMIDs and PMCIDs, arXiv for preprints, OpenLibrary for ISBNs, NASA ADS for bibcodes, WHO IRIS for institutional URLs. ISSN is accepted for journal-level checks. 3. **Act on the verdicts.** Keep the matched entries, fix the ambiguous ones (real paper, wrong identifier), and replace or remove every mismatch and not-found. ## Verdicts, read for AI output - **Matched** - the claimed title agrees with the resolved paper. Safe to cite (still read it!). - **Mismatch** - the identifier resolves cleanly but the title disagrees. This is the hallmark AI fabrication: real DOI, invented title. Do not cite. - **Ambiguous** - the claimed title matches a real paper found via search, but the identifier points to a different paper. The chatbot attached the wrong identifier to a real work - fix the identifier. - **Not found** - neither the identifier nor a title-search returns a match. The reference is invented end to end - remove it. Scope note: this is a **syntactic check** - identifier resolution plus bibliographic field agreement. It does not verify that the resolved paper supports the claim you are citing it for. ## Programmatic access - REST: `POST /api/verify` (single) and `POST /api/audit` (whole bibliography) - see [API documentation](/docs). - MCP: the `verifyCitation` and `auditBibliography` tools ship with the [scholar-sidekick MCP server](/mcp); call them from Claude Desktop, Cursor, or any MCP-aware client. - Inside ChatGPT itself: the [Scholar Sidekick ChatGPT app](/integrations/chatgpt) verifies citations without leaving the conversation. ## FAQ ### Why does ChatGPT make up references? Because a language model generates plausible text, it does not look anything up. A citation is produced the same way as any other sentence - token by token - so the output often blends real author names and real journals with an invented title or identifier. The result reads exactly like a genuine reference. Topaz et al. ([Lancet 2026](/citation-integrity)) audited 2.5 million biomedical papers and found the steep rise in fabricated references since 2023 strongly implicates LLM-assisted writing: by early 2026, about 1 in 277 biomedical papers contained at least one fabricated reference. ### The DOI opens a real page - doesn't that prove the reference is real? No, and this is the single most important thing to understand about AI references. The dominant fabrication pattern is a **real, resolvable DOI paired with an invented title**: the link resolves cleanly to an actual paper - just not the paper the citation claims. Clicking through only proves the identifier exists. The checker catches this by resolving the identifier and comparing the record it actually points to against the title you were given, field by field. ### Does this work for Claude, Gemini, and Copilot references too? Yes. The fabrication pattern is model-agnostic - the checker examines the citation itself, not which chatbot produced it. References from ChatGPT, Claude, Gemini, Copilot, Perplexity, or any other AI assistant are checked the same way: resolve the identifier, compare the resolved record to the claim. ### Can I check a whole ChatGPT reference list at once? Yes - the checker above opens in **Batch** mode. Paste a BibTeX, RIS, or CSL-JSON bibliography (or upload a `.bib` / `.ris` / `.json` file) and up to 10 entries are verified in one pass, with a per-row verdict table. Practical tip: ask the chatbot to *"output those references as BibTeX, including DOIs"* and paste the result straight in. For larger bibliographies, call [POST /api/audit](/docs) or the `auditBibliography` tool on the [MCP server](/mcp). ### A reference was flagged - what should I do? Depends on the verdict. **Mismatch** (identifier resolves, title disagrees) is the classic AI fabrication - do not cite it; check the candidates panel for the real paper behind that title, or ask the chatbot for a source you can verify. **Ambiguous** means the claimed title matches a real paper found via search but the identifier points elsewhere - the paper exists, so fix the identifier and re-check. **Not found** means neither the identifier nor the title resolves anywhere - remove the reference entirely. ### Is this free? Do you store what I paste? The web checker is free at the anonymous tier with a published rate limit (see [API documentation](/docs)). No signup. Nothing you paste is stored beyond standard server logs. Programmatic access via the REST API and MCP server is metered on [RapidAPI](https://rapidapi.com/scholar-sidekick-scholar-sidekick-default/api/scholar-sidekick) (free tier available; paid plans scale up). ### How accurate is the checker? On a 1,395-entry blind holdout - drawn from a recorded seed after the code was frozen, then measured exactly once (plus a repeatability re-run, 99.9% stable) - the verifier caught every fabrication on the dominant patterns (150/150 = 100%, Wilson 95% CI lower bound ~97.6%) and made high-confidence false-accusations on correctly-cited papers at 0.8% (95% CI 0.4-1.4%; just 0.17% were confident mismatches). We also report a measured blind spot - single-word near-miss semantic flips (caught 4/30). The fixtures, methodology, and downloadable receipts are published at [/citation-integrity](/citation-integrity). ## Related - [/tools/citation-verifier](/tools/citation-verifier) - the general verifier: single-citation mode with full per-field diff and candidate search. - [/citation-integrity](/citation-integrity) - explainer on the fabrication pattern, with measured validation numbers and downloadable JSON receipts. - [/tools/retraction-checker](/tools/retraction-checker) - after verification, check whether the resolved paper has been retracted. - [/mcp](/mcp) - programmatic access via the MCP server. ## Sitemap See the full [sitemap](https://scholar-sidekick.com/sitemap.md) for all pages.