A librarian receives a request for an ordinary-looking journal article: a real scholar, a plausible title, a volume number and page range. The catalogue returns nothing. Neither does the journal archive. After checking alternate spellings, subject databases and archival finding aids, staff discover that the citation came from an AI-generated summary—and may never have referred to a real publication.
This is no longer just a chatbot error inside a private conversation. Fabricated books, articles, archival records and URLs are entering student papers, reading lists, news work and reference desks. A polished citation is a description of a source, not proof that the source exists.
As an Amazon Associate I earn from qualifying purchases.
What librarians are being asked to find
Generative-AI systems can produce complete-looking bibliographic records without verifying them against a catalogue, publisher or archive. Requests brought to librarians may involve:
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstall- Academic books with plausible titles, publishers and dates.
- Journal articles with realistic volume, issue and page details.
- Real scholars assigned invented articles or books.
- Entire journals or journal issues that were never published.
- Government reports and institutional publications that do not exist.
- Archival collections, box numbers, folder references or catalogue identifiers that were never created.
- URLs that fail, redirect to unrelated material or point to a page with no such document.
- Real publications whose authors, dates, pages or titles have been blended with invented details.
Those cases are not interchangeable. A real work may be hard to access, incorrectly cited by a human, held in an unindexed collection or listed under a translated title. The first result of a failed search is therefore unverified, not automatically nonexistent.
#1 Best Overall
| What the citation might represent | What to check |
|---|---|
| Real source, difficult to find | Alternate titles, transliteration, print holdings, specialist catalogues and access restrictions |
| Real source, incorrect citation | Author, title, date, edition, journal, volume, issue and page range separately |
| Restricted or poorly indexed record | Institutional finding aids, local catalogues and an archivist or librarian |
| Completely fabricated source | Independent publisher, catalogue, DOI, ISBN or archival record |
| Real details combined into a false record | Whether every field belongs to the same work, edition or issue |
How common is the problem?
Sarah Falls, chief of researcher engagement at the Library of Virginia, estimated that about 15% of emailed reference questions received by her library were generated by AI. Falls said some included hallucinated published works and primary-source documents. That is an estimate from one library, not a national rate for reference questions or academic citations. Futurism reported Falls’s account on December 10, 2025.
The systems named in that reporting—ChatGPT, Google Gemini and Microsoft Copilot—are examples rather than an exhaustive list. The issue affects anyone who copies an AI-produced bibliography without checking it: students, journalists, teachers, policymakers and professional researchers.
Why an AI can make a citation look authentic
A language model generates likely sequences of words. Unless a system is explicitly retrieving and checking a record, it is not consulting a definitive catalogue for each sentence. Bibliographic language is especially predictable: author, year, title, journal, volume, issue and pages form a familiar pattern that a model can complete convincingly.
Rank #2
- Foundations of Library and Information Science
- Used Book in Good Condition
- Product type: ABIS BOOK
The model may recombine a real author with a real journal and a subject-appropriate title, then add plausible page numbers. It may continue writing when the correct answer would be “I cannot verify this” or “no record was found.” That behavior is fabrication in the technical sense: unsupported content generated from patterns, not a deliberate human lie.
The International Committee of the Red Cross has warned, as reported by Futurism, that incomplete or silent historical records create an opening for invented archival references, including catalogue numbers, document descriptions and references to nonexistent platforms. The organization directs researchers toward its own catalogue and archival resources rather than treating an AI-generated list as authoritative.
Why proving that a source does not exist takes so much work
Finding a record is often easier than proving that no record exists. Library and archival description is distributed across institutions, and many collections are not fully digitized. A serious check can involve union catalogues, national bibliographies, subject indexes, serials directories, publisher archives, institutional repositories and specialist finding aids.
Negative evidence is inherently limited
- A title may be translated, abbreviated or recorded with a different spelling.
- Names may use another transliteration or a variant form.
- An item may exist only in print or in a local catalogue invisible to ordinary web search.
- An archive may have an incomplete finding aid, unprocessed material or restricted files.
- A work may have been announced, withdrawn, left unpublished or issued under an institutional file number.
- A private collection or small repository may not expose records to public search engines.
Falls told Futurism that unique records are particularly difficult to disprove. Librarians are not simply checking Google; they are testing several independent explanations for why a citation cannot be found.
How fabricated references spread
- An AI system invents a citation or merges details from several real works.
- A user copies it into a paper, article, report, syllabus or reading list.
- Another writer encounters that mention and assumes the source was already verified.
- Search engines, scraped pages or AI summaries surface the repeated claim.
- The citation acquires apparent credibility through repetition.
- A librarian must determine whether the item is merely obscure or has been copied from another invented reference.
Secondary coverage has described this as a form of citation laundering: an unsupported reference gains authority because it appears in a real document and is then cited again. The Outpost reported an example involving a Chicago Sun-Times freelance reading list in which 10 of 15 recommended books were said not to exist. That account should be treated as a reported case, not evidence that every such list is fabricated.
The archival problem is different from a missing web page
Historical records are especially vulnerable to plausible invention because catalogues are descriptions of surviving material, not complete maps of everything ever created. A model can generate a convincing collection name or box number by extending the language used in real finding aids. When a historical record is incomplete or silent, there may be no quick public search that settles the question.
Rank #4
For humanitarian, government and institutional archives, use the holding organization’s catalogue and published finding aids first. If the record remains unclear, ask the archive whether the collection, series or file identifier is genuine. Do not treat an AI-generated archival citation as evidence merely because its terminology sounds professional.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.A verification workflow that works
Before relying on a citation, ask the AI system for identifying details. These requests do not establish that the source is real, but missing or inconsistent fields are warning signs.
- Publisher and place of publication.
- ISBN for a book; DOI or ISSN for a journal item.
- Journal volume, issue and complete page range.
- For archives, repository, collection, series, box and folder information.
- A direct link to the publisher, repository, catalogue or archive.
- The exact page or passage supporting the claim.
Check books and reports
- Search the exact title in quotation marks.
- Search the author and title separately, including spelling variants.
- Check major library catalogues and the publisher’s own catalogue.
- Verify ISBN metadata, edition, publication year and author list in a national library record or trusted bibliographic database.
- Compare the record with any institutional repository or government publication list.
Check journal articles
- Search the exact title and then the author-title combination.
- Open the journal’s own archive and verify the volume, issue, pages and publication date.
- Resolve the DOI through the DOI registry or the journal site; do not rely on a DOI-looking string alone.
- Check that the author affiliation and article title belong to the same issue, rather than being mixed from different records.
- Read the article itself before quoting or paraphrasing it.
Check archival and web sources
- Search the holding institution’s official catalogue and finding aids.
- Use the institution’s terminology, collection numbers and alternate names.
- Contact an archivist when a collection, series or file is not publicly described.
- For web material, confirm that the URL belongs to the claimed organization and that the page contains the cited document.
- Treat results found only on scraped pages, citation farms or AI-generated sites as unverified.
What to do when the search fails
Do not ask another chatbot to “fix” an unverified citation and then accept the replacement. Record the original output, tell the librarian or instructor that AI generated it, and replace it with a source located through a real catalogue, publisher, archive or database. Quote or paraphrase only passages you can inspect in the original.
A failed search can still reflect a misspelled name, an alternate title, an inaccessible database, an uncatalogued collection, a withdrawn work or a citation error introduced by a human. A librarian can help distinguish those possibilities, but the burden of proof remains with the person making the claim.
What this changes for libraries and researchers
Generative AI does not create unreliable citation practices from nothing; copying mistakes, paper mills, predatory publishing and careless reference management predate chatbots. It can, however, lower the cost and increase the speed and plausibility of producing false references. That shifts verification work to librarians, editors, teachers and researchers.
Libraries must balance thorough reference help with the time required to investigate records that may never have existed. The most useful response is neither automatic refusal nor uncritical acceptance: disclose the AI source, let staff test the citation, and treat a catalogue record, publisher page or archival confirmation—not the citation’s professional appearance—as the evidence.
Recommended Free Tools
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
