Gains:
- Ability to capture the hallucination by verifying the existence of each source, going to the primary source, and independent cross-confirmation steps
- Ability to maintain originality and academic integrity against fabricated quotations and documents in the humanities and social sciences
- Ability to teach users source verification and information literacy skills in the AI age
At the heart of librarianship's social mission is a single promise: reliable information. A librarian brings the user not only to information, but to verified information, and teaches him how to evaluate this information himself. In the age of artificial intelligence, this mission has become both more difficult and more important; because AI can produce misinformation in a highly convincing way. In this unit, we will cover hallucination recognition, the discipline of source verification, and teaching information literacy in the age of AI.
Let's define the concepts. Hallucination is when AI produces information, sources, quotes or events that do not actually exist as if they were real. Source verification is the confirmation that information is based on a real, reliable and primary source. Information literacy is a person's ability to find information, evaluate it, question its reliability, and use it ethically. This unit aims for the librarian to strengthen these skills both in his own practice and in teaching the user.
Step by step: source verification discipline
1. Request the source and verify its existence. When the AI gives a claim or quote, “what source is this from?” and check that the source actually exists in an independent catalogue, database or primary document.
2. Go to the primary source. The most reliable form of information is its first-hand source. Don't settle for a summary of AI or a secondary text; If possible, access the original document, data or publication.
3. Cross-confirm. Verify a critical claim with at least two reliable sources, independent of each other. One resource, especially AI resource, is not enough.
4. Evaluate the credibility of the source. Who wrote it, when, for what purpose, on what evidence? Is it up to date? The existence of information does not make it true; Also question the nature of the source.
5. Express uncertainty honestly. If information cannot be verified, say so clearly. It's always better to say "unconfirmed" than to offer a fabricated certainty.
Tip: When you get a quote or source from AI, first search the title and author in an actual catalogue. Non-existent resources appear immediately here. This 30-second check prevents a fabricated source from entering a bibliography, an assignment, or a publication.
Authenticity and verification in the humanities and social sciences
The humanities and social sciences are particularly vulnerable to the risk of hallucinations. In these areas, evidence is often based on a specific document, a date, a quote, or the source of a comment. AI can produce highly specific but completely fabricated claims, such as “that historian said” or “that document from 1923 says”. In a work of literature, history or law, this means a false chain of evidence.
Moreover, originality is the fundamental value of these fields: it matters to whom a text, an interpretation or a translation really belongs. Presenting a text produced by AI as an original work or relying on a "source" made up by AI violates academic integrity. The librarian is a guide who teaches users both verification and the limits of originality and plagiarism.
Caution: Even in cases where an AI-generated quote appears word-for-word accurate and is attributed to a real author, that quote may be fabricated. Verify a quote by finding it only in the author's actual work; The logic of "this author might have said something like this" is not enough.
three mini cases
Case 1 — Student was taught verification. A librarian saw 5 resources from AI in a student's assignment. Together they checked it in the catalogue: 2 sources were fabricated. The student experienced firsthand the lesson that "it is not right for it to look real" and once again gained the habit of verifying every source.
Case 2 — Fabricated historical document. A researcher received a "1919 newspaper report" from AI; The date, newspaper name and content were extremely convincing. The librarian scanned the newspaper archive: there was no such news at that time. A fabricated primary source was caught before it even made it into an article.
Case 3 — Misattributed quote. AI attributed a quote to a famous thinker. The librarian looked for this word in the thinker's works and could not find it; The word was not mentioned anywhere. Cross-checking prevented the dissemination of a misattributed quote.
The chain of spread of misinformation and the role of the librarian
Misinformation often starts with a single fabricated output, but it doesn't stop there. A user puts a fake AI-generated resource into an assignment; the assignment becomes a publication; the publication is cited by others, so that a fictitious source acquires the appearance of "truth" by being repeatedly cited. We can call this attribution contamination: false information gains legitimacy by being repeated again and again. The librarian stands at the most critical link in this chain; because it is the last control point where information is verified before entering the system. Just because a source "occurs in many places" does not make it true; The same fake information can appear in many places by being copied from each other. Therefore, cross-confirmation should be based on the question "how many reliable sources confirm it independently of each other", not "how many places does it appear?" Ten secondary sources repeating the same claim are not worth as much as a single independent primary source. The librarian should place this distinction (not confusing independence with repetition) at the center of both his own verification and his teaching to the user.
Attention: The appearance of information in a search engine or on many web pages is not proof of its accuracy. The same mistake may have become widespread by being copied from source to source. Prevalence and accuracy are different things; Confirmation requires an independent and reliable source.
Four copyable templates
1) Resource entity query:
List every source citation and citation in the text below. For each: author, title, year and publication information (if any). Extract item by item the information I need to verify; do not add.Text: [here]
2) Confirmability assessment:
For each of the following claims, isolate the factual elements (date, name, number, quote) that need to be verified. Suggest what type of source (catalog, primary document, database) it can be verified against. Don't judge accuracy, just show the verification path. Assertions: [here]
3) Welding reliability checklist:
Evaluate the credibility of the following source: who wrote it, when, for what purpose, is there evidence, is it current, is there possible bias? Just rely on the information I gave, write "unknown" where it is missing. Source information: [here]
4) Information literacy teaching plan:
Produce a brief 20-minute tutorial on source verification and hallucination in the age of AI for the following user group (e.g. undergraduates): objectives, 3 examples, 1 application. Group: [here]
Weak prompt / Strong prompt
Weak prompt:
Write a summary on this topic with reliable sources.
Saying “with reliable sources” encourages the AI to produce fake sources that appear to be real; There is no validation hook.
Powerful prompt:
Your role: information literacy assistant. WRITING a summary on the following topic; Instead, list the factual questions a user needs to verify and the type of reliable source they can verify them with. Do not fabricate any source citations; Mark the area you are not sure of as "must be verified". Subject: [here]
Powerful prompt; It uses AI as a guide that shows the way to verification, rather than producing answers, and structurally prevents fabrication.
Authentication layers table
layer
Question
Method
entity
Is this source real?
Catalog/database search
primacy
Is this the closest source?
Accessing the original document
Confirmation
Does another source confirm this?
Independent cross-check
reliability
How solid is the source?
Author, date, purpose, evidence
honesty
Am I sure?
State uncertainty clearly
Common mistakes
- Thinking that what is convincing is true. Looking good is not proof of righteousness.
- Being satisfied with a single source. Critical information requires independent cross-confirmation.
- Not looking for the quote at its source. This is how misattributed words spread.
- Hiding uncertainty. “Unverifiable” is better than false certainty.
- Not teaching the user to authenticate. Information literacy is the librarian's mission.
In summary
Source verification and information literacy are the most critical tasks of librarianship in the age of AI. Hallucination produces convincing but fabricated information and is particularly dangerous in the humanities/social sciences. The discipline is clear: verify the existence of each source, go to the primary source, conduct independent cross-checking, question the source's credibility, and express uncertainty honestly. The librarian spreads this culture of verification not only in his or her practice but also by teaching users; Originality and academic integrity are integral to this mission.
Application task
Request 6 sources and 2 citations from AI in your field. Verify each source by searching for it in an actual catalog and each citation in the author's actual work; Note how many are fabrications. Then, using the “Information literacy teaching plan” template, draft a 20-minute tutorial to explain source verification to a user group and include real examples you found yourself.
checklist
- [ ] I have independently verified that each source actually exists.
- [ ] I have used primary sources wherever possible.
- [ ] I have cross-checked critical claims with at least two independent sources.
- [ ] I verified the quotations by searching for them in the author's actual work.
- [ ] I clearly marked the unverifiable information as "unverifiable".