Gains:
- Knowing why the hallucination is hidden in the human domain and being able to pass each source through the five-step verification chain (existence, imprint, quote, context, attribution)
- Ability to confirm source authenticity with concrete identities such as DOI, ISBN and catalogue.
- Understand the discipline of banning artificial intelligence from making fabrications and never delegating verification of quotation and attribution honesty to it.
The value of a text in human and social fields depends on the authenticity and accurate transfer of the sources it is based on. A literary essay, a thesis, a lecture note—all rely on reliable sources and honest attribution (showing who/where you got information or a quote). AI offers great convenience in this field, but it also carries the greatest danger: hallucination, that is, producing a non-existent source, quote or fact as real. This unit is dedicated to establishing a "verification armor" against this danger.
One-sentence rule: No AI-generated source, quote, date or number can enter the text without first-person confirmation from a primary or reliable source. This rule is the most repeated and least flexible rule of this module.
Why is hallucination so dangerous in this field?
If AI is wrong on a technical question, it will often be noticed; but in the human sphere the hallucination is skillfully concealed:
- Realistic byline: AI combines the name of a real author with a title appropriate to the field and a reasonable year. The result is a source that doesn't exist but looks exactly real.
- Quotation appropriate to the style: AI can imitate the style of an author and present a sentence that does not belong to him as "his word".
- Close but wrong date: The year of a work may be shifted by one or two years; This escapes the careless reader.
- Confused attribution: May attribute an idea to the wrong person.
All of these can silently infiltrate the text of an author with honest intentions, and if not caught, damage both the author's reputation and the reader's trust.
Caution: Just because a resource appears "realistic" to the AI does not mean it is real. Realism and reality are two different things; AI is a master at the former, it cannot guarantee the latter.
Verification armor: step by step
For each source and quote, follow this chain:
- Is there? Search for the source in the library catalog, a reliable database, or the publisher's site. If it can't be found, it doesn't exist — don't use it.
- Is the imprint correct? Author, title, publisher/journal, year, page — all must match exactly.
- Is the quote real? Find the quote in the source itself, on the specified page, and compare it verbatim.
- Is the context correct? Is the quote used literally, or has it been distorted?
- Is the attribution honest? Is the idea attributed to the right person, is something taken from a secondary source portrayed as primary?
If even one link of this chain is broken, it cannot enter the source text.
Tip: Instruct the AI from the start: "Don't give out any source you're not sure exists; if you're not sure, say 'I couldn't find the source'." This doesn't end the hallucination completely, but it prompts the AI to say "I don't know" instead of making up.
Citation and citation discipline
Along with verification, honest attribution is also a discipline:
- Give the direct quote with quotes and page; Let it be literally.
- Also link indirect quotation (summarizing in your own words) to the source; If the idea is not yours, it needs attribution.
- Transparently state the AI contribution according to your organization's rule; AI is not a “resource,” but if used, it must be declared.
- Do not portray the secondary source as primary; If you took a quote from an author from another book, indicate this.
three mini cases
Case 1 — Two of the five sources turned out to be fictitious. A researcher went through the verification process of 5 sources given by AI. 2 were not in the catalog at all, 1 had the wrong year, 2 were real and correct. Only the 2 verified were used; article protected from fictitious attribution.
Case 2 — Quote distortion caught. One student found a quote from AI in an actual book, but the quote meant the exact opposite in the original context. The AI had taken the sentence out of context. The student corrected the quote with its correct context.
Case 3 — Incorrect attribution corrected. In one lecture note, AI attributed an opinion to the wrong author. When the teacher cross-checked it with two reliable sources, he found the rightful owner and fixed it. The note survived without going wrong to the student.
Four copyable templates
1) Fabrication ban instruction:
From now on: DO NOT produce any source, quote, date or number that you are not sure exists. If you're not sure, say "needs verification" or "I couldn't find it." Never give a tag that is realistic but you cannot guarantee its authenticity. Follow this rule with every response.
2) Source verification roadmap:
I will verify the source below myself. Tell me where to look for this resource (catalog, database, publisher) and with what key information. YOUconfirm that the resource exists; just give the verification steps.Source: [paste credit]
3) Quote context check:
Below is a quote and the actual paragraph in which it occurs. Is the quote used in its original context or has it been distorted? Just rely on the text I gave you; adding a comment.Quote: [...] Original paragraph: [...]
4) Attribution integrity audit:
In my text below: which sentences require a source but remain unattributed, which idea may belong to someone else, is there a place where I show the secondary source as primary? Suggesting fabricated sources; just point out missing citations.Text: [paste text]
Weak prompt / Strong prompt
Weak: "Give 5 academic sources and quotes on this topic."
Güçlü: "Tell me what kind of sources and keywords I should look for while researching on this subject. Do not produce any references or quotes that you are not sure exist; if you are not sure, say 'I couldn't find it'. I will give you the sources I find later; I will confirm the quotes from the books."
The powerful prompt closes the door to hallucination and leaves verification to the human. The weak prompt directly invites imaginary resource production.
Validation chain table
step
Question
verification location
entity
Does the source really exist?
Catalog / database
imprint
Are the author, year, page correct?
Publisher/imprint
quote
Is it verbatim in the text?
the source itself
Context
Is the meaning distorted?
main paragraph
Citation
To the right person?
cross check
Verification with permanent link and imprint
There are concrete ways to quickly and reliably confirm whether a resource actually exists. Most modern academic publications have a DOI (Digital Object Identifier—an unchanging unique ID permanently assigned to a publication; it acts like a link and leads directly to the publication). If you access the actual publication when you type the DOI of an article into a browser, that source exists; If you can't find it or the DOI doesn't exist at all, be suspicious. AI-generated “DOIs” are often made-up and go nowhere; This is one of the most practical tests of capturing the imaginary source.
ISBN (International Standard Book Number) provides a similar identification for books; You can verify the authenticity of the book by searching for it by ISBN in a library catalog or a reliable book database. For periodicals, check the journal name and volume/issue information on the publisher's official website. These concrete identities are the answer to the question "Is the source real?" It answers your question clearly within minutes.
These methods turn verification into a “transaction” rather than a “stock”: when you open and view each source in person, in the catalog, by DOI, or on the publisher site, the risk of phantom attribution is largely eliminated.
Tip: When you finish your bibliography, do a “click tour”: actually open the DOI or catalog record of each source. Even a single resource that does not open is a warning; either correct it with the correct imprint or remove it from the list.
Common mistakes
- Mistaking realism for reality. An AI's plausible identity does not prove its existence.
- Not verifying the quote from the source. Fabricated or distorted quotes may leak.
- Getting history from a single source. Cross-check with at least two reliable sources.
- Making the secondary look like the primary. State the transmission chain honestly.
- Citing AI as a source. AI is not a resource; its use is declared, the reference is made to the primary source.
In summary
Honesty in the humanities relies on source and citation accuracy. The biggest danger of AI is that it produces sources and quotes that look realistic but do not exist. Pass each source and citation through the five-step verification chain (entity, credit, quote, context, attribution). Ban the AI from fabricating from the start and never delegate verification to it. This discipline is the guarantee of your reliability in this profession.
Application task
Try having the AI suggest a few sources and a quote on a topic. Run each source through the verification chain: how many actually exist, how many have correct attribution, is the quote actually in the source? Write the results in a table. This experience will concretely show the limits of AI in this field.
checklist
- [ ] I confirmed the existence of each resource in the catalogue.
- [ ] I have verified each quote verbatim from the original source.
- [ ] I have checked that the context of the quote has not been distorted.
- [ ] I have cross-checked the dates with at least two sources.
- [ ] I banned the AI from fabricating and did not delegate verification to it.