Gains:
- Ability to use artificial intelligence for formatting and in-text citation support in citation styles such as APA, IEEE, Vancouver and MLA
- Ability to detect citations made up by artificial intelligence (hallucinations) and verify each source through DOI, imprint and database
- Ability to establish a reliable bibliography workflow by using artificial intelligence with reference management tools such as Zotero and Mendeley
Citation is the most concrete expression of academic integrity: by honestly citing the source of every idea, data, and claim, you both give credit to previous researchers and enable the reader to verify your claims. Accurate and consistent citation is the face of an article's credibility. AI offers specific assistance in citation tasks: style formatting, in-text citation layout, bibliography formatting. But the core warning of this unit is the most critical sentence of the entire module: AI makes up attributions that appear real but do not exist (hallucinatory attribution); Each source cannot be used without independent verification through DOI and database.
Attribution styles and the safe role of AI
Different fields use different citation styles. APA (psychology, education, social sciences — author-year), MLA (humanities), Vancouver (medicine — numbered), IEEE (engineering — bracketed number), Chicago (history, mixed). Each style has its own in-text citation and bibliography format: punctuation, ordering, italics, abbreviation rules. AI knows these formal rules well, and if you already have the correct imprint information, it can be safely used to format it into the desired style. So the safe role of the AI is to translate or format your authentic and verified byline from one style to another — not to find or remember the source.
The danger starts with telling the AI to "suggest a source on this topic" or "add a reference to that claim." Instead of remembering a real source at that moment, the AI makes up an author name, year, journal, and even a DOI that seems formally valid, saying "there might be such an article" statistically. These tags are so realistic that they cannot be distinguished by eye; The only way is to verify. This risk is not theoretical: published studies have shown that a significant portion of the references produced by language models are either entirely fabricated or misassembled of real parts. The name of a real author can be combined with the title of a real journal and the title of a non-existent article; The parts are familiar, the whole is fake.
Caution: Even a DOI that appears formally valid may be fabricated. Parse DOI via https://doi.org/: valid if it goes to an actual article and the byline matches; If it says "DOI not found" it is fake.
Verification and reference management tools
The validation chain for each citation is: (1) search for the citation in a database (Google Scholar, Scopus, PubMed, CrossRef); (2) parse DOI; (3) check that the author, year, title, and journal match; (4) most importantly, verify from the text that the source actually supports the claim you attribute. The fourth step is often skipped: the source may exist but not say what you say; AI confuses both.
Reference management tools make this job safe. Zotero and Mendeley (free), EndNote (institutional) save the article directly from the database to your library with its real citation, and produce in-text citations and bibliography in the style you want. The safest workflow is this: you import resources into these tools from the database, not from the AI; You use AI only for formal questions (style difference, fixing missing field in an imprint). Thus, the bibliography is created from real records, no fabricated citations can be included.
CrossRef, withdrawal and imprint hygiene
The infrastructure behind DOI validation is registrars such as CrossRef; When you parse a DOI via doi.org, you navigate to these records. CrossRef's search tools and "metadata search" services can be used to find the actual DOI of a piece of text or title; This is a reliable way to find out the truth of the imprint that the AI can make up. Searching a database for the title in quotes before adding a byline will uncover most bogus attributions in seconds.
A frequently overlooked aspect of verification is retraction control: a paper may have been retracted after publication due to serious error or ethical violations. Citing a retracted study as if it were valid weakens your argument. The Retraction Watch database and warning labels on many publishers' article pages indicate this; Check your key sources in this regard as well. Finally, pay attention to imprint “hygiene”: correct order and spelling of author names, full form of the journal name (not abbreviation, if the style does not require it), completeness of volume-issue-page and DOI. AI is helpful in catching formatting inconsistencies in these areas, but accurate information always comes from the actual recording.
three mini cases
Case 1 — Fake DOI. A researcher asked the AI for “appropriate attribution” to a claim and received the byline with its DOI. When I parsed the DOI on doi.org, it came up as "not found"; The author's name was real, but there was no article with that title. The AI had attached an article that did not exist to an existing author. The researcher found the real source on Scholar and added the byline with Zotero.
Case 2 — Source exists but does not support it. A student verified an article suggested by AI: it was real, the DOI was working. But when he opened the article and read it, he saw that his sentence did not contain the finding he claimed; The AI had mismatched the source. The student removed the reference. Lesson: It is not enough for the source to exist, it is also verified if it supports the claim.
Case 3 — Style translation safe use. One researcher had 40 fact sources on Zotero, but the journal wanted Vancouver instead of APA. Zotero automatically translated the style; At a few hesitant points he asked the AI “how do I write this byline in Vancouver?” This was completely safe since the sources were already real; The AI only helped form.
Copiable templates
1) Style formatting (from actual imprint):
Below is the verified imprint information I HAVE. Format them to APA 7 style. ADDING new source, FITTING imprint; Just format the information I give you. Imprints: [author, year, title, journal, volume, page, DOI ...]
2) Validation control table:
Generate an empty validation table for the following citation list:[No, Author, Year, Title, DOI, Found in Scholar?, Does it support the claim?]. You DO NOT verify; I will fill it in manually.List: [paste references]
3) Style translation:
Write the verified citation below in both APA 7 and Vancouver style, briefly explaining the differences. The imprint is real; change information. Imprint: [paste]
4) In-text citation order:
In the following paragraph, arrange in-text citations according to APA 7 guidelines (number of authors, et al. convention, multiple citation order). ADD/REMOVE Source; just fix the formatting of existing citations.Paragraph: [paste]
Weak prompt / Strong prompt
Weak prompt:
Add 3 academic sources to this sentence. [sentence]
AI creates three fake imprints; They all look realistic but may not exist.
Powerful prompt:
Below are the identifiers of 3 sources I HAVE ALREADY FOUND AND VERIFIED. Place these as APA 7 in-text citations in sentences and write bibliography entries. Suggesting new sources, fabricating imprints. Sentence: [...]Imprints: [real imprints]
business
Is it safe?
rule
Formatting the real imprint
Yes
Change information
Translation between styles
Yes
Imprint already verified
In-text citation layout
Yes
No adding/removing resources
"Suggest/add sources"
no
Risk of fabricated attribution
Asking to generate DOI
no
Risk of fake DOI
Common mistakes
- Asking for resources from the AI. Fabricated attribution is the most devastating error in academia; leads to withdrawal.
- Not parsing the DOI. A seemingly valid DOI may be fraudulent; Test it at doi.org.
- Assuming that the existing source supports the claim. Even if the source is real, it may mismatch.
- Not using a reference tool. Manual imprint entry carries the risk of both error and fabrication.
- Not checking the style rules one by one. Inconsistent bibliography undermines trust.
Tip: Golden rule: "Source from database and Zotero; format from AI." Never retrieve the resources themselves from the AI's memory; Use AI only to edit real tags.
In summary
Citation is the concrete face of academic integrity. AI can be used safely to format your real and verified imprints into the style you want. But if it is used to find sources, it produces false citations; This is the most critical risk of the module. Verify each source against DOI and database, verify that it supports the claim, and manage your sources from the actual record with tools like Zotero/Mendeley. The shortest rule of thumb: the source is from the actual record, the format is from the AI, the validation is from you.
Application task
Add your 5 actual sources to a reference tool (Zotero/Mendeley) and produce a bibliography in APA and IEEE styles. Then do an experiment: Tell the AI to "suggest 3 sources" on your topic and check the suggested citations one by one with DOI analysis. Note how many turn out to be real and how many are fake; This gives you a concrete view of the risk of spurious attribution.
checklist
- [ ] Have I verified each source via DOI and database?
- [ ] Have I confirmed that the source actually supports the claim I attributed?
- [ ] Did I get the sources from the real record and not from the AI?
- [ ] Did I use a reference management tool?
- [ ] Have I applied the citation style consistently?