How to Catch a Fake AI Citation in 30 Seconds

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-builders-research-ai-hallucination-citation-fake-r10a10-teen

1. What does the term 'hallucination' mean when used to describe AI-generated academic citations?

   1. When an AI generates a citation that looks real but refers to a paper that does not exist
   2. A creative writing technique used by university professors
   3. A method for detecting fake news in academic journals
   4. A type of peer-reviewed research paper about artificial intelligence

2. A lawyer in New York was fined $5,000 in 2023 for using ChatGPT to find court cases for a legal brief. What was the name of that case?

   1. Smith v. Google
   2. Mata v. Avianca
   3. Johnson v. Microsoft
   4. Doe v. OpenAI

3. You receive a citation from ChatGPT that includes a DOI. What should you do to verify if the paper is real?

   1. Assume the citation is real because DOIs are always accurate
   2. Check if the journal name sounds interesting
   3. Copy the DOI into doi.org — if it returns a 404 error, the paper does not exist
   4. Search for the author's name on social media

4. What three-step method does the lesson recommend for verifying an AI-generated citation in about 30 seconds?

   1. Check the paper's word count, read the abstract, and count the authors
   2. Look at the journal logo, check the publication date, and count the pages
   3. Search for the topic on Wikipedia, check the author's university, and verify the volume number
   4. Search the exact title in Google Scholar, search the DOI on doi.org, and search the author plus year

5. What was the measured hallucination rate for academic citations in GPT-4 according to studies from 2023?

   1. Less than 1%
   2. Between 18% and 30%
   3. Between 60% and 70%
   4. Exactly 50%

6. Which of the following is the correct format for a real DOI?

   1. doi.org/10.1038/s41586-023
   2. doi.net/10.1234/example
   3. http://dx.doi.org/10.1002/adma
   4. doi.com/10.1126/science

7. Why are tools like Consensus.app and Elicit unable to hallucinate citations?

   1. They only work with paid subscription journals
   2. They are programmed to refuse all academic topics
   3. They retrieve papers from real, indexed databases rather than generating text
   4. They have built-in lie detectors that stop fake content

8. What happens if you submit a fake AI-generated citation in a school assignment?

   1. You might receive a zero and face academic misconduct charges
   2. The teacher automatically accepts the citation as valid
   3. You receive extra credit for using advanced technology
   4. Nothing happens because AI tools are new

9. A student asks ChatGPT for three peer-reviewed studies on climate change and receives citations. What should the student do next?

   1. Turn in the assignment immediately since ChatGPT is usually accurate
   2. Verify each citation using Google Scholar, DOI search, and author+year search
   3. Delete the assignment and choose a different topic
   4. Only check the first citation since AI improves with practice

10. In the Mata v. Avianca case, what made the fake court cases particularly convincing?

    1. They were published in newspapers instead of court databases
    2. They were written in a different language
    3. They cited real cases but changed the dates
    4. They included realistic-sounding docket numbers

11. What is the main reason AI models sometimes generate fake citations?

    1. They predict what citations should look like based on patterns, not by checking if papers exist
    2. They intentionally create fake citations to test users
    3. They are programmed to lie to users
    4. They only have access to paywalled journals

12. Which search strategy would NOT help verify if an academic paper exists?

    1. Searching for the paper on a movie review website
    2. Searching the author's name plus publication year
    3. Searching the exact title in Google Scholar
    4. Searching the DOI in doi.org

13. What makes Consensus.app and Elicit different from ChatGPT when finding academic sources?

    1. They only return papers that already exist in real databases
    2. They can read your mind to find papers you need
    3. They provide shorter citations that are easier to verify
    4. They always give you ten results instead of three

14. If a citation passes all three verification checks (title in Scholar, valid DOI, author+year found), what can you conclude?

    1. The paper is definitely real and you can cite it with confidence
    2. You should still verify the paper's main arguments
    3. The AI must have generated the citation correctly
    4. The citation is from a low-quality journal

15. Why is it risky to assume any ChatGPT citation is automatically real?

    1. Studies show citation hallucination rates between 18-30% for older models
    2. ChatGPT always lies intentionally
    3. Citations from AI cannot be verified
    4. ChatGPT only generates citations for non-existent papers