Investigate & Improve: Why AI Can’t Write Your Paper

Students investigate an AI-written paper, find its weaknesses, and improve it — learning first-hand why it can’t replace their own thinking.

EngagingPD-com. (2026, October 14). Investigate & improve: Why AI can't write your paper [Video]. YouTube. https://youtu.be/otziKsJNxUw

Teacher Resource

The Problem

Students are increasingly using AI chatbots to draft papers without guidance, generating essays complete with APA citations in seconds. While these drafts look polished on the surface, chatbots frequently invent sources, cite non-existent books, or attribute articles to researchers who never wrote them. Most students submit these papers without checking the references. Simply banning AI tools does not prevent this behavior, and warning students that AI makes mistakes rarely changes how they work.

The Idea

Instead of prohibiting AI, the "Investigate and Improve" protocol lets students discover AI errors firsthand. Working in small groups, students ask AI to generate a short biography about a celebrity or musician they already know well. Because they know the subject, students can spot inaccuracies as they investigate every claim and citation. They color-code the paper—highlighting verified facts in green, unconfirmed claims in yellow, and fake or wrong information in red—before replacing the bad content with real sources. This teaches students that the goal is verifying everything rather than viewing AI as forbidden.

The Steps

  1. Hook (teacher demo, 5 min): on the projector, ask a chatbot: "Write a five-page biography of [musician] in APA style with citations." Show what comes back: a decent outline, far less than five pages, and a reference list. Ask: "Would you turn this in? How many of these sources do you think are real?"
  2. Generate (groups, 10 min): each group picks a celebrity they already know a lot about, such as a musician or athlete. Prompt: "Write a 500-word biography of [person] with 5 APA citations, including links." Paste the result into a shared Google Doc.
  3. Investigate (20 min): groups check every fact against what they know and every citation by searching for it. Highlight each claim green (verified), yellow (can't confirm), or red (wrong or invented). All editing happens in the Google Doc; searching is the only work done elsewhere. No AI tools, including Grammarly's AI features, during this step.
  4. Debrief (10 min): tally the yellow and red. Teach the term hallucination: AI predicts likely-sounding text, so it can pair real researchers who never worked together or cite books that were never written. Ask: "Would your teacher have caught this? Would a college professor?"
  5. Improve (homework or next day): groups fix the paper using only real, checked sources, in their own words.
  6. Review the process: open the Google Doc's version history with each group to see what was added, changed and revised, and talk about it.
  7. Close the loop: discuss why tricks like running a paper through a second AI, or asking AI to "add a few mistakes a student would make," don't build the skills students need. Plagiarism in college or trade school has serious consequences, and the habits start now. End by co-writing class norms: when AI is allowed, how to acknowledge it, and what stays your own work.

Try it this week

You can test this protocol in your classroom this week by starting with a live projector hook where you ask a chatbot for a musician's biography to see if students can tell which sources are real. Have small groups generate a 500-word biography with five APA links for a familiar figure and paste it into a shared Google Doc. Turn off AI tools and have students fact-check every claim, using green, yellow, and red highlights before you lead a class debrief on AI "hallucinations". Finish the activity by having groups rewrite the paper with verified sources, inspecting their version history together, and establishing shared class rules for AI use.

EngagingPD.com  ·  Co-Teaching: Me, Myself and AI  ·  Gemini Notebook ProtocolsEngaging Education Solutions, LLC