AI Reading Investigation
Rembrandt, A Scholar Seated at a Desk, 1634. National Gallery Prague.
Rembrandt, A Scholar Seated at a Desk, 1634. National Gallery Prague.
In my history of diet and health course, students read Steven Shapin’s “Trusting George Cheyne.” I wanted them to consider why people trusted Cheyne and what that trust tells us about expertise. I gave them a sequence of AI prompts to help investigate the article, followed by a short post explaining what they checked in the text.
This is part of a remote, asynchronous course on the history of diet and health.
Students need the full article and relevant course discussions about expertise. Test that the tool can read the supplied text and demonstrate checking a suggested passage. The proposed submission below consists of initial notes and a 400-word post; the linked handout preserves the assignment as given.
Before using AI, students write two or three notes about what they think the article is doing and what confuses them. These provide something to compare with their later interpretation.
Each stage of the sequence has a different purpose. Students can adapt it to the reading and follow up where necessary:
This sequence gives AI work we often ask students to practice: identifying an argument, selecting passages, and proposing interpretations. It gives students suggestions to test against the article, but limits their practice making those first judgments. If that independent reading is the aim, an adaptation would be to begin with students’ interpretations and use only step 7.
For a shorter exercise in assessing AI’s reading, use orientation, evidence-finding, and interpretation-testing: steps 1, 4, and 5.
first prompt in the sequence
I am reading [author, title] for a course on [subject]. Give me a concise orientation to the article: what question is the author asking, what is the main argument, how is evidence used, and what historical problem is the author trying to solve? Do not provide a flat, high-level summary of topics. Focus on the argument and evidence.
last prompt in the sequence
I need to write my own reading reflection. Here is my current possible angle: [paste your own rough idea]. Based on our discussion, give me a few ways I might sharpen or complicate it. Do not write the reflection for me. For each suggestion, tell me what evidence from the article I would need to verify before using it.
For another run, I’d stop requiring students to report a useful suggestion and a passage they understood better. Those requirements make an unsuccessful exchange difficult to describe honestly.
Ask for about 400 words addressing:
Ask for the initial notes too, so students can explain what changed or stayed the same. A well-supported account of an unhelpful exchange should meet the assignment’s requirements. The post should spend its space on passages and interpretations rather than retelling the chat.
Constructed example — no quotations from Shapin or student work. AI proposes that readers trusted an expert solely because his recommendations were effective. A student finds a passage discussing the expert’s character and relationships. They quote that passage, explain how it complicates “solely,” and revise the account of credibility.
If the passage instead supports the proposed interpretation, the same process can justify accepting it. The post needs the evidence and reasoning, not a required story about catching AI out.
Look for specific passages, an explanation of what was checked, and a defensible answer to the historical question. Completing the post doesn’t establish that a student read carefully. Their explanation needs to show why a passage supports or complicates a claim.
Give the same attention to agreement and disagreement with the model. Accepting a claim needs evidence; rejecting it needs reasons. Students shouldn’t have to manufacture an error or a benefit to complete the task.
Try the sequence yourself first. Answers can differ considerably depending on whether the tool has the article or is answering from its title. Make access to the text explicit in the instructions, and check the passages the model recommends. Knowing the reading’s particular difficulties will help you respond when students encounter them. State which tool or account students can use and how they should supply the article. Another adaptation would be to provide one generated response for everyone to examine, avoiding the need for individual accounts.
This sketch is licensed under Creative Commons BY-NC-SA 4.0. You're welcome to share and adapt it for noncommercial purposes, with credit, as long as you share your adaptation under the same license.
Image Rembrandt, A Scholar Seated at a Desk, 1634. National Gallery Prague. The image's own license applies to it, separately from the sketch.
Suggested citation
Fred Gibbs. "AI Reading Investigation." AI Sketchbook, Amaranth, University of New Mexico, September 25, 2026. https://amaranth.unm.edu/ai-sketchbook/teaching/ai-reading-investigation.html. Licensed under CC BY-NC-SA 4.0.