AI as Second Opinion

Old books are hard to read cover to cover and easy to ask AI about. This assignment uses both facts. Students sample a small set of primary sources (in my case, nineteenth- and early twentieth-century diet books) and form their own first impression. Then they ask AI for its overview, treat it as a rough second opinion, and go back to the originals to see where it holds up. They write a comparison of the texts, but the real discovery is the gap between what AI noticed and what they saw for themselves.

The Setup

Offer a landscape, not a reading list. Give students four or five digitized sources from the same period and ask them to pick two. They’re sampling, not reading five books: title page, preface, table of contents, and pages from throughout, read like artifacts. Who is the author talking to? What problem do they think they’re solving? What kind of authority are they trying to project?

First impression before AI. Students spend a few minutes with each text and write down their own quick take before prompting anything. Everything else gets measured against this baseline.

Ask AI for an overview. What does each author seem to argue, who is the audience, what authority does each project? The answer is a second opinion, not an interpretation.

Go back to the originals. Students find specific passages that confirm, complicate, or challenge the AI’s overview, paying attention to tone, voice, audience, and evidence.

The post includes their own first impression, one useful AI impression, one place the original confirmed it, one place the original complicated or corrected it or made it look generic, and a comparison of the two texts built on specific examples.

Run it again, with more independence. A week or two later, with a new set of sources (three this time, from a different period), students write their own prompts instead of relying on a generic “compare these texts.” They might ask AI to compare rhetorical strategies, surface assumptions about bodies or evidence, or identify what each author seems most anxious about. The post now includes the prompts themselves, and the comparison is organized around one or two lenses: rhetorical style, continuity with earlier sources, or how each author establishes expertise.

The Prompt

A basic prompt is fine for the first round. You want the kind of competent overview the originals will complicate:

In the second round, students write their own prompts. The examples to give them are lenses, not wording: compare rhetorical strategies, surface assumptions about evidence, explain unfamiliar terms, or name what each author is most worried about.

Why It Works

The first impression before AI is the key move. Without it, students read the originals primed by AI’s version and mostly find what it told them to find. With it, there are three perspectives on the page (theirs, AI’s, and the text’s), and the analysis starts where they disagree.

Old texts make a natural test case. AI can usually say what a nineteenth-century diet book argues. It’s much shakier on how the book sounds: one author’s breezy confidence, another’s defensive hedging, the appeals to personal experience or laboratory science. Those details tell a historian who a text was for and why it persuaded anyone, and students find them only by going back to the page.

Running it twice turns a structured exercise into a habit. The first round shows students what the back-and-forth looks like; the second asks them to design it.

What to Grade

The grade rests on the comparison and on the three-way conversation among the student’s impression, AI’s, and the text.

• Strong: the first impression is clearly the student’s own; the place where the original complicated AI’s answer is specific and explained; the comparison is organized around a lens and built on quoted or cited details from both texts.
• Middling: all the parts are there, but the comparison drifts toward two summaries, or the “complication” of AI’s view is generic.
• Weak: mostly summary, with little sign the student went back to the originals.

In the second round, add the student’s prompts to the criteria: did they interrogate the sources, or just ask for a comparison?

What to Watch For

• The comparison collapses into two summaries side by side. Asking for one or two organizing lenses helps.
• Students skip the first impression or write it after the fact. Ask for it first in the post.
• Old typography, like the long s that looks like an f, slows students down for the first few pages. Warn them in advance.
• A short walk-through of how to sample a long book (what to look at, in what order) pays off in both rounds. I used a video.

Using This Sketch

Print

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 Cover of Lulu Hunt Peters, Diet and Health, with Key to the Calories, 1918. The image's own license applies to it, separately from the sketch.

Suggested citation

Fred Gibbs. "AI as Second Opinion." AI Sketchbook, Amaranth, University of New Mexico, September 25, 2026. https://amaranth.unm.edu/ai-sketchbook/teaching/ai-as-second-opinion.html. Licensed under CC BY-NC-SA 4.0.

Related Sketches

activity · tested Narrate the Slides What does a smooth narrative leave out of the images it strings together? activity · lightly tested How Else Could This Look? What did this article decide to be about, and what did that decision cost?