The Machine Agreed With Me. That’s When I Got Suspicious.
I asked Gemini whether AI book summaries cost students the learning. It said yes — fluently, confidently, and on a foundation that doesn’t hold.
I asked Google’s Gemini a question I thought I already knew the answer to: if a student uses AI to summarize an assigned book, are they still learning what the reading was meant to teach?
It told me no. Summaries hand over the plot but bypass critical thinking, thematic analysis, and the slow comprehension that only comes from wrestling with the text. I agreed with every word.
Which is exactly the problem.
An answer that flatters what you already believe is the one you’re least likely to check. So I ran it through CERIC (Claim, Evidence, Reasoning, Implications, Context), the same five questions I’d want a student to put to anything a machine tells them. Here’s where a “correct” answer turns out to be hollow.
Claim
Watch how much it hedges. Summaries typically deprive students. They give the basic plot. And then, a few lines later, the same answer concedes that summaries can be “a useful supplement,” even helpful “for students with learning disabilities.” So which is it: a theft of learning or a study aid? The answer wants credit for the strong position and the safety of the soft one. A claim you can’t restate in a single sentence as declarative knowledge isn’t a claim; it’s a mood.
Evidence
This is where it falls apart. The answer arrives armed with six citations, which looks like rigor until you actually read them. One is a blog from a fax software company. One is a business that sells book summaries; asked whether book summaries are bad for you, it is not a neutral witness. One is the marketing page for Evernote’s own AI summarizer, which says AI captures fiction well—a source cited in support of a claim it directly contradicts. Not one is a study. The machine dressed a stack of content marketing in the costume of a literature review, and the footnotes are doing fake work.
Reasoning
Notice the structure: a tidy “What Students Miss” set against “What They Gain (At a Cost).” It’s a rigged scale. The losses are stated as established facts: reading “strengthens memory” and “builds analytical skills,” with no mechanism and no measurement. The gains are scare-quoted into suspicion, and the conclusion was loaded into the format before any evidence showed up.
Implications
Then comes the tell. After warning that summaries gut learning, the answer cheerfully offers to help me use AI “as a supplement” to generate discussion questions and clarify confusing chapters. The critique dissolves into an upsell for the very behavior it just supposedly diagnosed. If you took the answer seriously, you’d close the tool. The attention economy would prefer you keep it open.
Context
And it answers all of this about no book in particular, no student in particular, and no course in particular. A plot summary of an airport thriller and a summary of Beloved are not the same reading situation. What “the reading was meant to teach” depends entirely on what was assigned and why, and the machine, knowing none of that, answered as if it were one thing.
Here’s the part that should unsettle us. The students I’ve been calling the honor-roll offloaders are doing exactly what Gemini did here: producing a fluent, well-formatted, plausibly cited artifact that looks like understanding and was assembled without any of the work understanding requires. The summary and the summarizer share a flaw. The answer made some points I agreed with, and yet, it was still hollow. If you can’t tell those two things apart, the machine has already won an argument it never actually made.
That’s what CERIC is for. Not to reach the right conclusion, but to know whether the argument holds.