Gemini Notebook: Read Opinions Without Adopting Them
Nine prompts for letters to the editor, bold scientific claims, and any text trying to persuade you
The most dangerous text isn’t the one you disagree with. It’s the one you agree with from the first sentence.
The letters page is ideal practice: short, passionate, openly biased.
Take the New York Times letters page of October 27, 2026, headlined “Let Us Choose Wisely.” Readers write about the U.S. congressional elections. One letter uses a string of questions to set two visions of America against each other. Another, responding to an essay by Jeffrey Toobin, argues that even an opposition win in both chambers wouldn’t change much. A third advises letting the president turn the election into a referendum on himself. At the bottom: a letter on stuttering and one asking whether AI can be conscious.
The same day, the Times ran Cade Metz’s piece “Did the Hardest A.I. Problem Just Get Much Easier?” In a 26,000-word essay, former OpenAI researcher Nick Cammarata argues that interpretability, understanding why AI models do what they do, is far easier than most people think. Other researchers reply that it’s more impression than evidence.
The first source openly takes a side. The second looks neutral but carries a contested claim.
Before the first prompt
Save both pages as PDFs or paste the text into two text sources. Add them to a new notebook in Gemini Notebook. When a prompt targets one text, select only that source.
Prompt 1: map the voices
For each letter in the source, list: the author and city, the main claim in one sentence, the tone (angry, resigned, ironic, calm), and the text or event it responds to. Support each item with a quote. Present it as a table.
Get Mihailo Zoin’s stories in your inbox
Join Medium for free to get updates from this writer.
The three political letters don’t share a tone. One calls for action, one is resigned, one offers tactical advice.
Prompt 2: rhetoric or argument
Extract every sentence framed as a question or as a choice between two options. For each, assess: are the options offered the only possible ones, or are there others the author doesn’t mention? Don’t give your own view on the topic; analyze only the structure of the sentence.
“Is it X or Y?” is the oldest tool of persuasion. When Gemini Notebook takes such a sentence apart, you see where the argument ends and the appeal to values begins.
Prompt 3: fact, judgment, prediction
Sort every claim in the letters into three groups: (a) verifiable facts, (b) value judgments, © predictions. For group (a), say what should be checked and where. Don’t check them yourself and don’t add data that isn’t in the source.
The last sentence keeps the model on the page and stops it from “helping” with data of unknown origin. When a letter says something will most likely be over, that isn’t a fact. It’s fear in the future tense.
Prompt 4: the strongest counterargument
For the main claim of each letter, write the strongest counterargument an intelligent reader who disagrees would make. No irony and no straw man. Clearly mark that the counterargument does not come from the source.
This is the prompt we skip most often, because it’s uncomfortable. Here Gemini Notebook must step outside the sources, and you must see exactly where.
Prompt 5: claim and evidence
From the interpretability article, extract Cammarata’s main claims. For each, state what evidence the article offers (data, example, authority, none), who disputes it, and in what words. Present it as a table with quotes.
A news article doesn’t cheer, but its claims don’t carry equal weight. Three years ago Cammarata estimated that under 2 percent of neurons could be understood; now he believes he can explain over 90 percent. Constantin Venhoff replies that the essay is “largely made up of Nick’s vibes”: beautiful examples, no empirical results. The table also shows who praised the essay: venture capitalists and entrepreneurs, Sam Altman among them. Applause isn’t peer review.
Prompt 6: the quiet letter at the bottom
Analyze the letter about AI consciousness. What criteria for consciousness does the author implicitly use? Sort them into biological (body, hunger, pain), social (family, attachment), and functional. Which criterion would be hardest to apply to a machine, and which to a person in a coma?
The shortest letter asks the deepest question. The author lists what a machine doesn’t do: it doesn’t breathe, hunger, bleed, or hug its mother. He treats it as self-evident that consciousness is tied to the body. The coma question is the test: a good criterion must survive the inconvenient case too.
Prompt 7: when two sources meet
Compare the letter about AI consciousness with the interpretability article. The letter asks what a machine is; the article asks what a machine does inside. Do these questions touch? Answer only from the sources, support every connection with quotes from both texts, and say clearly if there is no connection.
The letter writer is certain the machine isn’t conscious. The researchers admit they don’t know what individual neurons do. One draws conclusions about the machine’s interior without looking inside; the others look inside and don’t dare conclude. Elegance without citations is invention.
Prompt 8: don’t trust the explanation, ask for the citation
Find everything the interpretability article says about whether an AI system’s explanation of its own decision can be trusted. Then write a five-question checklist a reader should apply to every AI answer, including this one.
For any user, this is the most important line in Metz’s article: when an agent explains why it made a decision, the explanation may be completely bogus. That applies to Gemini Notebook too. Its advantage is that it ties every claim to a citation you can open. Don’t ask the model why it thinks what it thinks. Click the citation number and read the source.
Prompt 9: the neutral rewrite
Summarize both sources in 150 words for a reader who follows neither U.S. politics nor AI research. No evaluative adjectives and no taking sides. Then list the words from the originals you had to remove to make the text neutral.
The list of removed words is a map of emotional charge. In the letters you’ll find fear and anger. In the science piece, “revolutionary” and “solved.”
What you’re actually practicing
None of these nine prompts tells you who is right. Gemini Notebook isn’t the judge here. It’s the magnifying glass. It shows structure, tone, the line between fact and fear, and the gap between claim and evidence.
A reader’s letter persuades you openly. A science story persuades you quietly, with numbers and authority. The same method works for both.
The better you understand the text you agree with, the less it leads you.
For further exploration
- Gemini Notebook Mastery (formerly NotebookLM) — Protocols and Decision Systems
The full membership: every protocol and decision system, plus new HOW TO guides. - NotebookLM Mastery: The Protocol Library for Gemini Notebook
30+ protocols for Gemini Notebook, plus every protocol I publish from now on. Buy once, own the whole library forever. - Gemini Notebook for Writing Complete Edition
Turn your sources into articles, essays and books. - NotebookLM for Writing — The Prompt Card
The essential writing prompts on one card. - Gemini Notebook for Beginners: 101 Days to Mastery
A step-by-step path for beginners, one small skill a day.
