Critical Reading in Practice: Using AI Review Assistants to Deconstruct Arguments

“That article was pretty good” or “that argument feels wrong” — that’s as granular as most people get after reading an opinion piece. It’s not that we lack judgment. It’s that critical reading is a skill that requires deliberate practice. It demands doing several things in parallel while reading: identifying premises, tracing logical chains, evaluating evidence strength, detecting overgeneralizations. Nobody is born doing all of this simultaneously.

Yomitomo’s Review Assistant matrix breaks the dimensions of critical reading into distinct AI role specializations. You don’t ask one AI for a generic evaluation. You ask different AI roles for feedback from their respective angles. Ultimately, you integrate those perspectives and form your own judgment.

Not Making Judgments for You — Helping You Notice What You Missed

The design philosophy of Review Assistants is clear: they don’t hand down conclusions. What they do is prompt — prompt you that this claim might have an unexamined premise, prompt you that this causal chain might have a missing link, prompt you that this data source might be insufficient.

You remain the one making the judgment. Review Assistants simply help you see what you might have overlooked.

Four Review Assistants, Four Critical Dimensions

Liang Zhengyan: Evidence Steward

When you read a passage containing factual claims — “Market X is worth Y billion,” “Company Z’s data shows” — mark it as “Key Point” or “Assumption,” then type @Liang Zhengyan in the discussion area.

Liang Zhengyan won’t search the internet to verify facts (that’s impossible — model knowledge has a cutoff date). What she does is methodological review: Is this data from primary research or a second-hand report? Was sample size disclosed? Was there a control group? Does the data’s time range match the conclusion?

What you get isn’t necessarily a “this is fake” verdict. It’s a prompt: “the evidence strength behind this claim may be insufficient — take note.” These prompts make you more sensitive to similar issues in subsequent reading — that’s the training effect of critical thinking.

He Mingheng: Logic Reviewer

When you encounter a causal claim — “Because of X, therefore Y,” “X’s growth caused Y’s decline” — type @He Mingheng.

He Mingheng checks for several common reasoning problems: Is there a skipped intermediate variable between X and Y? Is correlation being mistaken for causation? Are necessary and sufficient conditions being conflated? Does the conclusion overreach beyond the premises?

A seemingly tight argument, under He Mingheng’s deconstruction, often reveals leaps — especially those gaps authors paper over with rhetorical flourish.

Su Dingbai: Risk Reviewer

When you read an article full of “inevitably,” “obviously,” “beyond doubt” — type @Su Dingbai.

Su Dingbai flags potential cognitive biases and risks: Is this conclusion overgeneralizing from a few cases? Is there survivorship bias? Is the author’s stance compromising objectivity? Does this prediction underestimate uncertainty?

These prompts are especially useful when reading investment reports, industry analyses, and policy commentary — texts that often package themselves as highly professional, but whose core assumptions may not withstand scrutiny.

Tang Jian: Final Editor

Tang Jian plays a slightly different role — she doesn’t review the author. She reviews you.

When you’ve written a long block of your own thoughts in an annotation, or finished a distillation draft, type @Tang Jian. She compresses redundancy, clarifies expression, and makes your judgment sharper and more readable. This looks like editing, but it’s fundamentally an extension of critical reading — you’re using AI to scrutinize your own expression, ensuring that your critique is structured rather than just emotionally “I feel like this is wrong.”

Integrated Critique in the Distillation Window

Review Assistant feedback in annotation discussions is scattered — one issue per highlight. The Distillation window provides integration — you gather all your critical annotations about an article and write a consolidated critical analysis.

After writing your analysis draft, you can invite Review Assistants to review your analysis. This is a meta-level operation: you’re using AI to check whether your own critique is rigorous.

The final published distillation isn’t a “this article was good/bad” book report. It’s a structured critical analysis — containing premise examination, logic analysis, evidence assessment, and your final judgment.

Who This Is For

Anyone who wants to improve their critical thinking — students, researchers, investors, product managers. If you’re unsatisfied with “the article feels off but I can’t articulate why” and want to turn critical reading from intuition into a trainable skill.