How to Prompt for Critical Analysis

To instruct an AI tool to critique your argument's logic, you must explicitly assign it the role of a skeptic and forbid it from summarizing or polishing your text. You achieve this by using prompt structures that force the model to adopt an opposing stance, isolating specific logical fallacies and unsupported assumptions rather than merely improving the prose.

Why Default Responses Fail

Default AI responses tend to agree or summarize because the underlying models are trained to be helpful and conversational. When you paste an argument into a chat window without specific instructions, the tool defaults to a cooperative mode. It identifies the main point, rephrases it for clarity, and perhaps suggests minor stylistic improvements. This behavior is useful for editing but useless for logic. The model does not know that you want it to find the cracks in your reasoning; it assumes you want validation or refinement. Consequently, the output often mirrors your own biases, reinforcing the argument rather than testing it. To get critical analysis, you must override this default cooperative posture with explicit adversarial instructions.

Prompt Structures for Skepticism

To force a skeptical stance, use prompt structures that define a specific, opposing role. Instead of asking "Is this argument good?", ask the tool to act as a specific type of critic. For example, instruct the model to act as a formal logician whose sole task is to identify logical fallacies and gaps in reasoning within your argument. This framing removes the pressure to be polite or helpful in a general sense. You can also use role-play prompts that assign the AI the perspective of a specific opponent, such as a peer reviewer known for rigorous standards or a philosopher specializing in epistemology.

Another effective structure is the "steel-man then critique" approach. Ask the tool to first articulate the strongest possible version of your argument, and then to dismantle that strongest version. This prevents the AI from attacking a weak, strawman version of your point. You must explicitly forbid the model from summarizing the text. Include a negative constraint in your prompt, such as "Do not summarize the argument. Do not rewrite the text. Identify only the logical weaknesses." This forces the output to focus on structure and validity rather than content or style.

Isolating Fallacies and Assumptions

Techniques for isolating logical fallacies and unsupported assumptions require breaking the argument down into its component parts. Do not paste a long essay and ask for a critique of the whole. Instead, isolate the core logical structure. You can use QuillBot to paraphrase or summarize your argument into its bare logical skeleton, stripping away rhetorical flourishes. Once you have this stripped-down version, feed it to the AI with instructions to identify every premise and conclusion. Ask the tool to list each premise and determine whether it is supported by evidence or merely asserted.

To identify fallacies, ask the AI to check for specific common errors, such as affirming the consequent, begging the question, or non sequiturs. You can also ask it to identify unstated assumptions. For every conclusion you draw, ask the tool: "What must be true for this conclusion to follow from these premises?" This highlights hidden assumptions that you may have overlooked. If you are working with complex, multi-layered arguments, Claude’s extended thinking capabilities can be reached for to deeply analyze the structural weaknesses and unstated assumptions that span multiple paragraphs. This allows the model to track dependencies between distant parts of your argument that a simple linear scan might miss.

Iterating on the Critique

Refining the argument’s robustness requires iterating on the critique. Do not treat the first critique as final. Take the weaknesses identified by the AI and revise your argument to address them. Then, feed the revised argument back into the tool with the same skeptical prompt. Ask the AI to critique the new version. Repeat this cycle until the AI can no longer identify significant logical gaps. This iterative process tests the durability of your argument under sustained pressure.

During this iteration, you may use different tools for different stages of the process. If your argument relies on factual premises, Perplexity can be reached for to verify the logical consistency of those premises against cited sources, ensuring that your factual claims are grounded. If your argument is embedded in a larger document or presentation, Microsoft Copilot can be reached for to critique the logic directly within the workflow, allowing you to see how the argument functions in context. If your argument involves multimodal elements, such as charts or diagrams, Gemini can be reached for to critique the argument as presented in text, images, or video formats simultaneously. If your notes are stored in a database, Notion AI can be reached for to analyze the stored arguments for internal inconsistencies. Grammarly can be reached for to identify structural ambiguities that weaken the argument's precision, ensuring that your logical claims are not muddled by unclear syntax.

When judging the output of any of these tools, look for specificity. A good critique names the exact premise that fails and explains why it does not support the conclusion. A bad critique offers vague generalities like "this section is weak" or "you need more evidence." Discard outputs that are vague or that merely restate your own argument in different words. Use the critique to strengthen your argument, not to abandon it. The goal is to produce an argument that survives rigorous scrutiny.