Critical Thinking Bot

What it is

A chatbot that refuses to answer you.

That is the entire design, and its creator says so directly. Shae Omonijo, a historian and PhD candidate in History at Harvard, built it because "it's very easy to offload your critical thinking" to a general assistant. Her description of what she made: "I coded a critical thinking chatbot that only asks you questions and never gives you answers. It's a chatbot that refuses to think for you."

The front page gives you one instruction: "State one idea, claim, or argument. Discuss it until it becomes clear, concise, and defensible."

How it works

- No answers. It will not solve the problem, take a position, or write anything for you.

- Persistent questioning. It keeps asking targeted questions to clarify what you actually mean, surface the assumptions underneath, and press you for evidence.

- Productive friction. The point is to slow the work down at the stage where slowing down is worth something.

You start by picking how you want to be questioned:

- Clarify: work out what you actually mean.

- Strengthen: build the case.

- Test: attack it.

- Explore: open the surrounding territory.

- Decide: get to a call.

At the end you get a recap rather than a verdict, under headings like "Where it landed", "How the thinking changed", and "Assumptions still carrying it". Note what that last one is doing. It hands you back the things you never justified, still unjustified. Most people cannot list their own assumptions, which is exactly why arguments collapse on contact.

The site carries its own disclaimer: "Experimental and educational only. Not advice."

Why it matters

Every other AI interface you touch is optimised to reduce friction. This one adds it, deliberately and in one specific place, and that inversion is the whole reason to look.

The complaint about chatbots as thinking aids is legitimate and well evidenced: an assistant tuned for helpfulness will tend to agree with you, elaborate your existing position, and hand back a tidier version of what you already believed. You finish the session more confident and no more correct. A tool that structurally cannot do that is a different instrument, even if the model underneath is the same.

Declaring your mode up front does real work too. Choosing Test is a commitment made before you know what it will find, which makes it harder to quietly slide back into having your priors flattered.

Be clear about the limit. It tests whether your argument holds together, not whether it is true. Nothing here checks your facts, and "defensible" means internally coherent. Use it on the structure of a claim, not on whether the evidence exists.

Where it genuinely earns its place is the hour before you publish something with your name on it, when the useful question is which load-bearing assumption you never actually articulated.

What to try first

Take a claim you already believe and are slightly annoyed at having to defend. Run Test, not Strengthen. Read the assumptions before the recap. That is where the discomfort lives, and the discomfort is the product.

Key takeaways

- Built by Shae Omonijo, Harvard History PhD candidate, as part of her Critical Thinking in the Age of AI project.

- It only asks questions. It will not answer, decide, or write for you.

- Five modes make you declare what kind of pressure you want before you get it.

- It tests coherence, not accuracy. Fact-checking is still your job.

- Free, and deliberately narrow. The friction is the feature.