How to Use ChatGPT Without Losing Critical Thinking

Short answer

To use ChatGPT without losing critical thinking, keep three jobs human: the first answer, the strongest objection you can think of, and the final decision. Use ChatGPT between those jobs as a critic, not as the owner of the work.

The danger is not that ChatGPT gives answers. The danger is that it gives the first answer so often that your own first answer stops appearing.

The safe sequence

Use this sequence for serious work:

1. State your own view first.
2. Ask ChatGPT to attack it.
3. Check claims against sources.
4. Revise in your own voice.
5. Teach back the result without looking.

If you skip step one, the model becomes the frame. If you skip step five, you may have a polished output that you do not fully own.

Prompt pattern: critic, not ghostwriter

Use this:

I wrote the draft below. Do not rewrite it yet. First, identify the weakest claim, the strongest counterargument, the missing evidence, and any place where the voice becomes generic. Then ask me three questions before suggesting edits.

Avoid starting with:

Write this for me.

The second prompt may save time. The first prompt saves the circuit.

For writing

Before ChatGPT enters, write the messy version yourself. It can be short:

• the point
• the audience
• the claim
• the evidence
• the thing you are unsure about

Then ask the model to challenge the structure. When it suggests rewrites, do not paste blindly. Decide which objection is real, then rewrite manually.

For research

Ask ChatGPT to map possible sources, but do not treat the answer as evidence. The useful flow is:

1. Ask for source candidates.
2. Open the primary source.
3. Read the relevant section.
4. Write the claim yourself.
5. Use AI to test whether your claim overreaches.

This is why the Anti-AI Brain site keeps a public research ledger. A claim is only useful if a reader can inspect what carries it.

For decisions

The rule is decision first, AI second.

Write:

My current decision is X. I am choosing it because A and B. The risks I see are C and D.

Then ask:

What would make this decision wrong? What evidence would change it? What am I underweighting?

That keeps AI in the adversarial-review role.

For learning

Do not ask for the summary first. Try recall first.

After a lecture, article, meeting, or chapter, write five remembered points before asking ChatGPT to summarize. Then compare your recall to the model’s summary. The gap is useful. It shows what you did not encode.

The simplest daily rule

For every serious prompt, add a ten-minute wall:

I will think for ten minutes before asking.

The ten minutes do not need to produce brilliance. They only need to produce a human first move. That is where critical thinking stays alive.