Challenge Our Assumptions · Using AI for Research

This mode sits opposite “as fast and as big as possible.” You want AI to talk back — critique a recruiting plan, push on a discussion guide, look for overreach in early insights, or test an idea against real constraints.

Friction on purpose — before claims harden.

Remove all friction and you may remove the engine of real learning. Confidence that never met a counterargument is not rigor.

AI can list risks and alternatives. You own the ethical call. Do not treat generated dissent as field evidence.

Core principles

Make hunches explicit enough to see — and to test.
Attack early insights: alternative explanations, over-weighted voices, what you would have to ignore.
Red-team discussion guides for leading questions and performed answers.
If the model only agrees, ask it to argue the opposite, then steelman your view.
Success often feels slower — if the critique changes a question, recruit, or claim, that is the win.

Go deeper: AI Judgment — verification tiers

Check yourself

What is the point of Challenge assumptions mode?

Do this in Copilot

Run the early-insight attack on your themes. Change one thing in your plan, guide, or claim. Note what you changed in one sentence.

Paste this into Copilot Chat and work through it before moving on.

Pressure-test early insights

Here are candidate insights from research on [TOPIC]:
[PASTE]

Attack them. For each:
- Strongest alternative explanation
- Whose voice is over-weighted?
- What would we have to ignore for this to stay true?
- Business / technical / ethical / ecosystem constraint that could kill or reshape it
Then rank: which insight is most fragile, and why?
Do not soften the critique. Do not add new “insights” of your own.

Techniques: Red team · Assumption surfacing

For a guide: paste the outline and ask for leading questions, missing emotion/workaround probes, and baked-in assumptions.

Open Copilot

Did you run this in Copilot? Mark complete when you have tried it.