Ideas vs Facts: Choosing the Right Mode

A single chat box pretending to be good at everything is a quiet source of bad answers. Brainstorming and fact-finding are different jobs with opposite rules: one rewards imagination, the other punishes it. Trying to do both in the same mode means one of them is always being done wrong.

That's why there are two modes — Ideas and Facts — and why picking the right one is half the work.

Ideas: think and create from an open brief

Ideas mode is for when the answer doesn't exist yet and you're trying to make it. Drafting, naming, planning, reframing a problem, exploring options you haven't thought of. Here you want the model to range freely, propose, and improvise. Invention is the point.

Reach for Ideas when you'd say things like:

• "Help me outline a talk about…"
• "Give me ten angles on…"
• "Rewrite this so it sounds warmer."
• "What am I not considering here?"

The mediator still does its job up front — turning your rough request into a clear, well-scoped prompt — but the conversation that follows is generative by design.

Facts: answer strictly from sources you trust

Facts mode is the opposite discipline. The answer already exists, in a document or a source, and your job is to extract it faithfully. Here improvisation is a bug. You don't want a plausible figure; you want the real one, with a quote you can check and arithmetic that's exact.

Reach for Facts when you'd say things like:

• "What does this contract say about termination?"
• "Total the figures in this spreadsheet."
• "Does this report support the claim that…?"
• "Summarize these three documents — and cite where each point comes from."

In Facts mode, "the sources don't say" is a correct answer, and every claim can be tied back to a verifiable quote.

A simple rule of thumb

If you're making something, choose Ideas. If you're checking something, choose Facts. When a task has both — say, drafting a proposal from a real brief — start in Facts to pin down what's true, then move to Ideas to build on it.

The mistake isn't using AI for the wrong task. It's using the wrong mode for the right task.

Two modes, two jobs. Knowing which you're in is the difference between an answer you can trust and one you only hope is right.