Beginner8 min

Generate Ideas People Actually Search For

AI is a brilliant brainstorm partner and a terrible judge of demand. It will cheerfully hand you twenty ideas nobody is searching for. Your job is to use the model for volume and angles, then validate against the real world. This lesson teaches that two-step rhythm.

Step 1: Mine real questions, not topics

Topics are vague and unclickable. Questions are specific and map to what people type into search. Ask the model to turn your niche into the exact questions your viewer would ask out loud.

idea prompt
Niche: I help renters cook one-pan dinners in under 20 minutes.
List 20 specific questions my viewer would search for, phrased
the way they would actually type them. For each, add:
- the real frustration behind it
- a sharper title angle
- whether it suits long-form or a short

Step 2: Validate before you commit

  1. Search the exact phrase on your platform. Are the top results old, thin, or missing?
  2. Check whether they truly answer the question or just dance around it.
  3. Confirm you can add something real: a result, a teardown, a faster method.
Claude Sonnet 4.6
You
Give me 20 one-pan dinner questions a broke renter would search, with a sharper title for each.
Agent
1) 'one pan pasta no draining' -> 'The lazy pasta you cook in one pan (no straining)'. 2) 'cheap protein one pan' -> ... (18 more).
You
Now rank the top 5 by how easy it is to beat the current videos on this.
Brainstorm in chat, then take the survivors to real search results.
The model brainstorms, the data decides
Never publish an idea just because the AI suggested it. Treat its list as candidates. The search results and, later, your own analytics are the only honest judges of demand.

Example result: from twenty AI ideas, maybe four survive validation. Those four are worth more than the original twenty because you know real people are already looking for them.

Hands-on tasks