You're getting generic ideas back because you're asking a generic question. Here's the three-step fix — constraints in, real curiosity mined, ranked by what actually sparks a reaction.

Why "give me content ideas" always disappoints

Vague ask, vague output. It's not a flaw in the AI — it's a flaw in the prompt. When you give the model nothing to work with, it reaches for the middle: the safest, most averaged-out interpretation of your niche. That's where "10 tips for productivity" and "why mindset matters" live.

The problem isn't that AI can't generate good content ideas. The problem is that a blank prompt is an instruction to be generic. You're not giving it a reason to be specific, opinionated, or timely — so it isn't.

The three steps in detail

1. Feed it constraints before you ask for anything. Tell it your niche, name three things people in that niche are genuinely arguing about right now, and specify the exact format you're making — Reel, carousel, newsletter, thread. The constraint is what forces specificity. Without it, the model defaults to broad. With it, the ideas have nowhere to hide in the middle.
2. Ask it what questions your audience is actually asking this week — then turn each one into a hook. This is the move most people skip. Instead of inventing ideas from nothing, you're mining real curiosity: what are people typing into search, asking in DMs, debating in comment sections? Each genuine question is already a hook waiting to be written. You're not brainstorming — you're excavating.
3. Ask it to rank its own ideas by which would spark the most disagreement or curiosity. Don't let it hand you a flat list and walk away. The ranking step is what separates scroll-stoppers from filler. An idea that sits at number seven on an unordered list might be the sharpest thing in the batch — but you'd never know because nothing forced a comparison. Make it compare.

The exact ideation prompt 💬

Here's the prompt structure that runs all three steps in one go. Swap the bracketed parts for your own niche and format — everything else stays as written.

I create content for [YOUR NICHE].

Here are three things people in this space are actively debating right now:
1. [DEBATE 1]
2. [DEBATE 2]
3. [DEBATE 3]

The format I'm making is: [e.g. 60-second Reel / carousel / newsletter intro]

Do two things:
First, list 8 questions that people in this niche are genuinely asking or searching for this week — real curiosity, not textbook topics.
Second, turn each question into a hook for the format above.

Then rank all 8 hooks from most likely to spark disagreement or strong curiosity to least. Explain in one sentence why you ranked the top three where you did.

Before and after: same niche, different prompt

To make the difference concrete — here's what the output looks like for a personal finance creator, with and without the structure above:

Same tool. Same creator. The constraint is doing all the work.

How to get the most out of it 🎓

• Update your three debates regularly. What people are arguing about in your niche shifts week to week. The more current your debates, the more timely the hooks — run this fresh every few weeks rather than recycling the same prompt.
• Be specific about format. "Reel" is better than "video." "5-slide carousel" is better than "carousel." The model will calibrate hook length and energy differently depending on the container.
• Push back on the ranking. If you disagree with where something landed, say so. Ask it to argue the case for moving a lower-ranked idea to the top. That back-and-forth often surfaces the angle you actually want.
• Use the questions layer as a research step too. The eight questions it generates are worth keeping in a doc. They're a map of what your audience is actually curious about — reusable signal, not just one session's output.

The honest bit ✅

• The prompt doesn't know your audience better than you do. It's working from what's publicly available and statistically common. Your lived read on what your specific audience is feeling this week will always add something the model can't supply — treat its output as a starting point, not a finished brief.
• Ranked by "most disagreement" can skew toward provocation. Controversy and genuine curiosity aren't the same thing. Check that your top-ranked hook is interesting, not just inflammatory — especially in niches where trust is part of the value you offer.
• Real debates need you to actually know your niche. If you're not already reading what your audience reads, watching what they watch, or seeing what they're arguing about in comment sections, the debates you feed this prompt will be stale. The prompt amplifies your signal — it doesn't replace it.
• This is an ideation tool, not a publishing pipeline. A hook generated here is a rough shape, not a finished caption. The idea is still yours to sharpen, contextualise, and make sound like you.