Run the AI Niche-Research Assistant (Nancy)

The manual version of niche research is a spreadsheet, and you already know the one I mean. Twenty tabs open. A column for CPM, a column for competition, a column for whether you could realistically make the videos without a camera. Days of it. Sometimes weeks. And at the end you've narrowed a list of forty down to a list of thirty-one.

Nancy collapses that entire pass into one automated run.

She starts from a master list of niches already known to do two things: pay well, and make people curious enough to click. High CPM on one axis, high curiosity on the other. Then she narrows, and not to another maybe-pile. Down to five scored opportunities, so the shortlist arrives with an opinion attached instead of leaving you to break the tie. She also leans toward hybrids, two verticals fused into one lane rather than a single broad category, because that fusion is usually where the room is.

Each of the five comes back as a card. Faceless-friendliness, so you know the thing can actually be made without you on screen. Monthly search volume, so you know people are looking. Competition, so you know who's already in there. CPM, so you know what a view is worth. Every one of those is a number you'd otherwise hunt down by hand, one niche at a time.

But the numbers aren't the useful half. Underneath them, the card explains itself. Why the niche is currently underserved. Why a viewer clicks the thumbnail. Why YouTube would push the video into recommendations at all. Read that part slowly. The CPM tells you what a win pays, and the reasoning tells you whether a win is even available to you.

Then there's the money model. Nancy projects earnings at named subscriber milestones, so you can see the shape of the thing before you've made video one. And she goes past ad revenue into digital products that would suit the audience and sponsors matched to the topic. That matters, because for a lot of niches ad money is the floor rather than the ceiling, and a niche that looks ordinary on CPM alone can look very different once the off-platform side is on the page.

There's a second assistant worth knowing about. Sam takes vidIQ data and builds content pillars out of it, each one with a share of your upload schedule attached and a reason it exists, on both sides: what the audience wants from it and what the algorithm does with it.

Now the part people skip, and it's the part that actually finds you something.

Don't hand the whole job over. Pair the AI research with your own browsing. Ask the model for a handful of research keywords per niche, then go search those keywords on YouTube yourself. You'll surface channels neither method would have found alone, and more importantly you'll start to understand why the niche works instead of just being told that it does.

Keep that as a loop, and run it deliberately. Pick one object inside your niche. Ask ChatGPT for ideas around it. Organise what comes back into pillars. Search those on YouTube. If most of the competition is on camera, that's your opening, because the demand is proven and nobody's serving it faceless. Then go again.

Two ways to make the AI half heavier. Ghost nodes: take one obscure fact and expand it into dozens of related topics. And straight mass-generation: prompt for unique faceless niches, for cross-products between two categories, for entire content pillars at once. You'll get lists of titles nobody has covered.

Five scored options with the homework done. Which one you can stand to make videos about for a year is still yours to answer.