Popularity is the least important of four factors. The other three are harder to reverse, so they decide the niche.
Niche selection is the one decision that is genuinely hard to reverse, because it determines what you can make for years. The useful frame is not which niches are popular but which ones are sustainable for a channel that will not be on camera.
Three things. Whether you can produce it consistently without a subject-matter presence. Whether the subject supports a hook formula that is not simply a lie. And whether the platform's policy treats the topic as sensitive.
Popularity is the least important of the four and the one most tutorials lead with. A niche with modest demand and forty episodes of runway beats a saturated one where you are the twentieth channel.
History, mythology, science and nature. These carry narrative structure, do not require the creator to be an authority, and sit outside the sensitive-topic categories in AI labelling and monetisation policy.
The structural advantage is that the story carries the video. The viewer is watching because of what happens, not because someone is explaining it to them.
Finance, legal, medical and current affairs are treated as sensitive by YouTube's inauthentic content policy when presented with a synthetic persona, and by TikTok's AI spam detection in politics and finance.
This is not a prohibition. It is a materially higher bar, and the safe approach if you want to cover these topics is to appear on camera, which places you outside the category entirely.
The risk in picking one of the obvious niches is that everybody else picked it too, and the resulting catalogue looks identical across channels. That is a monetisation risk before it is a policy one.
Look for the structural angle rather than the subject. The same true-crime format with primary-source analysis is a different channel from the same format with a stock-footage montage, and the difference is visible in retention.
ReelsAudio renders faceless episodes with consistent voice and music, varied structure, and loudness normalised per platform.