How to Choose a YouTube Niche: The Four Numbers That Decide It, and Why the CPM Tables Should Not

The question gets asked as an identity question. What should my channel be about? What am I, on camera, for three years? That framing is why the answer takes months, and why it is so often a subject the creator likes rather than one that works.

Underneath it is a far more tractable problem. A niche is a bet on four quantities: how many people want this, how hard it is to be the one they pick, what one video costs you to make, and what a thousand views is worth when it arrives. Three of those can be estimated before you commit. The fourth — the money — is the one every listicle leads with and the one you control least.

What follows is that estimate worked properly: what a niche is on a platform that recommends videos rather than channels, how to read demand and competition with free tools, why the CPM tables should not decide this, and a four-week test that beats another month of deliberation.

A niche is an audience, not a topic

Start with how the machine sees you, because it does not see a category. YouTube's public description of its recommendation system lists the signals it works from: what a viewer has watched and searched for, what they subscribed to and liked, the topics they have shown affinity for, and what they do when a video is offered — click, ignore, dismiss, finish. YouTube describes the system as learning from tens of billions of these signals. None of them is a field on your channel where you wrote down your genre.

The practical consequence is the whole of this article. Your videos are not shown to a niche; they are shown to individual people whose viewing history suggests they will stay. Two subjects that share nothing on a category chart can be one niche if the same person wants both, and two videos in the same category can be aimed at people who would never watch each other's. So the definition worth using is behavioural: a niche is a group of viewers who reliably want the same kind of next video. The test is not "is this a category" but "can I name the viewer, and can I name the next four videos they would click".

This is also why the most common fear about niche choice is overstated. Changing subject does not trip some permanent genre assignment, because there is no such assignment. What it costs is the immediate advantage of publishing to people who already liked the last one — your subscribers and recent viewers are the first audience a new upload is offered to, and a video they do not want has to find colder surfaces to survive. That is a real cost, measured in one video's performance rather than in a channel-wide penalty.

The four numbers

Write these down for each candidate subject before you argue with yourself about which you like more.

Number	What it really asks	How to estimate it free	What bad looks like
Demand	How many people are already looking for or being shown this	YouTube autocomplete, Google Trends set to YouTube Search, view counts on the top twenty results	Nothing published in six months and nothing above a few thousand views
Competition	Whether a channel with no audience can be the one they pick	Sort similar-sized channels' uploads by views-to-subscriber ratio; study the outliers	Every big video belongs to a channel above a million subscribers
Cost per video	Whether you can still publish in month twelve	Make one video and time every stage honestly, including packaging	Twenty hours per video against competitors publishing twice a week
Revenue per view	What a thousand views converts into across all revenue lines	Ad RPM benchmarks, sponsorship rate guides, and whether the audience buys anything	Ad-poor audience with nothing to sell them and no sponsor category

Demand and competition are not independent, which is what catches people out: high demand is exactly what attracts competition, and the emptiest fields are usually empty for a reason. What you want is not the extreme of either column but a gap — real demand where the supply is weak, badly packaged, or aimed at a slightly different viewer than the one you can serve.

Demand: how to check it before you commit

There are two kinds of demand on YouTube, and a new channel can only reach one of them cheaply. Search demand is people typing a question. It is reachable with no audience at all, because the query does the targeting, and it compounds: a video that answers a question keeps answering it. Browse and suggested demand is YouTube choosing to show your video to someone who was not looking for it — where the enormous numbers live, and much harder to enter cold. The breakdown of traffic sources covers how each surface behaves; for niche choice, the point is that a subject with no search demand gives your first fifty videos no floor.

To estimate search demand without paying for a tool: type the subject into YouTube search and read the autocomplete list, which is generated from real queries. Then set Google Trends to YouTube Search rather than web search and compare candidates over five years — the shape matters more than the level. A flat line is evergreen demand. A repeating annual wave is seasonal, and has to be published into four to six weeks early. A single spike that never returned is a dead trend wearing a niche's clothes.

Then look at the supply. Search the three or four queries you care about most and read the top twenty results for two things: whether the view counts are meaningful, and how old they are. Old videos holding the top of a query are the best signal a new channel gets: demand exists and nobody has answered it properly since. The full keyword method sits in the search guide.

Competition: read the outliers, not the giants

The instinct is to check whether big channels already cover the subject. Wrong measurement: big channels cover everything, and their presence says nothing about whether a new video can win an impression.

The measurement that works is the ratio between a video's views and the subscriber count of the channel that published it. Find fifteen channels roughly the size you will be in a year, list their recent uploads, and sort by that multiple. A video doing three or four times its channel's subscriber count was chosen on the strength of the video, not the brand. A field with several of those a month is one where packaging and topic still beat audience size, and you can enter it. A field where every large video belongs to a household name is one where you are competing on production budget.

The second thing to read is packaging quality. If every thumbnail on the first page is the same arrangement of the same face with the same yellow text, that homogeneity is an opening — the viewer has no way to choose and will pick whatever looks different. If the packaging is already excellent and varied, you are late, and the cost of entry is craft rather than ideas. The outlier scan and the log that makes it useful are set out in the trending ideas piece.

One caution about scale. Pew Research Center's 2019 study A Week in the Life of Popular YouTube Channels identified 43,770 channels with at least 250,000 subscribers as of late 2018, and those channels alone published nearly a quarter of a million videos — 48,486 hours — in the first seven days of 2019. That is the volume you are choosing to be legible inside, and it has not got smaller since.

Cost per video: the number that ends channels

Nobody quits because they picked an unprofitable subject. They quit because the subject they picked costs nine hours a video and they have four hours a week.

Do this arithmetic before anything else. Make one complete video in the candidate niche — scripted, shot, edited, packaged, published — timing every stage without rounding down, then multiply by the cadence your competitors maintain. If the answer exceeds the time you will genuinely have in month eight, when the novelty has gone and the channel is not yet paying for itself, the niche is wrong for you however good its other numbers look. What a sustainable cadence is, and why it matters more than the consistency creators obsess over, is in the piece on how often to post.

Production cost varies more between niches than revenue does, and it varies visibly in advance. Software tutorials cost a screen recording. Commentary costs a microphone and a strong opinion. Travel costs travel. Anything involving a build, a recipe, a location or another person costs a day of logistics per upload and cannot be caught up on a bad week. A niche where the cheapest possible video is still expensive has no room for the experiments that teach you what your audience wants.

Revenue per view, and why the CPM tables should not decide this

This is the number the internet wants to talk about, and it is the least reliable of the four. Published CPM-by-niche tables for 2026 put finance somewhere between the mid-teens and sixty-five dollars per thousand impressions, and gaming somewhere between one and eight. The direction is consistent across sources: bids for audiences with money and buying intent are multiples of bids for entertainment audiences. The magnitudes are not consistent at all — the tables disagree with one another by several times on the same niche in the same year, which marks them as estimates assembled from samples rather than measurements.

They also measure the wrong thing. CPM is what advertisers pay for a thousand impressions. What lands in your account is RPM, after YouTube's share and after the fraction of your views that carried an ad at all, and it moves on variables unrelated to your subject: where your audience lives, whether your format admits mid-roll breaks, whether you publish in November or February. Long-form ad revenue and the Shorts pool are split on different terms too, which is why a Shorts-heavy channel and a long-form channel in the same niche report different worlds. The arithmetic, including the seasonality that makes Q4 the best-paid quarter, is in the piece on what YouTube pays per thousand views.

Two corrections, then, before you let money into the decision. Sponsorship, not ads, is where niche choice pays: published rate guides put finance and B2B integrations several times above gaming on the same view count, because the advertiser is buying a customer rather than a brand impression, and a sponsor's rate does not depend on ad fill. If you take that route, the disclosure rules apply.

And revenue per view is one term in a product where the other term is bigger. A niche paying four times as much per thousand views is worse than a niche you can get ten times the views in. The fields that look ad-poor are frequently the ones where the audience buys directly — a template, a programme, a course — and revenue lines like memberships and affiliate shopping are indifferent to what advertisers think of your subject.

How wide should the niche be

Too narrow and you run out of videos; too broad and no single viewer has a reason to subscribe. Two tests settle it.

The first is the thirty-idea test. Write thirty specific video titles in the candidate niche — titles, not themes, the kind you would actually publish. If you cannot reach thirty, the niche is too narrow to carry a channel and should be a series inside a wider one. If thirty comes easily but you cannot describe one person who would watch twenty of them, it is too broad, and you have written a calendar for four channels.

The second is the ceiling test. Multiply roughly how many people could plausibly care by how often they would watch. A small possible audience that watches weekly is a real channel. A large possible audience that would watch one video, once, is a good video and a bad channel. This is where format-shaped niches fail: reviews, reactions and Shorts are production methods, and no viewer's interests are described by a method.

The shape that works is a beachhead. Start narrow enough that a specific viewer recognises the video as made for them, then widen along the audience rather than the topic — the adjacent subject that same person already watches, which is what interest affinity means in the signals above. Widening along the topic instead, into things that merely share a category with your work, is how channels end up with an audience that no longer recognises their uploads.

You are also choosing a design problem

Every niche comes with a visual inheritance, and it decides how much of your outcome rests on packaging. Some subjects hand you a package: food gives you the dish, gaming gives you the character and the moment, transformation content gives you a before and an after, which is a thumbnail with the argument already in it. Others give you nothing photographable at all — software, finance, law, most B2B, most audio-first formats. There the tile has to be constructed rather than captured, and the channels that win built a visual language instead of screenshotting a dashboard.

That is not a reason to avoid the unphotogenic niches, where the competition is weaker for exactly this reason, but it is a reason to budget for it. Which template fits a subject, how much of the frame a face should hold, and which colours read as credible rather than cheap vary sharply between fields; the per-niche version sits on the niche pages, and the underlying patterns in the six thumbnail templates.

The niche also decides whether you can stay anonymous. Faceless formats work, but they give up the cheapest attention device on the platform and have to replace it deliberately — the trade is set out in the piece on faceless channels.

What 2026 changed about this decision

Two changes alter the choice rather than decorate it.

The first is the monetisation floor under mass-produced content. In July 2025 YouTube rewrote the channel monetisation policy on repetitious content, renaming it inauthentic content, effective on 15 July; trade coverage including Social Media Today's report stressed that the underlying rule was not new but that the wording now plainly captures templated, mass-produced and duplicated uploads. YouTube's stated position on AI is that the tools are welcome and the output has to carry original human value. As a niche filter, that rules out a whole family of subjects built on volume — automated quiz, slideshow and narrated-listicle formats — not because they are AI-assisted but because nothing distinguishes one upload from the next.

The second is that packaging has become cheap to test. Testing inside Studio, which began as a three-thumbnail experiment, now covers titles as well and has been extended to the broad population of channels with advanced features, with the winner decided on watch time per impression rather than clicks alone. The packaging assumptions you make on entering a field no longer have to be guesses held for a year. How the test works and how long to run it are in the piece on Test & Compare.

A four-week test that beats another month of thinking

Deliberation has a poor return here. Three published videos tell you things no amount of research will.

1. Week one: evidence. Thirty titles. Fifteen similar-sized channels logged, their outliers sorted by views-to-subscriber multiple, ten written down with a sentence each on why the package worked. Autocomplete and Trends for the three queries you most want.
2. Week two: publish three. Same viewer, same package family, three different topics. One video tells you nothing, because variance between videos inside a niche is larger than the difference between most niches.
3. Week three: read them. Impressions tell you whether the system found anyone to offer them to. Click-through rate tells you whether the package earned the click, against the benchmarks in the piece on what counts as a good CTR. Average view duration tells you whether the video paid for the click. Returning viewers, small as that number will be, tells you whether anyone wanted a second one.
4. Week four: decide. Impressions but no clicks is a packaging problem, the cheapest of the three to fix. Clicks but no retention is a promise mismatch. No impressions at all, repeatedly, where similar-sized channels are getting them, is the answer you were looking for.

Three videos is not a statistically strong sample and nobody should pretend otherwise. What it produces is a direction and a list of specific problems, which is more than any spreadsheet of CPM estimates will.

Pivoting without burning the channel

If the test says no, changing subject costs less than the folklore claims and more than nothing. What it costs is the head start: a new upload is offered first to the people who liked the last ones, and in a pivot those people are the wrong audience, so early videos underperform what they would do on a clean channel until the audience turns over. Expect two or three uploads of that and do not read it as a verdict on the new niche.

What it does not cost is a permanent mark, because recommendation works per video and per viewer. Three moves make the transition cheaper: keep the old catalogue rather than deleting it — unlist anything that actively misrepresents the channel, for the reasons in the piece on deleting old videos; rebuild the packaging first, so the new direction is legible from the first upload rather than the fifth; and publish two or three bridge videos both audiences would click, or take the clean break and accept the slow month.

Five ways this decision goes wrong

• Choosing on passion alone. Passion decides whether you last, not whether anyone watches. It belongs in the cost column.
• Choosing on CPM. A high rate on views you never get is zero.
• Copying a large channel's format. Theirs works because their audience exists. Yours has to work on a stranger.
• Picking a cadence you cannot hold. A field whose competitors publish weekly is not available to a monthly channel.
• Choosing an audience that buys nothing. If nobody there ever spends money on the subject, ads are the only revenue line you will have.

Frequently asked questions

Does YouTube punish a channel for changing niche?

There is no channel-level genre penalty in anything YouTube publishes about how recommendations work; the system is described as matching individual videos to individual viewers. The real cost is that your subscribers and recent viewers are the first audience a new video is shown to, and if they do not want it, the video has to earn colder impressions. That shows up as two or three weak uploads, not a permanent handicap.

Should I pick a high-CPM niche I do not know well?

Generally no. The credibility bar in finance, law and B2B is the thing the audience is actually paying attention to, and it is visible within thirty seconds. If the subject genuinely interests you and you are willing to do the reading, the economics are good. If the only attraction is the rate table, you will be outcompeted by people for whom it is not.

Can one channel cover two niches?

It can if one viewer wants both, and the test is whether you can name that person. Where the audiences genuinely do not overlap, a second channel costs you your whole starting advantage and all your remaining time, which is why it is usually the wrong answer before the first channel works.

How long should I give a niche before deciding?

Three videos is enough to spot a packaging or promise problem; a dozen is a reasonable floor for judging the niche itself. What you should not do is give it a year without reading the middle of it. Impressions, click-through rate and retention will tell you which of the four numbers you got wrong long before the subscriber count does.

Sources

Checked in September 2026. Rate tables are third-party estimates that move constantly and disagree with each other; confirm current figures and current YouTube documentation before acting on any of them.

• How YouTube Works — recommendations
• Pew Research Center — A Week in the Life of Popular YouTube Channels (2019)
• Social Media Today — YouTube clarifies its inauthentic content monetisation update
• Plagiarism Today — YouTube targets inauthentic content
• YouTube Help — channel monetisation policies
• YouTube Help — A/B test titles and thumbnails
• vidIQ — highest-paying YouTube niches, CPM and RPM by category
• Google Trends (set to YouTube Search)

The decision, in one page

Pick the subject where real demand meets weak or homogeneous supply, that you can publish into every week for a year without dreading it, aimed at an audience that eventually buys something. Check the money last, knowing the tables are estimates and that view volume moves your revenue further than the niche label does. Then stop deciding and publish three videos, because the first honest data about a niche arrives the week you enter it.

The input you control completely is the one most people treat as an afterthought. Demand and competition are given to you, cost per video is mostly given to you by the format, and the package is decided fresh on every upload — in the unphotogenic niches it is the whole competitive position. That is the job Thumblore exists for: a tile that looks like it belongs in your niche without a design afternoon per video, so the cadence you promised yourself in week one survives to month twelve. If you are still choosing, the niche pages show what converts in eighteen of the biggest fields on YouTube, and the guide to a channel's first ninety days covers what to do once the choice is made.

One niche chosen badly and worked for a year beats four chosen well and abandoned in six weeks. The four numbers exist to stop you doing the second thing.