YouTube Channel Analyzer — Reach, Audience Language & Sponsorship Fit

Give it a list of YouTube channels. Get back what you actually need before you offer anyone a sponsorship.

Subscriber counts do not tell you what a video will get, and they do not tell you who watches. This Actor reads a channel's recent uploads, a sample of its real comments, and its recent video descriptions, and returns the numbers a brand needs — each one with the evidence behind it.

What you get, per channel

Field	What it means
Expected views	The median of recent uploads — not the average, and not subscribers
Typical low and high	Where the middle half of their uploads land, so you can see a steady channel apart from a lottery
Outliers	The uploads that beat their own median by 2×, named, so you can see what goes big
Formats	What share of recent titles declares a review, tutorial, listicle, vlog, challenge or Q&A — and what share says nothing
Days since last upload	So a big number from a dormant channel cannot mislead you
View trend	Whether recent uploads are getting more views than the ones before them, with the two medians and the windows behind it
Audience language	Inferred from a sample of real comments, with the sample size and confidence
Takes sponsorships	Whether recent videos carry brand deals — and separately, affiliate links, memberships and their own products
Deal structures	What share of their recent videos carries each kind, so you can see whether the arrangement you want to offer is one they already use
Comments per 100 views	The median across the videos we read, next to the sample it rests on
Sponsors named	Who has sponsored them, and how long ago — with the last sponsorship dated in its own column
Competitors seen	Any brand you named, and how long ago — with the sentence it appeared in, read across the last ten videos, not just the last three
Estimated price range	Expected views × your CPM. Arithmetic on measured data
Upload cadence and engagement	How often they post, and their like rate
Recent titles	The titles of the uploads every number above was measured on, so you can see what the channel actually makes
Channel description and keywords	What the creator says the channel is about, in their own words, off the channel page — read these before you trust a search term

Why the median, not the average

One viral upload drags a channel's average 20–75% above what a normal video gets. We measured it on a real channel: it averages 28,933 views and its median is 16,500. Price a sponsorship off the average and you overpay by that much.

The range, not just the median

The median says a typical upload gets 137,500 views. It does not say whether the channel delivers that every time or wins a lottery twice a year — and those are different products at the same price. So the row carries where the middle half of recent uploads landed, and the ratio between the two ends.

Measured live on two large channels: one runs a spread of 1.7×, the other 2.1×. The quartiles are taken inside the observations, never extrapolated past them: a "typical high" no video ever reached would be worse than useless.

What goes big, and what kind of video it was

An outlier is an upload that beat the channel's own median by two times. Beating the median alone is half their uploads and tells you nothing.

The row names them. On one homelab channel we tested, the single outlier was a tour of everything the creator runs at home — and reading that one title tells you more about what works there than any label could.

We also read a format off the title where the creator states one: review, tutorial, listicle, vlog, challenge or Q&A, in English, Spanish and French. And we say what share we could not read:

review 58% | challenge 8% | unlabelled 33%

We do not guess at the rest, and on many channels the unlabelled share is the biggest number in the cell. A science explainer titled as a plain question is a real format, and its title does not say so. Inventing a category and then grading a creator against it would tell you about our taxonomy, not about them — so where the titles are silent, you get the outlier titles and your own judgement.

The view trend, and what it is not

Recent uploads against the ones before them: the newest third of the videos we read, against the two thirds before, compared on the median views of each. The cell beside it says exactly what was compared — "3 newest (26-102 days) vs 6 before (122-233 days)" — because the same −18% means something very different over three weeks than over three quarters.

Why videos and not a fixed 30 days. Because a fixed window is unanswerable at both ends. One real channel put one video in the last 30 days; another put nine, and nothing at all in the 60 before. A figure drawn from one video against one video is noise with a decimal point in it.

Videos under a week old are left out. Views arrive in a rush and then taper, so a video posted yesterday is not a small video, it is an unfinished one. On one real channel the newest upload had 13,483 views at a day old next to siblings at 1.7 million.

Read it as a floor, not as a verdict. Every video in the recent window has had less time to gather views than every video in the older one. So a figure above zero is understated — the channel is doing at least that well. A figure below zero is part real decline and part videos that are simply younger, and nothing observable from outside can separate the two. We publish the number, the two medians and the windows. The judgement is yours.

When it is blank, the cell says why. A channel that uploads daily will not have a comparable baseline inside twelve videos, and one that uploads monthly may not have enough videos at all. Raise "How many recent descriptions to read" to 30 and a daily channel becomes measurable.

Deal structures, as a share rather than a count

A brand offering an affiliate arrangement wants to know whether the creator already runs them. So the row says what share of recent videos carries each kind:

affiliate 60% | patronage (footer)

Counting matched links instead would be useless: one real channel's ten descriptions contained 91 affiliate links, because a single rack-build video listed thirty-five parts. Six of the ten videos carried any at all, and six of ten is the answer to the question.

A link that appears in every description is marked (footer), not a percentage. A Patreon in the standing footer is one arrangement the creator has, not one deal per upload, and reporting it as "patronage 100%" would bury the deals that really are per-video. It is still shown — just not counted as something it is not.

Comments per 100 views

Published as a median across the videos we read, because the newest upload distorts it: comments arrive faster than views do in the first hours, and on one real channel a day-old video sat at 0.97% against a channel that otherwise runs near 0.15%.

The industry's rules of thumb are 0.3% and above is healthy, under 0.1% is worth a second look. Those are their numbers and we print them here so you can apply them — we do not. Two large, obviously genuine channels we measured came in between 0.12% and 0.26%, below the "healthy" line, which is a good reason not to let anyone's threshold hand out verdicts on your behalf.

A video with comments switched off produces no figure rather than a zero. A zero would read as exactly the signal those thresholds are hunting for.

Why comments, not titles

YouTube translates video titles into whichever language you ask for. Request English and a Mexican Spanish-language creator's titles come back in English — the titles describe your request, not their audience. Comments are not translated, so that is what we read.

Why the dates matter

Brands are told to check for a direct rival in the last 90 days. "Promoted a competitor" cannot answer that. "Promoted a competitor 243 days ago" can, and it is the difference between a live conflict and a stale one — the same creator, a different decision.

So every sponsor and every competitor mention comes back dated, from the exact publish date on the video's own page. Where that date is missing we fall back to YouTube's relative label and mark the number with a ~, because the label is coarse exactly where it matters: everything between 90 and 119 days old reads "3 months ago". Measured on one real channel, the label understated the true age of six consecutive videos by 5 to 32 days — and two videos both reading "3 months ago" were really 102 and 122 days old.

A ~ figure is a floor, never an estimate: YouTube rounds down, so the video is at least that old and possibly older.

We publish the number of days and say how precisely we know it. The 90-day line is yours to draw.

Two windows, on purpose

Audience language comes from comments, and three videos is plenty — the language a sample points to is stable from about ten comments, so sampling stops as soon as it has enough.

Sponsor history and competitor mentions come from descriptions, and that window is wider by default, because "has this creator promoted my competitor recently" is not a question three videos can answer. On one real channel, reading three descriptions found no competitors and reading twelve found three of them.

Both are adjustable. Raising the description window makes runs slower, since each extra video is one more page fetched.

What it will not do

• It will not tell you who to pick. It returns facts and one piece of arithmetic. The weighting, the shortlist and the budget are yours — you know your market and we do not.
• It will not guess a CPM. Rates vary enormously by niche. Supply your own, or leave them blank and get every other field.
• It will not claim a language it cannot support. When a comment sample is too small or too mixed, the language is left blank with the reason stated, and the full distribution is still there for you to read. Audiences that write their own language in Latin script — romanized Hindi, for example — reliably defeat language detection, and we would rather say nothing than say "English".
• It does not read transcripts. A sponsor mentioned aloud and never linked in the description is invisible to it.
• It will not date a standing footer. A sponsor line that appears on every video says nothing about when the deal happened, so it is named without a date rather than credited with today's.
• It will not score an audience for authenticity. A bot score from one scrape is a guess. We publish the observable ratios and name the thresholds the industry uses; the judgement is yours.
• It will not tell you a channel is dying. The view trend is arithmetic on what we could read, with the windows printed beside it, and it is blank rather than guessed when the uploads do not support a comparison.
• It does not report audience age or gender. Those exist only in a creator's own YouTube Analytics or in paid panel data. Anyone selling them from the outside is modelling, not measuring.
• It does not report what an audience "talks about". We tested it: the top terms in a video's comments are that video's own title read back, plus the creator's first name. It measures the video, not the audience, and we would rather publish nothing than label it as something it is not.
• It does not invent a format for a video whose title does not state one.

Input

Paste channel handles (@name), full channel URLs, or channel IDs (UC...). Repeats in your list are charged once. Optionally add your CPM range and the competitors you want flagged.

Leave the input empty and it runs a small sample so you can see the output shape.

Output

Two dataset views: Shortlist (the decision columns) and Evidence (sample sizes, sponsor names, and the reason any answer was withheld). Export the Shortlist view for a clean CSV — exporting the unfiltered dataset sorts all columns alphabetically.

Every row carries the same columns, including rows for channels that could not be read, so a spreadsheet built on the output never changes shape halfway down.

Each channel is fetched on its own proxy session, and a channel YouTube refuses is read once more on a fresh one; attempts says whether that happened. A channel that fails twice is an error row with the reason, and it is not billed.

Is the channel even about your subject? No number here answers that, and a search term will happily return a channel with "lab" in its name whose recent uploads dose livestock medicine. recent_titles, channel_description and channel_keywords are on every row so you can check in one glance. They are published as found, unjudged — the subject of a channel is your call.

Pricing

Pay per channel analyzed. Channels that could not be read are not charged — our failure is not your bill.