01
Counted, not guessed
How often a word comes up, how the wording splits, which days the thread moved — all of it is arithmetic on comments we actually read.
YOUTUBE COMMENT ANALYZER
What the comments actually say.
Read a public thread as a whole: the mood, the words people keep repeating, and the questions nobody has answered. The numbers come from counting the comments we read. A deeper reading by a model is available as a separate, clearly labelled step.
Public videos only · Signing in is required — the scan runs against your account and reports are saved to your history
04
Nothing analyzed yet
Paste a public video above. You will get the measured report first; the deeper reading is a second step you choose to take.
01BUILT FOR THE NEXT QUESTION
02
Every claim has a link
Each theme and each unanswered question carries the comments behind it, so you can check the source before you act on it.
03
Deeper reading on demand
A model can read a sample of the thread and explain what it means. It is a separate step, clearly labelled, and it never rewrites the numbers.
02READING THE PATTERN
See the shape before you read.
Before you read a thousand comments, look at the proportions. Pick a theme to see the comments that produced it.
Interactive example · sample data, nothing is fetchedPositive 58%Neutral 27%Negative 15%
Battery life2 representative comments
The most repeated subject, and the one most often framed as a question rather than a complaint.
01
Paste a public video
A watch URL, short link or video ID all work.
02
Read the measured part
Signing in is required — the scan runs against your account, so reports are kept in your history.
03
Ask for the deeper reading
Optional. A model reads a sample and explains it, with links back to the comments.
Want the raw comments behind a pattern? Export them with the comment downloader.
Is the sentiment analysis done by AI?+
No, and the page says so wherever it appears. The mood split is produced by matching the wording of each comment against a fixed list of positive and negative words, with negation handled ("not good" counts as negative). It is repeatable and you can see which word decided each label. A model is only used in the separate deeper reading, where it explains what it read rather than producing any number.
How many comments do you actually read?+
Up to 500 top-level comments per analysis — five pages of a hundred. A video can have far more than that, so every report states how many it read out of how many exist. Comments are read in YouTube's relevance order, and replies are not scanned (each reply thread would need its own request).
What does the deeper reading do that the numbers cannot?+
It reads tone and intent. Counting cannot tell that "great, another ad" is a complaint, nor separate a genuine question from a rhetorical one. The model is given the measured numbers as facts it may not alter, plus a sample of 200 comments, and it explains the themes, the disagreements and what the creator could do next. Anything it cites must be a comment from that sample — citations that are not are dropped, and the count of dropped ones is shown.
Why does a question count as unanswered if it has no visible replies?+
Because that is all the public data can tell us. The reply count reported by YouTube is what is visible, so a reply that was deleted or held looks exactly like no reply at all. The list is a shortlist of things worth answering, not proof that nobody did.
Could you test the battery life next?
Battery life is the only thing holding me back.