How Do You Analyze YouTube Comments Without the API? Top Comments, Questions and Common Words
Read a video's first comments in YouTube's order and count likes, questions and repeated words, with no Data API quota. What that sample is, and is not.
7 min read · 19 Oct 2026
You can read a public video's comments, and count what is in them, without a Google Cloud project, an API key or a quota. What you cannot do is read all of them at once, and the part people get wrong is what the first batch represents.
This post covers what a video's first comments tell you, what they do not, and which counts are honest to compute. The numbers come from a real public video read on 2026-10-19.
You can follow along with the YouTube comment analyzer: paste a video link and it does everything below. It needs no sign-up.
What does the first page of comments actually contain?
Twenty top-level comments, in the order YouTube chose to rank them. Three things follow from that sentence.
It is not the newest comments. YouTube's default view is its own ranking, so the first page leans toward comments that drew likes and replies. The tool does not offer a newest-first view.
It is not all of them. There is no total in the response. A video's headline comment count is on the video page itself, which the YouTube video analyzer shows, but the comment pages do not carry it.
It is not in like order either. We read three pages of Rick Astley's Never Gonna Give You Up (60 comments). The best of the first page had 568,000 likes. The most-liked comment of the 60, at 1,000,000, was on page three. Anything you compute from the first page alone describes those twenty comments and no more.
That is why the tool states its sample on the result: based on the N comments YouTube returned first, in YouTube's order, not every comment on the video. A percentage with no sample size next to it reads as a fact about the video. It is a fact about the sample.
What can you count honestly?
Only what is sitting in the comments you read. No sentiment score, no "audience mood": a model's guess presented as a measurement is the thing to avoid. These are plain counts:
| Tally | What it is | How to read it |
|---|---|---|
| Comments read | How many came back | The denominator for everything else |
| Likes on them | Sum of the like counts YouTube published | Dominated by a few comments; see below |
| Share with a like | Comments with at least one like, out of those that published a like count | Says how broad the approval is |
| Most liked, most replied to | Top five of each, ties in YouTube's order | The comments worth reading first |
| Pinned, from the creator | Comments the creator pinned or wrote | Usually zero; a quick check for engagement |
| Questions | Comments ending in a question mark | A rough list of what viewers want answered |
| Words used more than once | Lower-cased, three letters or more, common words dropped, with counts | A cue for themes, not a topic model |
For the 60 Rick Astley comments: 6,604,267 likes in total, every one of the 60 with a like count, 2 questions, no pinned comment and none from the creator. The most-replied comment had 963 replies. The words that came up most were rick and link (12 each), then song and rickrolled (9 each).
Look at what the sample size did to the words. On the first page alone, rick appeared 7 times and link 5. On the 60, rick appeared 12 times and link 12. A ranking built on twenty comments moves when you add forty.
Two details keep the counts honest. A comment that published no like count is left out of the like totals rather than counted as zero, because "not published" and "zero" are different facts. And the question count is a punctuation rule: it misses a question with no question mark and counts a rhetorical one.
How does this compare with the YouTube Data API?
The official route is commentThreads.list, which returns up to 100 top-level comments a page and costs one quota unit a page against the default 10,000 units a day, as the quota table in YouTube without an API key lays out. For a handful of videos that is plenty, and if you need every comment on a video it is the right tool.
The reasons to skip it are setup and shape, not capability: a project, a key and a consent screen before the first call, versus pasting a link. The no-key route returns twenty at a time and follows a cursor for the next page, so a deep read is more requests. The comment analyzer stops at three pages, which is about 60 comments and three credits on our key. That is deliberate: every page is a real fetch.
What does not work
- Comments turned off. A video with comments disabled and a video nobody has commented on look the same from outside. We tried Baby Shark Dance, a video aimed at children, and the answer was: This video has no comments, or comments are turned off. The tool shows that wording rather than an empty table.
- Few comments. One video we read returned ten comments on a single page and no cursor, so there was nothing more to fetch. None of the ten carried a like count, so the like total and the share with a like showed a dash and a note, not zeros.
- No total count. See above: the comment pages do not say how many exist.
- Replies. Every comment shows how many replies it has, but the replies themselves are a separate request per comment. The tool reads top-level comments only.
- Dates. The comment timestamp is often missing, so there is no "comments per day" view.
Where do the rest of the YouTube tools fit?
Comments are one angle on a video. The others each answer a different question:
- The YouTube video analyzer gives views, likes, comment count and engagement rate, with the formula shown.
- The YouTube transcript extractor gives what was said, as text or subtitles. For a whole channel, the bulk transcript tool does the same across many videos.
- The YouTube channel viewer gives the channel's subscribers, links and recent uploads.
Browse the full set at /tools. The same questions come up on other platforms: how to check TikTok engagement rate without logging in and what you can see on Instagram, LinkedIn and Snapchat without an account cover them.
Doing it in code
The endpoint behind the tool is GET /v1/youtube/video/comments with a video URL. It returns twenty comments, a cursor in pagination.cursor and a hasMore flag. Pass the cursor back to read the next page, and stop when hasMore is false. Each page is one credit; a cached page is free, and a request that returns nothing is not charged.
If you want the tallies yourself, they are a few lines over the items array: sum likeCount while skipping nulls, count isPinned and isAuthorReply, and test text for a trailing question mark. The one rule worth copying from the tool is to carry the sample size next to every figure.