YouTube Comments Scraper: Comments, Questions & Sentiment

Scrape YouTube comments from any video or Short, with no API key and no login, and get a report for every video that tells you what the comments actually say:

• Every comment: text, likes, reply count, date, link, and whether it is pinned, hearted or written by the creator
• Replies under each comment (optional)
• Top or newest first: what YouTube shows first, or what people are saying right now
• Video report: top comments, the questions viewers ask, the complaints, the moments people mention ("1:23:24 ..."), keywords and phrases, sentiment, emoji and how often the creator hearts or replies
• Video details: title, channel, views, likes, publish date and comments per 1,000 views

What does this YouTube comments scraper do?

Learn Python - Full Course for Beginners · freeCodeCamp.org
 49.4M views · 46,100 comments · 0.93 comments per 1,000 views
 500 newest comments: 35.6% positive, 3.2% negative
 Most-mentioned moment: 1:23:20 (30 comments)
 Top question: "Want more from Mike? He's starting a coding RPG/Bootcamp…" (4,700 likes, pinned)
 Top phrases: learn python · python course · hello world · try except

1. Paste video links (watch pages, Shorts, youtu.be links or plain video ids).
2. Choose how many comments per video, top or newest, and whether to include replies.
3. You get every comment in the dataset and one report per video, plus a CSV of all reports.

flowchart LR
    A[Video links,<br/>Shorts, ids] --> B[YouTube's own<br/>comment feed]
    B --> C[Comments<br/>+ replies]
    C --> D[Dataset<br/>one row per comment]
    C --> E[Video report<br/>questions, complaints,<br/>moments, keywords,<br/>sentiment]

Who is it for?

• YouTubers and channel managers: what your viewers ask, what they complain about, which moments they talk about, and how often you reply
• Marketers and brands: what people say about a product launch, an ad, a review or a competitor's video
• Product and market researchers: real questions and objections, in the audience's own words, from thousands of comments
• Agencies and analysts: comment sentiment and engagement across a list of videos, as a table
• AI and data teams: clean comment text for training, topic modelling or sentiment models

What data do you get?

Comments (dataset, one row per comment or reply)

Field	What it tells you
videoId, videoTitle	Which video
commentId, parentCommentId, isReply	The comment, and for replies the comment it answers
text	The comment
likes, replies	Engagement
publishedTimeText, approxPublishedAt, edited	When it was posted ("3 days ago" and the date it works out to)
isPinned, isHearted, isByCreator	Pinned by the creator, hearted by the creator, written by the creator
sentiment, isQuestion, timestamps	Positive, negative or neutral; a question; the video times it mentions
url	Direct link to the comment

Video report (REPORTS and REPORT_<videoId>, one per video)

Field	What it tells you
title, channel, channelUrl, channelSubscribers, views, likes, publishedDate	The video
totalComments, commentsPer1kViews, likesPer1kViews	How much people engage
commentsRead, repliesRead, medianCommentLikes, top10LikeSharePercent	What the report is based on, and how concentrated the likes are
sentiment	Positive, negative and neutral share
creator	Pinned comment, hearted comments, creator replies
commentAge	Comments from the last 24 hours, 7 days, 30 days, year, older
topComments	The 10 most-liked comments
topQuestions	The 10 most-liked questions
topNegative	The 5 most-liked negative comments
topMoments	Video times mentioned by 2+ comments, with a link to that moment
topKeywords, topPhrases, topEmoji	What people talk about

REPORTS_CSV has one flat row per video, ready for a spreadsheet. RUN_SUMMARY lists any problems and the limits of the data.

Example: 2 October 2026

Two runs: 6 videos at 300 top comments each, and 5 videos at 500 newest comments each.

Video	Views	Comments	Per 1,000 views	Positive	Negative
Learn Python - Full Course (freeCodeCamp)	49.4M	46,100	0.93	35.6%	3.2%
But what is a neural network? (3Blue1Brown)	24.6M	9,350	0.38	49.4%	4.0%
Steve Jobs' 2005 Stanford Commencement Address	49.2M	29,194	0.59	42.8%	2.4%
Inside the Mind of a Master Procrastinator (TED)	62.0M	79,654	1.29	29.6%	7.6%
Let's build GPT: from scratch (Andrej Karpathy)	8.0M	3,306	0.41	41.2%	2.6%

What the reports surfaced:

• Moments: 30 of the 500 newest comments on the Python course point at the same 10 seconds (1:23:20). On Karpathy's GPT video the most-mentioned moments are 57:00 (one viewer flags a possible slip) and 1:20:00, the self-attention formula.
• Phrases: "panic monster" and "instant gratification monkey" for the TED talk; "stay hungry, stay foolish" for Steve Jobs; "self attention", "head size" and "block size" for Karpathy.
• Top comments (top sort): "Inside the mind of master procrastinator. Me: adds to 'watch later'" with 74,000 likes; a 73-year-old learning Python with 35,000.
• Activity: the neural network video still gets about 11 comments a day, eight years after it was posted.

Pricing

$0.0005 per comment saved ($0.50 per 1,000) and $0.02 per video report.

• 1 video, 200 comments: $0.12
• 5 videos, 500 comments each: $1.35
• Reports only (turn off "Save comments"): $0.02 per video, still based on every comment read

Good to know

• No names: the Actor does not collect commenter names, handles or channels, and replaces @handles inside comment text with @user. You get what was said, not who said it.
• Dates are approximate: YouTube shows "3 days ago" or "1 year ago", so approxPublishedAt is only as precise as that.
• Like counts are rounded by YouTube above 1,000 ("1.2K"). Video likes are exact when YouTube sends them, otherwise rounded (likesRounded: true).
• Top vs newest: top comments skew old and popular; newest shows what people say now. Pinned comments come first in both.
• Sentiment is a simple rule (English words and emoji). It works for the share of positive and negative comments across hundreds of comments, not for judging a single comment, and it misses sarcasm.
• Large runs keep going: YouTube blocks an address after a few hundred pages. The Actor then switches to Apify Proxy on its own (datacenter first, residential only if needed), at no extra cost to you. A test run read 13,289 comments and replies from 10 videos this way. If every address is blocked, the run stops cleanly, keeps what it read and says so in RUN_SUMMARY.
• Comments turned off (for example videos made for kids), private and removed videos are listed in RUN_SUMMARY, and nothing is charged for them.
• Data comes from YouTube's public web comment feed, the same one youtube.com loads under every video.

Input

Field	Meaning
Videos	Video links, Shorts links, youtu.be links or ids
Comments per video	1 to 20,000 (default 200)
Sort	Top comments or newest first
Include replies	Off by default
Replies per comment	1 to 500 (default 20)
Save comments	On by default; off = reports only

Integrations

Export to JSON, CSV, Excel or Google Sheets, or connect through the Apify API, webhooks, Make, Zapier and n8n. Schedule it with "newest first" to follow the comments on your latest videos every day.