YouTube Comments Scraper

Scrape public YouTube comments and replies from one or more video URLs, then export the results as JSON, CSV, or a spreadsheet. This YouTube comments scraper collects the conversation text, public commenter identity, displayed engagement, reply relationships, and video context without requiring a video data key.

Use it when you need a repeatable way to capture a discussion around a specific video. Each result is a flat record with a stable comment identifier, so records can be joined, deduplicated, compared across runs, or loaded into a table. The actor follows public comment pagination and optional reply threads until it reaches the limit you set or the public continuation ends.

What data can you extract from YouTube?

Each comment and reply has the same set of fields. A reply has type set to reply and replyToCid set to its parent comment ID. A top-level comment has type set to comment and a null parent ID. The dataset may also contain an error item for an input that could not provide comments; error items are described below.

Field	Description	Example
comment	Public comment text.	But the link said free robux 😢
cid	Stable comment or reply ID supplied by the site.	UgwmOpWybGLAIIhhA854AaABAg
commentUrl	Direct watch-page link to the comment.	https://www.youtube.com/watch?v=dQw4w9WgXcQ&lc=UgwmOpWybGLAIIhhA854AaABAg
author	Public display name.	A public handle
authorChannelId	Public commenter channel ID.	UCdOyiHRpksv1dAMVQWmss3A
authorChannelUrl	Public commenter channel link.	A public channel URL
authorAvatarUrl	Public profile image link.	An image URL
authorIsVerified	Whether a verified badge is shown.	false
videoId	Video ID shared by all results from that video.	dQw4w9WgXcQ
pageUrl	Canonical video watch link.	https://www.youtube.com/watch?v=dQw4w9WgXcQ
commentsCount	Video-wide count, if the public page exposes one.	null
replyCount	Displayed number of replies to this comment, parsed as a number.	1
voteCount	Displayed likes parsed as an approximate number.	320000
voteCountText	Original displayed like count.	320K
publishedTimeText	Relative age as shown on the page.	22 minutes ago
authorIsChannelOwner	Whether the commenter owns the video channel.	false
hasCreatorHeart	Whether the creator heart is shown.	true
isPinned	Whether a top-level comment is marked as pinned.	true
type	comment or reply.	comment
replyToCid	Parent ID for a reply; null for a top-level comment.	null
title	Video title.	The title shown on the watch page

The site currently supplies relative comment age in the public comment response. publishedTimeText preserves that wording; it is not an exact publication timestamp. The site also abbreviates large like and reply counts. voteCountText preserves the displayed value, while voteCount expands abbreviations such as K into an approximate integer. Do not treat 320K as an exact measurement of 320,000.

commentsCount is included for a stable shape alongside other comment exports, but the tested public page did not provide the video-wide count. It is null when unavailable. A null count does not mean the video has zero comments; the dataset can still contain many individual records.

How to scrape YouTube comments

1. Paste one or more public video URLs into Video URLs. A normal watch link is the simplest input. Short links and links to Shorts, live pages, or embedded videos are also accepted.
2. Set Maximum comments per video. The default is 100. This limit counts both top-level comments and replies when replies are enabled.
3. Choose Newest first to monitor recent discussion, or Top comments to collect the featured ranking. Leave Include replies on when conversations beneath comments matter.
4. Run the actor. It reads the watch page, follows comment continuation pages, and saves each result as a dataset item.
5. Download the dataset in your preferred format. Use cid to remove duplicates if you combine multiple runs for the same video.

For a first check, use a video with a visible public comment section and a limit of 40 or 100. A video with disabled comments, a region restriction, an age wall, or an unavailable watch page may produce an error item. Those conditions are tied to the video and the public visitor view, not to the requested limit.

Input

Parameter	Type	Default	Description
startUrls	URL list	Required	Public video URLs. Each item can contain a url property. The actor also accepts a bare 11-character video ID in a list item.
maxComments	Integer	100	Maximum number of comment and reply records per video. Minimum is 1.
sortCommentsBy	Choice	NEWEST_FIRST	NEWEST_FIRST or TOP_COMMENTS for top-level comments.
includeReplies	Boolean	true	Follow public reply threads when they are available.
oldestCommentDate	Text	Empty	Optional UTC date, such as 2026-09-01, or a relative age, such as 7 days. This filter uses displayed relative age and is approximate.
maxItems	Integer	0	Overall dataset cap across all input videos. Zero means there is no overall cap.
proxyConfiguration	Object	Automatic	Optional network configuration for watch pages and comment requests.

A realistic input for recent public discussion is:

{
  "startUrls": [
    { "url": "https://www.youtube.com/watch?v=dQw4w9WgXcQ" }
  ],
  "maxComments": 40,
  "sortCommentsBy": "NEWEST_FIRST",
  "includeReplies": true
}

The per-video and overall limits work together. With three URLs, maxComments set to 100, and maxItems set to 180, the actor can return at most 100 items for any one video and at most 180 items for the whole run. If the first two videos fill the overall cap, the third video is not fetched. If one video exposes only 12 public comments, the actor moves on after those 12; a requested maximum is a ceiling, not a promise of that many records.

The date filter changes the top-level comment order to newest first. A date such as 2026-09-01 is interpreted in UTC. A relative value such as 7 days is calculated from the start time of the run. Because the source says things like 5 days ago instead of returning an exact timestamp, the cutoff is approximate near its boundary. A reply to an older top-level comment may also be absent when its parent was excluded by the cutoff.

Output example

The following values come from a successful 40-item run against a public video. This shortened record omits the profile image URL and several other fields shown in the table above.

{
  "comment": "But the link said free robux 😢",
  "cid": "UgwmOpWybGLAIIhhA854AaABAg",
  "commentUrl": "https://www.youtube.com/watch?v=dQw4w9WgXcQ&lc=UgwmOpWybGLAIIhhA854AaABAg",
  "videoId": "dQw4w9WgXcQ",
  "pageUrl": "https://www.youtube.com/watch?v=dQw4w9WgXcQ",
  "commentsCount": null,
  "replyCount": 1,
  "voteCount": 1,
  "voteCountText": "1",
  "publishedTimeText": "22 minutes ago",
  "authorIsChannelOwner": false,
  "hasCreatorHeart": false,
  "isPinned": false,
  "type": "comment",
  "replyToCid": null
}

Each record can be read independently. pageUrl identifies the video, while commentUrl points to that specific comment. For replies, replyToCid connects the reply to its parent. This makes it possible to build a thread view after export without relying on dataset order. The video title and public commenter fields are repeated on each item so a CSV export remains useful without a separate lookup table.

Use cases

• Community monitoring: Collect recent comments on your own videos to see recurring questions, requests, and points of confusion. Use a modest limit on a regular schedule, then compare comment IDs against earlier runs.
• Campaign response tracking: Capture the public discussion beneath a launch video and separate top-level reactions from replies. The reply relationship helps distinguish a new opinion from a response to an existing conversation.
• Editorial research: Review comments on a specific topic video for language viewers use, examples they mention, and questions left unanswered. Keep the original comment URL when passing a finding to an editor.
• Moderation review: Export public text and engagement to prioritize manual review. The actor does not classify harmful content or make moderation decisions; it supplies the source records for your own workflow.
• Audience analysis: Compare the questions and reactions under several public videos. Use videoId and cid as stable keys, while remembering that featured ordering and visible counts can change between runs.
• Creator interaction review: Identify comments marked as pinned, posted by the channel owner, or hearted by the creator. These flags describe the public page state at collection time.

For longitudinal work, keep the run date with your exported file. Relative ages such as 2 days ago will read differently in a later run even for the same comment. A stable comment ID lets you identify the same item over time, but the visible like count and creator signals may change.

How much does it cost to scrape YouTube comments?

The configured result charge is $1 per 1,000 dataset items, or $0.001 per delivered item. A normal comment, a reply, and an error item each occupy one dataset item. The actual bill can depend on the account and run settings; check the run estimate before collecting a large dataset.

Delivered items	Result charge
100	$0.10
1,000	$1.00
5,000	$5.00

For example, ten videos returning 100 items each produce up to 1,000 records and a $1.00 result charge. Five videos returning only 20 public items each produce 100 records and a $0.10 result charge, even if the requested per-video maximum was higher. An overall maxItems cap is useful when you need a hard ceiling on record volume. It stops data collection as soon as the dataset reaches that cap.

Reply-heavy videos can require more requests for the same number of output records. In a verified 40-item run, the actor completed successfully with all required fields filled. Run time and infrastructure usage vary with video availability, public page behavior, reply depth, and network conditions. Do not multiply the short verification run time into a guarantee for thousands of comments.

Tips for best results

Choose the sort order for your question. Newest first is suitable when you want current reactions; Top comments follows the site's featured ranking and can surface older high-engagement entries. Sort order applies to top-level comments. Replies are collected from the reply threads attached to those comments, so a reply is not globally sorted against every top-level entry.

Leave replies enabled when the discussion itself matters. The actor gathers a spread of top-level comments before traversing reply threads, so a single popular pinned comment does not consume the entire small output limit. If you only need independent top-level opinions, turn replies off. That generally reduces requests and gives your limit entirely to top-level comments.

Start with a small maximum to verify the video and fields, then increase it. Popular videos may expose long chains of continuations. The actor stops at the requested per-video limit, the overall cap, or the end of public continuations. It also tracks comment IDs while processing a video so repeated entries from pagination are not added twice.

Use more than one URL when comparing videos, but remember that the overall cap is processed in input order. Put the highest-priority videos first if maxItems may stop the run early. If a particular video has restricted or disabled comments, other input videos can still return normal records. Inspect error items before treating a low result count as a sign of low audience activity.

Keep both voteCount and voteCountText when analyzing engagement. The text field records what a visitor saw; the numeric field is convenient for sorting and charting. Large values are rounded by the source display, so use them as approximations. The same caution applies to replyCount where the display uses a compact value.

Limits

Only public comments visible to an anonymous visitor can be collected. Private, deleted, hidden, disabled, age-gated, or region-restricted discussions may not appear. Moderation and ranking can also change what a visitor sees. The actor does not sign in, reveal private details, or bypass permissions attached to an account.

The source currently provides relative publication age rather than an exact UTC timestamp for individual comments. oldestCommentDate is a convenience filter based on that visible age, not an exact historical query. For strict time-series boundaries, retain the run time and apply your own review to comments near the cutoff.

The tested watch page did not disclose a video-wide total comment count, so commentsCount is null. A requested maximum may exceed the number of public items reachable through the site's continuation chain. Reply counts and like counts can be abbreviated in the interface. A value such as 1K is expanded approximately in the numeric output.

An unavailable source produces an item with input, url, error, and note. These fields let downstream processing separate failures from normal comments. Check error before interpreting an item as comment data. A failed video does not erase records already collected from other URLs in the same run.

The comment endpoint and response shape are controlled by the site and can change. A layout or response change can temporarily affect extraction, pagination, or a field's availability. The actor uses the current watch-page configuration rather than a fixed client version or hard-coded continuation token, which helps it follow routine changes but cannot guarantee permanent compatibility.

Structured requests

The input names are intended to make scheduled and programmatic runs straightforward. A request such as “collect the newest 100 comments and replies from these two video URLs” maps to startUrls, maxComments, sortCommentsBy, and includeReplies. A request such as “capture up to 500 top-level comments across five videos” maps to startUrls, includeReplies: false, and maxItems: 500.

When passing the output into another system, key each record by cid and group it by videoId. Keep type and replyToCid so replies do not become disconnected from their parent. Check for error items before counting comments, and treat null fields as missing public values rather than zero. These conventions work for both one-off exports and repeated monitoring.

FAQ

Can I scrape comments from more than one video?

Yes. Add multiple entries to startUrls. The actor processes them in order and applies maxComments separately to each video. maxItems limits the entire run. Duplicate video IDs in the input are skipped during a run.

Are replies included?

Yes, by default. Set includeReplies to false when you want only top-level comments. Every included reply has type: "reply" and a parent ID in replyToCid. The per-video comment limit includes both kinds of item.

Can I get the exact date a comment was posted?

The tested public response exposes relative wording such as 22 minutes ago or 1 year ago. The actor preserves that wording in publishedTimeText. It does not invent an exact timestamp from an imprecise relative label.

Why is the video-wide comment count null?

The current public watch page and comment response did not expose a reliable total for the tested video. The commentsCount key stays in the output for a stable schema and is null when no total is supplied. The number of dataset items is the number actually collected, not the total number on the video.

What happens if comments are disabled or a video is unavailable?

The actor emits an error item for that input when it cannot retrieve public comments. The item includes the input URL and a short explanation. Other valid URLs in the same input can still produce comments.

Do I need a proxy?

The default network configuration is set up for public watch-page and continuation requests. The successful verification runs returned comments without a challenge. Availability can differ by network, country, and time, so a blocked request may require a different route or may remain inaccessible to an anonymous visitor.

Is it legal to collect comments?

Use public data for a legitimate purpose. Respect the site's terms, applicable privacy rules, and any obligations around storing or sharing people's comments and profile details. Avoid collecting or using personal data in ways the commenters would not reasonably expect.

Integrations

The dataset can be downloaded directly or read programmatically. Schedule repeated runs to monitor a video, send a completion notification, or load the flat records into a spreadsheet or database. Use cid for deduplication and videoId for grouping when combining exports across runs. A webhook can signal that fresh comments or an error item are ready for processing.

Support

Open an issue on the actor's Issues tab with the video URL, the input used, and the observed error or missing field. Include a run link when available so the failing public response can be checked. Requests for additional public fields are welcome.