YouTube Comments Scraper – Comments, Replies, Likes & Authors

YouTube Comments Scraper extracts public comments and replies from one video or a batch of videos without a user-supplied API key. It follows the public YouTube web client's HTTP/InnerTube continuation flow and writes clean, exportable rows to an Apify Dataset.

Use it to scrape YouTube comments, turn public YouTube comment data into a structured YouTube comment export, inspect public author signals, compare TOP with NEWEST results, and collect replies linked to their parent comments. This is a focused YouTube comment extractor and a practical HTTP-based YouTube comments API alternative for public videos.

What it delivers

Each valid Dataset row contains:

• video ID, title, channel name, and direct comment URL;
• comment ID and text;
• author name, public channel ID, and thumbnail when available;
• observed likes and reply count;
• YouTube's published-time label;
• pinned, hearted, creator, and verified signals when exposed;
• parent_comment_id and reply_level for YouTube replies;
• selected TOP or NEWEST sort and row warnings.

Unavailable optional values are null. The Actor does not invent an exact publication date when YouTube only supplies relative text such as 3 hours ago.

Input

Provide video_urls, video_ids, or both. The Actor validates in Python that at least one is present and deduplicates all video IDs while preserving their first-seen order.

{
  "video_urls": ["https://www.youtube.com/watch?v=dQw4w9WgXcQ"],
  "max_comments": 25,
  "sort": "TOP",
  "include_replies": true,
  "max_replies_per_comment": 5,
  "max_videos": 10
}

max_comments is the maximum total rows per video, including replies. Supported URL shapes include watch, youtu.be, Shorts, Live, and embed links. One run accepts at most 25 unique videos and 2,000 results per video.

Outputs

• Dataset — one record per valid comment or reply, ready for JSON, CSV, Excel, XML, and other Apify exports.
• OUTPUT — run status, requested/completed videos, counts, deterministic summary metrics, warnings, and storage references.
• REPORT.html — responsive report with videos, channels, comment table, likes, replies, public creator/pinned/hearted signals, most-liked comments, warnings, and methodology.

Summary metrics include delivered top-level comments and replies, public signal counts, comments with replies, observed likes, most-liked delivered comments, and reply share. No sentiment analysis, demographic inference, toxicity classification, topic extraction, authenticity score, or invented engagement rate is performed.

Acquisition and reliability

The Actor uses only public HTTP requests:

1. load the public watch page;
2. read the public web client's InnerTube configuration and video metadata;
3. open the comments continuation;
4. select YouTube's TOP or NEWEST continuation;
5. paginate comments and optionally replies;
6. deduplicate by video_id + comment_id.

Requests use explicit 15-second timeouts and two short retries for transient failures. A failed video is isolated so other videos in the batch can still finish. There is no browser or Playwright fallback, login, user-supplied API key, paid API, proxy, or external paid service.

Private, removed, age-restricted, regional, or otherwise unavailable videos may return no results. Comments-disabled videos return a clean empty result with a warning. YouTube's TOP order is dynamic, abbreviated like counts are normalized, and InnerTube is an unofficial public web interface that may change.

Billing

The prepared PPE event is comment-result. It is emitted only after a valid comment or reply is delivered to the Dataset. Empty content, duplicates, errors, timeouts, unavailable videos, comments-disabled videos, metadata, OUTPUT, and REPORT are not billable. The event price remains unset until a real Cloud cost is measured.

Free-user friendly sample

The default evergreen sample uses one public video, 25 total results, TOP order, and at most five replies per parent. It requires no secret or external setup and is designed to finish comfortably inside five minutes.

Responsible use

Scrape only public information and follow applicable laws, platform terms, and privacy requirements. Public availability does not remove your responsibility to use personal data lawfully.