YouTube Video Details Scraper — Metadata, Engagement Metrics and Searchable Transcripts

Pull full details for any YouTube video, or point it at a channel or playlist and let it expand that into videos for you. Every record carries the real numbers, derived engagement metrics, creator chapters, and a transcript you can actually search — with timestamps and links that jump straight to the moment.

No API key. No quota.

---

What it does

1. Give it whatever link you have

Videos, channels and playlists all work, in every URL form people actually paste:

You paste	It understands
youtube.com/watch?v=…, youtu.be/…, /shorts/…, /live/…, /embed/…	A single video
A bare 11-character ID	A single video
youtube.com/@handle, or just @handle	The channel's recent uploads
youtube.com/channel/UC…, or a bare UC… ID	The channel's recent uploads
youtube.com/playlist?list=PL…, or a bare playlist ID	The playlist's videos

Channels and playlists are expanded through YouTube's own public feed — a documented endpoint that needs no key and no quota. Handles are resolved to a channel ID automatically.

Anything unrecognised is reported back to you by name rather than silently dropped, so a run never quietly comes up short.

2. Search inside the transcript

Give it search terms and every match comes back with a timestamp, the surrounding sentence, and a link that opens the video at that second.

The important detail: phrases are matched across caption boundaries. Captions break wherever the line ran out, so "machine learning" is routinely split into …about machine / learning and…. Matching fragment by fragment misses those. The transcript is joined first and matches are traced back to the fragment that contains them, which is how each hit still gets an accurate timestamp.

Alongside search you get the transcript as readable plain text (caption artefacts like [Music] and >> SPEAKER: removed), a word count, the speaking rate in words per minute, transcript coverage, and the most frequent meaningful words with their share of the transcript.

3. Engagement metrics, honestly derived

Field	What it is
engagementRatePercent	(likes + comments) ÷ views, as a percentage
likeRatePercent	likes ÷ views
commentRatePercent	comments ÷ views
viewsPerDay	views ÷ days since publication
videoAgeDays	days since publication

Sort by any of them, or filter on views, duration, publish date, keyword, or "must have a transcript".

Where the numbers come from, since it matters: the view count and like count are exact figures read from the page. The comment count is only ever published rounded ("2.4M") on videos of any size, so it is named commentCountApprox and the original label is kept in commentCountText — a rounded number is not presented as a precise one. Engagement rate is a conventional proxy computed from public counts; views are not unique viewers.

4. Chapters, tags and format

Creator chapter markers are read out of the description and returned with start time, end time, duration and a deep link. Timestamps that aren't chapters — links to other videos, track listings — are rejected by requiring them to ascend and to fall inside the runtime.

isShortFormat comes from the canonical URL, which is the only reliable signal. (Requesting /shorts/<id> directly proves nothing: it redirects for every video, Short or not.)

maxResolution and maxFps are read from the actual stream list, so a 4K video reports 4K.

---

Input

Field	Type	Default	Description
videoUrls	array	—	Videos, in any URL form or as bare IDs
channelUrls	array	—	Channel links, @handles or UC… IDs
playlistUrls	array	—	Playlist links or IDs
maxVideosPerSource	integer	10	Videos to take from each channel/playlist
includeTranscript	boolean	true	Fetch and analyse captions
transcriptLanguages	array	["en"]	Preferred languages, in order
transcriptSearchTerms	array	—	Terms to locate, with timestamps
includeTranscriptSnippets	boolean	false	Include every timed caption segment
keywordCount	integer	15	Keywords to report per video
includeChapters	boolean	true	Extract chapter markers
includeDescription	boolean	true	Keep the full description
minViewCount	integer	—	Drop videos below this
minDurationSeconds	integer	—	Drop shorter videos (61 excludes Shorts)
maxDurationSeconds	integer	—	Drop longer videos
publishedAfter	string	—	YYYY-MM-DD or ISO timestamp
filterKeywords	array	—	Title/description must contain one
requireTranscript	boolean	false	Only keep videos with captions
sortBy	string	discovery order	Views, likes, date, engagement, …
sortDescending	boolean	true	Highest first
concurrentRequests	integer	5	Videos fetched at once
emitRunSummary	boolean	true	Append the aggregate record
proxyConfig	object	Apify Proxy on	Proxy settings; effectively required on the platform

At least one of videoUrls, channelUrls or playlistUrls is required.

Example input

{
  "channelUrls": ["@RickAstleyYT"],
  "maxVideosPerSource": 10,
  "includeTranscript": true,
  "transcriptSearchTerms": ["never gonna give you up"],
  "minViewCount": 10000,
  "sortBy": "engagementRatePercent",
  "sortDescending": true
}

---

Output

One video record per video, then an optional summary.

Video record fields

Identity

Field	Description
videoId, videoUrl, canonicalUrl	The video's ID, a watch link, and YouTube's own canonical link
videoTitle, description	Title and full description (description omitted if you turn it off)
channelName, channelId, channelHandle, channelUrl	Who published it
subscriberCountText, subscriberCountApprox	Channel size as shown, and parsed to a number
discoveredVia	Whether this video came from a direct link, a channel feed or a playlist feed
scrapedAt	When this row was taken

Timing and format

Field	Description
publishedAt, uploadedAt	Publish and upload timestamps (they differ for scheduled videos)
durationSeconds, durationText	Runtime in seconds and as M:SS / H:MM:SS
isShortFormat	Whether it's a Short, from the canonical URL
isLiveContent, isPrivate, isUnlisted, isFamilySafe, allowRatings	Status flags as YouTube reports them
category, audioLanguage	Category, and the declared audio language when the video sets one
maxResolution, maxWidth, maxHeight, maxQualityLabel, maxFps	Highest quality actually offered
thumbnailUrl	Largest available thumbnail

Counts and engagement

Field	Description
viewCount, likeCount	Exact figures
commentCountApprox, commentCountText	Rounded comment count, and YouTube's original label
ratingAverage, ratingCount	Star rating, when the video arrived via a feed that carried it
engagementRatePercent, likeRatePercent, commentRatePercent	Ratios against views
viewsPerDay, videoAgeDays	Reach per day, and age

Content

Field	Description
tags, hashtags	Creator tags, and hashtags found in the title and description
chapters, chapterCount	Chapter markers with start, end, duration and title
hasCaptions, captionLanguages	Whether captions exist, and in which languages

Transcript (when includeTranscript is on)

Field	Description
transcriptAvailable, transcriptError, transcriptLanguage	Whether a transcript was retrieved, and why not if it wasn't
transcriptText	Cleaned, readable plain text
transcriptWordCount, transcriptCharacterCount	Length
speakingRateWpm, readingTimeMinutes, transcriptCoveragePercent	Pace, reading time, and how much of the runtime the captions span
topKeywords	Most frequent meaningful words with counts and shares
searchHits, searchHitTotal	Matches for your search terms, with timestamps and deep links
transcriptSnippets	Every timed caption segment (only when you ask for it)

Failure

Field	Description
fetchFailed, fetchError	Whether the video could be read, and the reason

Example video record

{
  "recordType": "video",
  "videoId": "dQw4w9WgXcQ",
  "videoUrl": "https://www.youtube.com/watch?v=dQw4w9WgXcQ",
  "videoTitle": "Rick Astley - Never Gonna Give You Up (Official Video) (4K Remaster)",
  "channelName": "Rick Astley",
  "channelId": "UCuAXFkgsw1L7xaCfnd5JJOw",
  "channelHandle": "@RickAstleyYT",
  "subscriberCountText": "4.53M subscribers",
  "subscriberCountApprox": 4530000,
  "publishedAt": "2009-10-24T23:57:33-07:00",
  "durationSeconds": 213,
  "durationText": "3:33",
  "viewCount": 1803235213,
  "likeCount": 19332287,
  "commentCountApprox": 2400000,
  "commentCountText": "2.4M",
  "engagementRatePercent": 1.2052,
  "likeRatePercent": 1.0721,
  "viewsPerDay": 293512,
  "videoAgeDays": 6136.0,
  "category": "Music",
  "isShortFormat": false,
  "isLiveContent": false,
  "isFamilySafe": true,
  "maxResolution": "3840x2160",
  "maxFps": 25,
  "tags": ["rick astley", "Never Gonna Give You Up", "nggyu"],
  "hashtags": [],
  "hasCaptions": true,
  "captionLanguages": ["en", "de-DE", "ja"],
  "transcriptLanguage": "en",
  "transcriptWordCount": 366,
  "speakingRateWpm": 103.1,
  "transcriptCoveragePercent": 99.2,
  "topKeywords": [{ "keyword": "gonna", "occurrences": 24, "sharePercent": 14.55 }],
  "searchHits": [
    {
      "term": "never gonna give you up",
      "matchCount": 6,
      "occurrences": [
        {
          "startSeconds": 43.0,
          "startTime": "0:43",
          "context": "…we're no strangers to love never gonna give you up never gonna let you down…",
          "linkUrl": "https://www.youtube.com/watch?v=dQw4w9WgXcQ&t=43s"
        }
      ]
    }
  ],
  "chapters": [],
  "chapterCount": 0,
  "discoveredVia": "direct link",
  "fetchFailed": false,
  "scrapedAt": "2026-08-12T09:00:00+00:00"
}

A video that cannot be read still produces a row with fetchFailed: true and the reason in fetchError. Where possible the title and channel are filled in from YouTube's public oEmbed endpoint, so a blocked page still tells you which video failed.

Summary record

{
  "recordType": "summary",
  "videosAttempted": 10,
  "videosSucceeded": 10,
  "videosFailed": 0,
  "failedVideoIds": [],
  "sourcesRequested": ["@RickAstleyYT"],
  "inputsRejected": [],
  "totalLikes": 21400418,
  "transcriptsRetrieved": 8,
  "shortsCount": 3,
  "totalViews": 2104882133,
  "averageViews": 210488213,
  "medianViews": 270575.0,
  "averageEngagementRatePercent": 2.41,
  "averageDurationSeconds": 128.4,
  "totalTranscriptWords": 4820,
  "channelBreakdown": { "Rick Astley": 10 },
  "categoryBreakdown": { "Music": 8, "Entertainment": 2 },
  "topTags": [{ "tag": "rick astley", "videoCount": 6 }],
  "mostViewedVideoId": "dQw4w9WgXcQ",
  "mostViewedVideoTitle": "Rick Astley - Never Gonna Give You Up (Official Video) (4K Remaster)",
  "mostEngagingVideoId": "JjI4o2w6D5A",
  "finishedAt": "2026-08-12T09:00:22+00:00"
}

---

Common use cases

• Content research — pull a competitor's recent uploads and sort by engagement rate to see what actually lands, not just what got views.
• Finding the moment — search a long interview or conference talk for a topic and get links straight to every mention.
• Repurposing — the cleaned transcript, chapters and keyword profile are the raw material for blog posts, show notes and clips.
• SEO and tag research — topTags across a channel's catalogue shows what they consistently tag for.
• Monitoring — run on a schedule against a channel and track viewsPerDay as videos age.
• Dataset building — transcripts plus metadata for a set of videos, in one pass.

---

Notes and limits

• The public feed is short. Channel and playlist expansion returns roughly the 15 most recent items and offers no paging, so maxVideosPerSource cannot exceed what the feed holds. To go deeper, supply video URLs directly. The run log says when the feed was the limiting factor.
• Comment counts are rounded by YouTube, not by this Actor. Large videos publish "2.4M" and nothing more precise. commentCountApprox carries the parsed number, commentCountText the original label.
• Transcripts depend on the video having captions. Many do not, and some creators disable them. transcriptAvailable and transcriptError say which case you hit, per video, instead of returning an empty string that looks like a silent video.
• Run this with a proxy. Apify's shared outbound addresses are heavily used against YouTube, and a run without proxyConfig is answered with "Sign in to confirm you're not a bot" for most videos. Measured on the platform: with no proxy, 1 of 4 videos returned data; with Apify Proxy enabled, 5 of 5 returned data and transcripts. Either the default datacenter group or a residential group works; residential holds up better across long runs.
• Videos that fail this way still produce a row. Titles, view counts, publish dates and the Shorts flag come from the public feed when a channel or playlist supplied them, so a challenged run degrades to less data rather than to none.
• Auto-generated captions contain errors. Speaking rate, word counts and keyword frequencies inherit them. Treat them as good estimates, not transcription-grade output.
• speakingRateWpm is words divided by video length, so long musical or silent passages pull it down. That is the honest figure for the video, not for the speech within it.
• Age-restricted, private, members-only and region-locked videos cannot be read without signing in, and are reported as failures with the reason.
• Legacy /c/Name and /user/Name channel URLs no longer resolve for many channels. Use the @handle or the /channel/UC… form.
• View and like counts change constantly. scrapedAt records when each row was taken.
• This Actor reads only public pages and public feeds. It does not sign in, and cannot reach anything that requires an account.

---

Local development

pip install -r requirements.txt
python -m tests.test_core     # 172 checks, including live requests to YouTube
apify run