YouTube Search Scraper

Fast, no-browser YouTube search scraper that turns any free-text query into a flat dataset of video, channel, and playlist results. Direct InnerTube API reads. PPE pricing.

What you get

For every search query you pass in, the YouTube Search Scraper returns one row per matching video with full metadata. When you allow channels and playlists in the result mix, those also flow into the same dataset, tagged with a clean type column so downstream filtering is one keystroke.

Per-video fields include the exact integer view count (not just "6.8M"), an approximate ISO date for the upload (computed from YouTube's "2 years ago" style string), exact duration in seconds, channel name plus channel ID plus channel handle plus verified flag, all thumbnail resolutions, the animated hover preview, the description snippet shown on the results page, and every video-level badge (4K, CC, Live, New, HDR, etc.). Live streams are flagged with an isLive boolean.

When to use this actor

• Content gap analysis for an Airbnb host, brand, or creator: pull every video for a topic, slice by recency, find under-served angles.
• Trend monitoring: dump 1000 results sorted by date for a query, run it daily, diff the result set.
• Competitor video discovery: scope to a single channel ID and search across its uploads.
• Ad campaign research: filter to HD only, sort by views, study the top performers.
• AI training data: bulk extract title, description, view counts, and metadata across thousands of queries.

Inputs

searchQueries is the only required input. Everything else is optional.

• searchQueries (array of strings): one or more queries. YouTube search operators work inside the query string, for example "iphone 15", before:2024, -plus, channel:mkbhd.
• maxResultsPerQuery (integer, default 100, max 1000): hard cap on rows per query. YouTube itself stops issuing pagination around 600 to 1000 results; values above are silently clipped.
• sortBy (relevance / date / views / rating): result order.
• uploadDate (any / last_hour / today / week / month / year): restrict to recent videos. Setting this forces type=video.
• duration (any / short / medium / long): under 4 min, 4 to 20 min, over 20 min. Setting this forces type=video.
• type (any / video / channel / playlist / movie): restrict result type.
• features (array): any of live, hd, 4k, subtitles, cc_license, 3d, hdr, 360, vr180, location, purchased. Setting any forces type=video.
• channelId: optional channel scoping. Accepts UC..., a /channel/UC... URL, a /@handle URL, or just an @handle. Free-text filters above are ignored when a channel is set.
• proxyConfiguration: Apify proxy options. Datacenter is sufficient for routine volume; switch to Residential at scale.

Output

One JSON row per result. The dataset has a built-in table view (Overview) with the most useful columns. Every row contains:

• type (video / channel / playlist)
• query, resultRank
• videoId, url, title, description
• channelTitle, channelId, channelHandle, channelUrl, channelIsVerified
• viewCount (exact int), viewCountText, shortViewCountText
• publishedTimeText (relative), publishedDateApprox (ISO date)
• lengthText, lengthSeconds
• isLive, badges
• thumbnails, richThumbnails
• For channels: subscriberCount, subscriberCountText, videoCount
• For playlists: playlistId, videoCount, videoCountText

Pricing

Flat $0.002 per result. No actor-start fee. A 1000-result run costs $2.00. The default 100-result run costs $0.20.

This actor undercuts the popular alternatives by 30 to 60 percent for the same query, because the InnerTube API is fast (under 1 second per page of 20 results) and the runtime never spins up a browser.

How it works

The YouTube Search Scraper bootstraps a session by hitting youtube.com once to pin the current InnerTube client version and visitor data, then issues POST requests to /youtubei/v1/search for each query. TLS impersonation (Chrome 131 fingerprint via curl_cffi) is used so the requests look identical to real browser traffic. Pagination is via opaque continuation tokens, capped at 60 pages or your maxResultsPerQuery, whichever is smaller. No browser is launched. No Selenium. No Puppeteer. No CAPTCHAs.

Tips

• For the largest result sets (around 1000 per query), expect 30 to 60 seconds per query end-to-end with Apify datacenter proxy.
• Pass many narrow queries instead of one broad one if you want more than 1000 results on a topic. Queries are scraped independently and de-duplication on videoId is up to you.
• Combine sortBy=date with uploadDate=week for a fresh-uploads digest of any topic.
• Combine sortBy=views with duration=long to find long-form best-performers.

Limits and notes

• YouTube's pagination ceiling is around 600 to 1000 results per query, regardless of how many estimated matches the response advertises. This is a hard wall on YouTube's side, not in this actor.
• publishedDateApprox is computed from the relative time ("2 years ago") and the run timestamp. Day-level precision at best. For the exact upload date, follow up with the per-video player API (out of scope here).
• Live streams and premieres legitimately lack viewCount, lengthText, and publishedTimeText. The actor leaves those fields null.
• Channel scoping uses the channel's "Search" tab, which is search-within-uploads only. Free-text filters above (uploadDate, duration, features, sortBy) are ignored when channelId is set.

API and integrations

Every Apify actor exposes a REST API for triggering runs and pulling datasets from any language. Python, Node.js, Go, curl: any client that can speak HTTP can drive the YouTube Search Scraper, schedule it on a cron, and stream the dataset into a database or analytics stack. The dataset can be exported as JSON or CSV directly from the run's Storage tab.