🔎 YouTube Search Scraper — every filter, every field in one request

Scrape YouTube search results by keyword and get back the whole row YouTube's own search response contains: video ID, title, description snippet, channel name and channel ID and @handle, view count, duration, thumbnail ladder, badges, and per-video chapters — plus the complete filter surface (4 sort orders × 6 upload windows × 6 result types × 4 durations × 11 feature flags).

No API key. No login. No cookies. No browser. One HTTP request per page of results.

---

🚀 What does YouTube Search Scraper do?

You give it keywords. It runs them through YouTube's own internal search endpoint, walks the continuation chain, de-duplicates, and writes one row per result to your dataset.

• 🎬 Videos — the full result row, in YouTube's own ranking order (searchPosition)
• 📺 Channels — resultType: channel returns channel rows with ID, @handle, subscribers, bio
• 📚 Playlists — ID, owner, video count, preview titles, first video
• 🍿 Movies — genre, release year, cast, director, runtime
• ⚡ Shorts — including the uploader's channel ID, which YouTube does not put in a Shorts row at all (recovered from the reel endpoint — see The wedges, below)
• 🔴 Live streams — with the live viewer count, correctly separated from view count
• 🧭 Chapters — title, timestamp and startSeconds, straight off the search response, without a second request per video

📊 Every field on a video row

Field	What it is	Fill on the 600-row proof run
type	video / short / channel / playlist / movie	100.0%
videoId	YouTube's 11-char ID	100.0%
url	https://www.youtube.com/watch?v=…	100.0%
title	Video title	100.0%
titleAccessibilityLabel	Spoken title incl. duration	100.0%
descriptionSnippet	The search snippet, query terms included	99.7%
descriptionSnippetHoverText	What the snippet was taken from	99.7%
channelName	Uploader name	100.0%
channelId	UC… channel ID	100.0%
channelHandle	@handle	99.8%
channelUrl	Canonical channel URL	100.0%
channelVerified	Verified badge present	100.0% (true on 273)
channelBadges	Verified / Official Artist Channel	45.7%
channelAvatarUrl	Largest avatar	100.0%
channelAvatarThumbnails	Full avatar ladder {url,width,height}	100.0%
collaborators	Every co-uploader of a collab video, with ID + handle + subscribers	0.3%
viewCountText	"3,315,126 views"	100.0%
viewCount	Integer 3315126 ("No views" → 0)	99.8%
shortViewCountText	"3.3M views"	100.0%
liveViewerCountText	"15 watching" — live rows only	0.2%
liveViewerCount	Integer live viewers	0.2%
publishedTimeText	"6 years ago" / "Streamed 1 day ago"	99.8%
publishedTimeApprox	ISO date derived from that — approximate, see limits	99.8%
durationText	"18:37"	99.8%
durationSeconds	Integer 1117	99.8%
durationAccessibilityLabel	"18 minutes, 37 seconds"	99.8%
isLive / isShort / isUpcoming	Row-kind flags	100.0%
scheduledStartTime	ISO start of an upcoming premiere	0.0% on this run
badges	New / CC / 4K / LIVE labels	48.7%
hasCaptions	CC badge present	100.0% (true on 76)
is4K	4K badge present	100.0% (true on 244)
thumbnails	Full ladder, 360×202 + 720×404	100.0%
thumbnailUrl	Largest thumbnail	100.0%
movingThumbnailUrl	Animated hover preview (.webp)	77.5%
chapters	[{title, time, startSeconds}]	31.7% (190 rows, 1,811 chapters)
chaptersCount	Integer	100.0%
playerParams	YouTube's player params for the row	100.0%
searchVideoResultEntityKey	Stable per-result tracking key	100.0%
inlinePlaybackUrl	Inline-preview watch URL	97.0%
searchQuery	Which of your keywords produced this row	100.0%
searchPage	Continuation page index	100.0%
searchPosition	Rank inside the result set — YouTube's own ordering	100.0%
estimatedResults	YouTube's estimate for the whole query	100.0%
scrapedAt	ISO timestamp of the scrape	100.0%

46 fields on a video row; 28 of them at 100.0% fill across the 600-row proof run.

Other row types (measured)

Type	Rows measured	Fields	Notable
channel	80	23	channelId 100%, @handle 100%, subscriberCount 98%, descriptionSnippet 90%
playlist	40	21	playlistId, owner ID + handle, videoCount, previewVideoTitles, firstVideoId — all 100%
movie	19	32	releaseYear 100%, genre 100%, cast 100%, director 89%
short	51	46	videoId 100%, viewCount 100%, channelId 100%

---

🎛️ Every filter YouTube has — not a hand-picked subset

Most YouTube search scrapers ship a fixed lookup table of pre-baked filter strings, so you get whatever combinations somebody typed out by hand. This actor encodes YouTube's search protobuf properly, so the entire cross-product is reachable:

Filter	Values
sortBy	relevance · uploadDate · viewCount · rating
uploadDate	any · lastHour · today · thisWeek · thisMonth · thisYear
resultType	video (default) · any · shorts · channel · playlist · movie
duration	any · under3min · from3to20min · over20min
features	live · fourK · hd · subtitles · creativeCommons · threeSixty · vr180 · threeD · hdr · location · purchased

And it checks itself. Every search response carries YouTube's own filter chips with their official encodings. On the first page of every run the actor decodes those chips and compares them to its own encoder. If YouTube ever renumbers an enum, the actor adopts YouTube's value, logs the drift, and re-runs the query — instead of silently handing you unfiltered results.

Measured on the proof runs: 24 of 24 encodings byte-identical to YouTube's own chips (type=video → EgIQAQ==, duration>20m → EgIYAg==, VR180 → EgPQAQE=, …), and the actor logged encoder confirmed against 23 live filter chips on every platform run.

---

🧩 The wedges — what this does that the search response does not hand you

1. Shorts get a channel. YouTube's shortsLockupViewModel has a title, a view count and a thumbnail, and nothing else — no channel name, no ID, no handle. The uploader's UC… ID is embedded in the reel endpoint's params protobuf, so it is recovered there. Verified against the videos' own watch pages: 4 of 4 exact matches, then 51 of 51 Shorts rows on the platform.
2. Collaboration videos keep their channel. When two creators co-upload, YouTube replaces the byline link with a "Collaborators" dialog — a parser that only reads the byline loses channelId and channelHandle on those rows. Here they are read from the dialog, and every collaborator ships in collaborators[] with ID, handle and subscriber count.
3. Chapters without a second request. 1,811 chapters across 190 of 600 rows, from the same response — no extra fetch, no extra latency, no extra charge.
4. "No views" is 0, not null. A brand-new upload has a real, meaningful zero.
5. "15 watching" is never mis-read as a view count. It lands in liveViewerCount.

---

📥 Input

{
  "searchQueries": ["plumber near me", "best crm software"],
  "resultType": "video",
  "sortBy": "relevance",
  "uploadDate": "any",
  "duration": "any",
  "features": [],
  "maxResultsPerQuery": 120,
  "maxResults": 0,
  "language": "en",
  "region": "US",
  "deduplicateAcrossQueries": false,
  "proxyConfiguration": { "useApifyProxy": true }
}

Input	Default	Notes
searchQueries	["web scraping"]	One per line. Volume comes from many queries, not one deep one — see the ceiling below.
resultType	video	video returns only videos. any reproduces youtube.com, Shorts shelf included — you pay for those rows too.
sortBy / uploadDate / duration / features	unfiltered	Server-side, combinable.
maxResultsPerQuery	200	Per-query stop.
maxResults	0 (unlimited)	Overall stop across every query.
language / region	en / US	hl / gl. YouTube ranks per country — this is how you scrape the German result set for the same keyword.
deduplicateAcrossQueries	false	Duplicates inside a query are always dropped before billing. This also drops a result that already appeared under a different keyword.
proxyConfiguration	Apify Proxy auto	Verified HTTP 200 through auto, through residential, and from a home IP.

---

💵 Pricing

Pay per event — one event, nothing else. No per-run start fee, no auto-events.

Event	Price
search-result-scraped — per search result returned	$0.0008 = $0.80 / 1,000 results

Billing runs through Actor.pushData(rows, 'search-result-scraped'), which is budget-aware: delivered == billed. If you set a charge cap, you are never handed rows you did not pay for, and never billed for rows you did not get. Duplicates from YouTube's continuation chain are removed before they are pushed, so you do not pay for repeats.

Measured on the platform: the 600-row run reported chargedEventCounts: {"search-result-scraped": 600} against a dataset of exactly 600 items.

Honest comparison — do not take a blanket "cheapest" claim from anybody, including us:

• vs api-ninja/youtube-search-scraper free tier ($0.0025/result + a per-run fee): we are 3.1× cheaper per result and charge no run fee.
• vs the same actor's paid tiers ($0.00025/result on Bronze and above): they are cheaper. If you are on a paid Apify plan and only want their field set, they cost you less per row. What you get here instead is the wider row, the complete filter cross-product, and one flat rate.
• vs khadinakbar/youtube-search-scraper ($0.003/result): we are 3.75× cheaper.

---

📈 Measured performance (real runs on Apify, 2026-09-02)

Run	Config	Result
Proof run	5 queries × 120, type=video, Apify Proxy auto	600 rows, 600 unique, 69.5s, 35 requests, 100% success, 0 blocks
Depth run	2 queries, maxResultsPerQuery: 1000	820 rows, 51 requests, 100% success — both queries exhausted their continuation chain
Mixed	resultType: any	80 rows: 51 Shorts + 28 videos + 1 channel
Channels / Playlists / Movies / Shorts	one run each	80 / 40 / 19 / 40 rows, all types parsed
Filter proof	sortBy: uploadDate + uploadDate: lastHour + hd	40 rows, freshest "21 minutes ago"
Health (prefill input) ×3		SUCCEEDED in 7.3s / 30.4s / 6.0s, 50 rows each
Empty input {} ×2		SUCCEEDED in 13.3s / 23.9s, 200 rows each — never fails on empty input

Sustained-load check: 86 consecutive InnerTube requests across the two deep runs, through Apify Proxy, zero non-200 responses and zero 403/429. Requests are logged with a running success rate, so you can see it yourself in any run.

---

⚠️ Honest limits — read these before you buy

• YouTube caps how deep one query goes. Measured to exhaustion on the platform: "plumber" ran dry at 473 unique results / 31 pages, "best crm software" at 347 / 20 pages. Then the continuation token goes null and the query ends (the log says no further pages). This is YouTube's ceiling, not this actor's — every YouTube search scraper hits it. Plan for volume by running many queries and filter variants, not one enormous one.
• YouTube's continuation chain repeats itself. Measured 151 duplicates in 971 raw rows (15.6%) on the depth run. They are dropped by ID before pushData, so you are never billed for them — but it is why "20 results per page × 30 pages" does not equal 600 unique rows.
• The publish date is relative, not absolute. Search returns "6 years ago", never an ISO date. publishedTimeApprox is that string resolved against the scrape time, so it is exact to YouTube's own granularity and no better. It is named Approx on purpose. An exact timestamp needs a second request per video, which this actor deliberately does not make.
• Live rows have no duration and no publish date. That is YouTube, not a bug — a live row carries liveViewerCount instead. Verified: 20 of 20 rows under the live filter.
• Shorts rows are thin by nature. YouTube gives a Shorts result a title, a view count and a thumbnail. This actor adds channelId (100% of 51 rows). It cannot give you the channel name or @handle for a Short, because they are not in the payload.
• videoCountText / videoCount on channel rows measured 0/80. YouTube currently puts the subscriber count in that slot instead. The parser reads those fields by content rather than by name, so they will populate again if YouTube restores them — today they are empty, and this README says so rather than pretending.
• No per-video enrichment. Like count, comment count, subscriber count on a video row, full description, keywords and tags are not in the search response. They need a separate /player or /next call per video. This actor is one request per page, on purpose — it does not claim fields it cannot see.
• Sustained very-high volume is unproven. 86 consecutive requests came back clean. Nobody has run this at tens of thousands of requests an hour. If you do and YouTube starts refusing, the run exits cleanly with a "blocked" status message — never as "no results found", and never as a FAILED run. Switch proxyConfiguration to RESIDENTIAL, or bring your own proxy.
• Residential proxy costs bandwidth. A result is roughly 36 KB of response. On Apify Proxy auto (the default) bandwidth is not billed. On residential at ~$8/GB it is about $0.29 per 1,000 rows — still comfortably inside the price, but do not switch it on out of habit.

---

⚖️ Legal & fair use

This actor reads YouTube's public search results — the same rows any visitor sees, with no login, no account, no cookies and no CAPTCHA solving. It collects no personal data: channels are public business/creator profiles, and no viewer or commenter information is touched.

https://www.youtube.com/robots.txt lists, under User-agent: *:

Disallow: /results
Disallow: /youtubei/

That is quoted here verbatim so you can make your own call. robots.txt is a crawling convention, not law, and it applies to indexing crawlers; you are responsible for your own use of the output and for compliance with YouTube's Terms of Service in your jurisdiction. Keep request volume reasonable, do not republish YouTube content wholesale, and use the data for research, analysis and internal tooling.

---

❓ FAQ

How many results can I get for one keyword? Measured: 473 for "plumber", 347 for "best crm software", then YouTube stops paginating. Use more keywords and more filter variants (each variant returns a genuinely different slice).

Do I pay for duplicate rows? No. Duplicates within a query are dropped before they are pushed or charged. Set deduplicateAcrossQueries: true if you also want a result de-duplicated across different keywords.

Can I get the exact upload date? Not from search — YouTube only exposes a relative string there. publishedTimeApprox resolves it against the scrape time and is labelled approximate for that reason.

Can I get like counts, comment counts, or subscriber counts on a video row? No, and no honest YouTube search scraper can do it in one request — they are not in the search response. Pair this with a video-detail actor if you need them.

Does it get YouTube Shorts? Yes. resultType: shorts returns only Shorts; resultType: any mixes them in as youtube.com does. Each Short comes with its videoId, view count, thumbnails and the uploader's channelId.

Can I search a specific country or language? Yes — region (gl) and language (hl). YouTube ranks per country, so region: "DE" gives you the German result set for the same keyword.

Can I sort by upload date or rating? Yes. sortBy: "uploadDate" and sortBy: "rating" are encoded straight into YouTube's search protobuf. So is uploadDate: "lastHour" — verified live returning videos 21 minutes old.

Do I need a proxy? The default (Apify Proxy auto) is enough and was verified at 100% success across 86 consecutive requests. Residential is there as an opt-in for very large jobs.

Do I need a YouTube API key or an account? No. No key, no login, no cookies, no quota.

What happens if YouTube blocks the run? It exits cleanly with an explicit "blocked" status message telling you how many requests were refused, and suggests residential proxies. It never reports a block as "no results found", and it never ends FAILED for a transport problem.

What happens if YouTube changes its response shape? If results come back but nothing can be parsed, the run fails loudly rather than quietly handing you an empty dataset. If a filter enum is renumbered, the actor detects it against YouTube's own filter chips and corrects itself mid-run.

Is the result order YouTube's real ranking? Yes. searchPosition preserves the order YouTube returned, per query, after de-duplication.

What is the cheapest way to run this? Leave resultType on video (so you are not billed for the Shorts shelf), keep proxyConfiguration on auto, and set maxResultsPerQuery to what you actually need.

---

🔬 How this was verified

Nothing in this README is estimated. Every number came from a run you can reproduce:

• The parser was validated offline first against captured live responses — 44 rows on page 1, all five result shapes, the live-row absence trap, and 24/24 filter encodings against YouTube's chips.
• Then on the Apify platform, build 0.1.2: 12 runs, all SUCCEEDED, covering every result type, every filter axis, the depth ceiling, the health-test input and empty input.
• Fill percentages are computed over the 600-row proof run, not a sample.