Facebook Reels Scraper — Views, Reactions & MP4 URLs

Pull every Reel a public Facebook Page has published, as clean rows. One row per reel: permalink, video_url_hd, video_url_sd, thumbnail_url, caption, posted_at, duration_seconds, view_count_approx, reaction_count, comment_count, share_count_approx, audio_track_title and the author Page (page_name, page_id, page_url, page_is_verified). No login, no cookies, no session token, no CAPTCHA solving.

{
  "pages": ["redbull", "gopro"],
  "maxReelsPerPage": 20,
  "proxyConfiguration": { "useApifyProxy": true }
}

Click Try for free and press Start — that block is literally the input this Actor ships with. It returned 40 reels, 0 duplicates, 100% fill on views, reactions, comments, shares, thumbnails and MP4 URLs, and cost $0.08.

Calling it from the API, an MCP client or an AI agent with an empty input is safe too: with no Pages supplied the Actor runs that same two-Page sample (20 reels each) and says so in the run status, instead of failing.

---

Read this before you buy rows

Six things that would otherwise turn into a refund request.

1. View counts are ROUNDED. Reactions and comments are EXACT. This is the single most important thing on this page. Facebook itself only publishes a bucketed play count on the reels surface — the literal string it renders is "3.2M", "807K", "9.8K". On a 360-reel sample, 93.9% of view counts came back bucketed (they carry a K/M/B suffix and are rounded to two significant figures); only the 6.1% under 1,000 are exact. You get both view_count_text ("3.2M" — what Facebook literally said) and view_count_approx (3200000 — that string expanded to a number). Do not present view_count_approx as a precise figure. reaction_count and comment_count, by contrast, are exact integers on 100% of rows. share_count sits in between: 85.6% of rows carried an exact integer, 14.4% were bucketed.
2. There is no keyword mode and no hashtag mode. This Actor reads a Page's public reels tab. Facebook puts hashtag pages, keyword search and Watch behind a login — verified 2026-08-15, all four surfaces (/hashtag/…, /search/videos/, /watch/search/, /reel/?q=) redirect a logged-out client to a login form. Anything that claims to search reels by keyword without a login is doing something this Actor deliberately does not do. You must supply Pages.
3. Some Pages have no public reels tab, and that is not a bug. Facebook answers "This content isn't available right now" for those — measured identically on datacenter and on residential IPs, so it is a content state, not a block. Of 20 varied brand Pages tested, 2 answered this way (tiffanyandco, Corona) and 2 more served a reels tab with an empty collection (NASA, thedodo). Those Pages are logged, skipped, and never billed.
4. video_url_hd and video_url_sd are signed, expiring CDN links. They are the real MP4s and they play, but Facebook signs them with an expiry (hours, not days) and ties them to the fetching context. Download what you need promptly; do not store them as permanent links. permalink is the stable one.
5. No follower count, no email, no contact details. This Actor returns reels, not Page profiles. If you need Page-level fields (likes, category, phone, email, website, bio), run our Facebook Page Scraper and join on page_id.
6. No comments text, no commenter list. comment_count is a number. The comments themselves are a different record and are not in this dataset.

---

What you get

Field	What it is
reel_id	Facebook's numeric reel/video id. Stable, and the de-duplication key.
permalink	https://www.facebook.com/reel/<id> — the durable public link.
page_slug	The handle you supplied.
page_name	The Page's display name, as Facebook prints it.
page_id	Numeric Page id — join key to our Facebook Page Scraper.
page_url	Canonical Page URL.
page_is_verified	Blue-check state of the author Page.
caption	The reel's caption text, with line breaks and emoji preserved.
posted_at	ISO 8601 UTC publish timestamp, to the second.
duration_seconds	Playable length, fractional.
view_count_text	What Facebook literally rendered — "3.2M", "9.8K", "415".
view_count_approx	That string expanded to a number. Rounded — see limit #1.
reaction_count	Exact total reactions (all types combined).
comment_count	Exact comment count.
share_count_text	Facebook's share string — exact below ~1,000, bucketed above.
share_count_approx	That string as a number.
thumbnail_url	Poster frame, full resolution.
video_url_hd / video_url_sd	Signed MP4 URLs. Expiring — see limit #4.
video_width / video_height	Source pixel dimensions.
audio_track_title	e.g. "Nike · Original audio". Sparse — see below.
is_original_audio	Whether the audio is the creator's own.
is_ai_generated	Facebook's own AI-content detection flag.
is_reshare	Whether the reel is a reshare of someone else's.
post_id	The underlying post id (differs from reel_id).
video_permalink_url	Facebook's /videos/ form of the same asset.
scraped_at	When this row was captured.

---

Field fill — measured on 360 live reels

360 unique reels across 12 public brand Pages (Cristiano, Red Bull, National Geographic, BuzzFeed, Sephora, GoPro, Starbucks, Tasty, LADbible, ESPN, LEGO, Airbnb), 30 reels per Page, pulled through Apify datacenter proxies on 2026-08-15. Counted, not estimated.

Field	Fill
reel_id	100.0%
permalink	100.0%
page_slug · page_name · page_id · page_url · page_is_verified	100.0%
posted_at	100.0%
duration_seconds	100.0%
view_count_text · view_count_approx	100.0%
reaction_count	100.0%
comment_count	100.0%
share_count_text · share_count_approx	100.0%
thumbnail_url	100.0%
video_url_hd · video_url_sd	100.0%
video_width · video_height	100.0%
is_original_audio · is_ai_generated · is_reshare	100.0%
post_id · video_permalink_url	100.0%
caption	94.2%
audio_track_title	43.6%

The sparse ones, plainly:

• audio_track_title — 43.6%. Facebook only names the audio when the reel is attached to an identified track or the creator's own original audio is registered. On the other 56.4% Facebook publishes no track name at all, and the field is null.
• caption — 94.2%. 5.8% of reels are published with no caption text.
• video_url_hd — 100% here, 98.9% on a wider 624-reel sample. A handful of reels are only distributed at SD; video_url_sd was present on every row in both samples.
• is_ai_generated and is_reshare were false on all 360 rows in this sample. The fields are populated (not null), the flags simply did not fire on these Pages. Treat them as a signal you can filter on, not as a distribution you can trust from this sample.

Value ranges in the same sample, so you know what the columns look like: views 316 → 71,000,000 (median 166,000); reactions 0 → 4,523,560 (median 1,001); comments 0 → 135,670 (median 47); shares 0 → 126,000 (median 56); duration 2.5s → 13,310s (median 36.2s — yes, Facebook files some multi-hour videos under the reels tab). Publish dates in the sample ran from 2018-12-14 to 2026-08-15.

---

How deep it goes

Facebook serves the reels tab in batches of 10, newest first, and this Actor follows the same pagination the browser does — contiguously, with no gaps.

• Sephora, maxReelsPerPage: 100 → 100 reels in 10 contiguous batches, 0 duplicates, 100% engagement fill, covering 2026-06-08 → 2026-08-14.
• 12 Pages × 30 reels → 360 rows, 0 duplicates, 36 contiguous batches.
• Across every measurement run on this page, in-run duplicate rate on a single Page was 0.0%. Duplicates only appear if you list the same Page twice under different aliases (e.g. gopro and GoPro) — those are de-duplicated on reel_id before billing, so you never pay twice.

Depth is bounded by what the Page has published, not by the Actor. Set maxReelsPerPage to whatever you need; the log tells you when a Page ran out of reels versus when it hit your cap.

---

Reliability — measured through Apify proxies, not from a laptop

A residential home IP tells you nothing about how a scraper behaves from a datacenter, so every number here was measured through Apify Proxy.

Rung	Calls	Usable	Notes
Apify DATACENTER, one pinned session per Page	20 Page fetches + 50 pagination calls	68 / 70 = 97.1%	The 2 misses were Pages with no public reels tab, not blocks
Apify RESIDENTIAL (spot check)	9	9 / 9 HTTP 200	Outcomes byte-identical to datacenter on the same Pages, including the empty ones
Across all runs on this page	122	118 / 122 = 96.7%	0 transport blocks observed. All 4 misses were Pages with no public reels

On Pages that actually publish reels, the through-Apify success rate was 118/118. That is why this Actor has no residential fallback rung: datacenter is not walled here, and a reels payload runs about 1.4 MB per reel, so a residential retry would burn several times the revenue of the rows it recovered. You can still select RESIDENTIAL in the proxy input if your account needs it — just budget for the data transfer.

Retries rotate to a fresh datacenter session, because the failure mode on Facebook is a served-but-empty response rather than an HTTP error code. There is one extra safeguard: the reaction/comment/share counters arrive in a separate deferred chunk of the same response, and Facebook occasionally omits it — when a Page comes back with zero engagement fill, the Actor re-fetches it once on a fresh session before accepting the rows, and logs how many rows still ended up without counters.

---

How it works

The reels tab at https://www.facebook.com/<page>/reels/ is server-rendered: the first batch of 10 reels is a plain JSON blob embedded in the HTML that a logged-out client receives. No headless browser is needed to read it, which is why this Actor runs on 2 GB instead of the 4–8 GB the browser-based reels scrapers ask for.

Later batches come from the one Relay query the browser fires when you scroll, ProfileCometAppCollectionReelsRendererPaginationQuery. That query is answerable logged out, but only if you send it exactly right: the request tokens Facebook prints into the same HTML (lsd, haste session, revision, jazoest), the connection node id and cursor from page_info, and Relay's 19 feature-gate "provided variables" with the precise values Facebook chose for that response. Guess any of them and the server returns noncoercible_variable_value, so this Actor lifts them verbatim out of the page's own preloaded-query blob rather than hard-coding them.

The persisted-query id is not in the HTML — it lives in a JS bundle. The Actor tries to discover it at run time from the bundles the page links, and falls back to a pinned id when it cannot. Measured 2026-08-19 on 8 platform runs: the bundle that defines this query is lazily loaded and is not among the ones the reels tab links, so run-time discovery does not currently fire and every run uses the pinned id. That pinned id paginated 8/8 runs, including a 300-reel / 30-batch pass on a single Page, so pagination is working — but the self-healing path is not carrying it, and the id is re-verified on each release rather than per run.

The engagement counters are not in the reels connection at all; Facebook streams them later in the same response as deferred chunks, keyed back to a reel by its video id. The parser reads both and joins them.

---

Pricing

$0.002 per reel returned. One event, reel-scraped. That is the cheapest per-reel price on the Apify Store for a Facebook reels Actor as of 2026-08-15 (live comparison: $0.0045, $0.00499, $0.00599, $0.006, $0.007 across the other per-reel actors).

Reels	Cost
20 (shipped default, 2 Pages)	$0.04
100	$0.20
1,000	$2.00
10,000	$20.00

You are billed for unique reels actually returned, and only for those:

• Reels are de-duplicated on reel_id before the charge, across Pages and across pagination batches.
• A Page with no public reels tab costs nothing.
• A Page that would not serve a payload after retries costs nothing.
• A reel removed by the postedAfter filter costs nothing — the filter runs before billing, so it costs you run time, not money.
• If a run ends with zero reels it finishes SUCCEEDED with a status message naming why (no public reels / no payload / everything removed by your date filter) and bills nothing — it does not fail the run on you, and it does not silently pretend it worked.

---

Input

Field	Type	Default	What it does
pages	array	["redbull","gopro"]	Handles, Page URLs, /profile.php?id=… URLs or numeric Page ids.
startUrls	list source	—	Bulk paste, a .txt/.csv upload, or a linked Google Sheet of Page URLs.
maxReelsPerPage	integer	100 (prefill 20)	Per-Page cap. Facebook pages in 10s.
postedAfter	string	—	YYYY-MM-DD. Keep only reels published on or after this date.
proxyConfiguration	object	Apify Proxy	Required. Datacenter default is what the numbers above were measured on.
requestDelayMs	integer	500	Pause between Facebook requests.

Handles, full URLs and numeric ids can be mixed freely, and duplicates across pages and startUrls are dropped before anything is fetched.

---

Who buys this

• Ad-creative and short-form research. Every organic reel a competitor ran, with the MP4, the caption, the exact reaction and comment counts and the posted date — a swipe file with real engagement attached, refreshed on a schedule.
• Brand and competitor monitoring. Point it at a set of Pages with postedAfter and a daily schedule to get posting cadence, format mix and engagement trend per brand.
• Creator and partnership sourcing. reaction_count / view_count_approx gives a per-reel engagement rate you can rank creators by — with the caveat from limit #1 that the denominator is rounded.
• Video and dataset work. video_url_hd plus caption and duration_seconds is a ready ingestion feed for transcription, captioning or vision pipelines.

---

When a run fails

Log line	What it means	What to do
no public reels ("This content isn't available right now")	The Page has no public reels tab.	Nothing — not billed. Check the Page in a logged-out browser.
Page has a reels tab but publishes no public reels	Reels tab exists, collection is empty.	Nothing — not billed.
no reels payload after retries	Three fetches on three sessions all came back without data.	Re-run; if it persists on one Page only, that Page is likely region-restricted.
reels pagination stopped after N reel(s)	Facebook ended the connection early.	The N reels already returned are yours; re-run to go deeper.
engagement counters missing from the first response	The deferred counter chunk was omitted; the Actor is retrying.	Nothing — informational.
No reels scraped — N Page(s) publish no public reels …	Every Page you listed has no public reels, or your date filter removed all of them.	Run ends SUCCEEDED, nothing billed. Check your handles or widen postedAfter.
Stopped early to stay within the run time limit	The Actor hit its own wall-clock budget and stopped before the platform could kill the run.	The reels already pushed are yours; raise the run timeout or lower maxReelsPerPage.
No Facebook Pages supplied — running the built-in sample	You started the Actor with an empty input (common from the API or an AI agent).	The Actor runs the two-Page demo (redbull, gopro, 20 reels each) so you get real rows. Fill in pages to scrape your own.

---

Legal & fair use

https://www.facebook.com/robots.txt states, for the generic crawler group:

User-agent: *
Disallow: /

Facebook's robots.txt disallows generic crawlers site-wide and reserves crawl access to a named allow-list of bots. This Actor reads only publicly visible pages — no login, no cookies, no session token, nothing behind an authentication wall — and it returns business/brand Page content, not private individuals' data. It does not download or re-host video files; it returns the URLs Facebook itself serves. Robots.txt compliance, Meta's Terms of Service and the applicable data-protection rules in your jurisdiction are your responsibility as the operator of the run. If you are scraping at volume, or for a commercial product, take your own legal advice first.

---

Our other Facebook Actors

• Facebook Page Scraper — Page name, category, likes, talking-about, phone, email, website, address and bio. Join to this dataset on page_id.

---

Feedback

Found a Page this Actor mis-reads, or a field you need that Facebook publishes and this does not return? Open an issue on the Actor page with the Page handle — measured bug reports get fixed.