We Measured 689 Facebook Video Downloads. 64% Failed — Then a New Engine Cut It to 26%

Every downloader site says the same thing when a Facebook link fails: make sure the video is public. It reads like a brush-off.

We had the data to check whether it was true, so we did.

Update, 2 October 2026: the engine behind our Facebook to MP4 converter changed in August (yt-dlp’s Facebook extractor had broken; we moved to a maintained one), and the numbers moved with it. Over the 14 days to 2 October, 684 Facebook jobs: 26% failed, down from 64%. By link type: Reels 14%, /share/r/ 12%, /share/v/ 30%, page /videos/ 50%, /watch?v= 44%. TikTok failed 7.6% and YouTube 6.7% on the same servers. What still fails includes posts that are not public. Facebook Stories never work from outside and are now refused the moment you paste one. The figures below are the August measurement, corrected on 4 October to count only visitors' jobs; the first version included our own automated test jobs.

The numbers #

These are visitors' jobs from our own servers over the 30 days to 4 August; our own automated test jobs are left out.

Splitting by where the link came from #

Splitting the Facebook jobs by the shape of the URL is where it gets interesting:

A /share/v/ link failed 60% of the time and a /reel/ link 48%. Both point at Facebook videos, and both went through identical code.

When you tap Share → Copy link inside the Facebook app, you are copying something out of a feed that Facebook assembled for you, while you were logged in. That feed is full of posts from friends, closed groups and restricted audiences. You can see them. A server visiting as an anonymous stranger cannot.

When someone pastes a /reel/ URL, they usually found it while browsing public content — so it is more likely to be public.

/share/p/ at 100% makes sense once you see it: p is for post, and most of those are photos, text or link posts. There was never a video to download.

We checked whether it was our own fault first #

The honest failure mode here would be blaming Facebook for our own broken code, so we tested that directly.

We took 8 Facebook URLs that had previously downloaded successfully on our system and ran them again:

The extractor still handled public Facebook video. It failed only on the links that were already failing. At the time we read that as the content, not the code. The engine change in August showed otherwise: Reels went from 48% failed to 14%.

We also ruled out the usual suspects: routing through a different network changed nothing (we A/B tested a second instance against the same URLs and got byte-identical results), and the extractor was on its then-current release.

Then we took 15 failing /share/v/ links and followed each redirect to see what they actually pointed at:

So it is not only closed groups. Plenty of apparently-normal Reels failed too — and after the engine change only 14% of Reels did, so most of those failures were down to software, not privacy.

The part that genuinely is tooling #

We are not going to pretend all of it is Facebook.

In July the widely-used open-source extractor that handles most platforms could not parse Facebook pages at all — it returned Cannot parse data on every reel we tested, with or without cookies. That was a real upstream problem, and it is why so many downloader sites got worse at Facebook at the same time. Since 2 October it works again and is our first choice for Facebook.

Two separate things were happening at once, and this was the bigger one: once we stopped depending on it, Facebook failures fell from 64% to 26%.

What this means if you are trying to download a Facebook video #

What we changed after seeing this #

Two things, both about being honest with people.

The first is speed of failure. A doomed Facebook job typically spent about 20 seconds on our servers, because the system retried three times before giving up — sensible for a flaky network, pointless when the answer is a definite "this is not public". Some hung for half an hour. Across 440 failed jobs from visitors that is roughly 15 hours of jobs running only to be told no.

The second is the message. "Something went wrong" is useless when the real answer is "this video was never public". Telling someone that immediately is worth more than another retry.

The uncomfortable conclusion #

A 64% failure rate looked like a broken product, and much of it was: after we changed engines it fell to 26%. We published the numbers anyway, because the alternative is the thing every downloader site does — quietly fail, blame the user, and let them try six more tools that will fail identically.

Facebook is not YouTube. It was built as a private network between people who already know each other, and a post that stays private cannot be downloaded by anyone.

Some of the failure is the platform. In July, most of it was software — and fixing the software is what moved the number.