Which platforms are actually blocked, country by country

Counting confirmed blocks produces a censorship ranking topped by Germany and the United States

The obvious way to build this page is to ask OONI how many measurements in each country came back “confirmed blocked”, add them up, and sort. Doing exactly that on this month’s data puts Germany fifth with 1,087 confirmed blocks, the United States sixth and the United Kingdom eleventh, and it leaves China in eighteenth place — below Canada, Switzerland and France.

The reason is that a confirmed count is an absolute number measured over wildly different volumes. The United States contributes 609,390 measurements in this window and the United Kingdom 130,918; Iran contributes 36,036. A large, heavily-measured country accumulates a long tail of confirmed blocks from school networks, corporate filters and parental controls, and that tail is arithmetically larger than the entire measured output of a country that actually censors.

So every country figure on this page is a rate. A platform counts as blocked at national scale in a country when confirmed blocks exceed 2% of that country’s measurements for it, and as widespread above 20%. Below the lower threshold, the block is real but local — one ISP, one school, one company — and it is excluded from every ranking here.

Practical ruleAny censorship statistic that reports a count rather than a rate is measuring probe volume, not censorship. Ask what the denominator was before you cite the number.

China’s blocking is invisible to a confirmed-only view

There is a second trap underneath the first, and it is worse because correcting the first does not fix it. OONI can only mark a measurement “confirmed blocked” when it observes positive evidence of blocking, such as a block page served in place of the real site. Blocking that works by DNS tampering or by resetting the connection produces an anomaly instead — a strong signal that something interfered, without the artefact that would confirm it.

China is the clearest case in the dataset. Across the 25 platforms measured, 23 come back with elevated anomaly rates and essentially none come back confirmed. A view built on confirmed blocking alone therefore reports China as freer than the United Kingdom, which is a conclusion no reader would accept and no honest method should produce.

One row in the China data deserves its own caveat: Netflix comes back as accessible on 65 measurements. That is above our 30-measurement floor but still thin, and more importantly it measures whether the domain is reachable — not whether the service will serve a viewer there. Network reachability and service availability are different questions, and this dataset only answers the first.

Iran blocks 21 of 25 platforms — including every VPN website we could measure

Iran is the one country in this window where the measured picture needs no hedging. 21 of the 25 platforms are confirmed blocked on most networks measured, spanning every category in the test list: social, messaging, news and video, circumvention tools, and the websites of VPN providers themselves. The rates are not marginal — Facebook is confirmed blocked in 74.7% of Iranian measurements.

The VPN-provider category is the part no competitor publishes, and it is the reason we included it. These are the nine providers in our own database — the set we recommend from — and we measured their public websites the same way we measured everything else. In Iran, 8 of them are confirmed blocked at rates between 81.7% and 87.1%.

The practical reading is narrow and worth stating precisely, because the tempting version is wrong. This does not show that a VPN fails to work in Iran; it shows that the pages you would use to obtain one are unreachable from most measured Iranian networks. The consequence is about sequence rather than capability — the moment to install and pay for a tool is before you are behind the border that blocks its shop.

Practical ruleA country that blocks circumvention websites has made the download, not the connection, the scarce step. Judge a tool by whether it can be acquired in advance and used offline-first, not only by whether it connects.

Russia’s blocking is partial by design, and that is a finding rather than a caveat

Russia sits second on the measured ranking with 12 platforms blocked at national scale, but the shape of its numbers is completely different from Iran’s. Where Iranian confirmed rates cluster above 80%, Russian rates sit between roughly 15% and 20% — real, repeated, confirmed blocking that reaches a minority of the networks measured rather than nearly all of them.

That is what a network-by-network blocking regime looks like in measurement data: some ISPs implement an order and others have not, or implement it differently. Reporting it as “Russia blocks Facebook” without the rate would be true in spirit and misleading in practice, because the measured experience of a Russian user depends on which provider they are on. Four further platforms — X (x.com), Telegram, Discord, YouTube — show elevated anomalies without confirmation, and are listed separately below rather than folded into the count.

Germany and Mexico look like censors in the raw output, and are not

Two countries rank third and fourth on the measured scale with 8 platforms each above the national-scale threshold: Germany and Mexico. Both are flagged by the collector and excluded from the ranking, and the reason is visible in the category breakdown rather than in the totals.

The confirmed rates make the same point quantitatively: in Germany the VPN-provider domains sit between roughly 2.5% and 6.2% of measurements, and every social, messaging and news platform in the list is either accessible or shows only isolated confirms — 11 of them. A state blocklist does not look like that. A corporate or campus web filter with an “anonymisers” category switched on looks exactly like that, and such networks are well represented among volunteer OONI probes.

We publish the flag rather than silently dropping the rows, because the alternative — a censorship map that quietly decides which inconvenient countries to omit — is not inspectable. A reader who disagrees with our reading can see the same numbers and reach their own.

The noisiest signal in the dataset is anomalies on VPN and Tor websites

If anomalies were a reliable blocking signal, the list of countries showing them on circumvention domains would look like a list of authoritarian states. It does not. In this window, Kazakhstan (9), Switzerland (8), Kenya (6), Denmark (5) all show elevated anomaly rates only on VPN-provider and Tor domains, with no confirmed national-scale blocking anywhere in the list. Some of that is filtered networks in the probe pool again; some is plausibly the domains themselves, which sit behind aggressive anti-abuse infrastructure that can produce connection failures indistinguishable from interference at measurement level.

The contrast is what makes the flag useful. When elevated anomalies appear on social and news platforms instead — which they do in Mongolia (5), Tanzania (5), Guinea (4) — the ordinary explanations stop fitting, and the case for looking harder gets stronger. Neither pattern is a blocking claim. One is worth an investigation and the other mostly is not.

The country-by-platform matrix

Below is every country in the pull that showed any signal at all — 31 of 181 — against all 25 platforms measured. Countries not listed showed no confirmed blocking above the isolated threshold and no elevated anomalies; that is a measured result for the ones with coverage and an absence of evidence for the ones without, which the next section separates out.

A third of the grid was never measured, and pretending otherwise would be the real error

Of the 3,586 country-by-platform cells in this pull, 1,238 — 35% — fall below the 30-measurement floor. OONI runs on volunteer probes, so coverage follows where volunteers are, not where censorship is. 64 countries have usable coverage on all 25 platforms; others have almost none.

This is also the reason the matrix above uses a distinct mark for unmeasured cells rather than leaving them blank. A blank cell in a censorship table reads as “fine here”, and for roughly a third of this grid that reading would be unsupported.

Method, in enough detail to be checked

Every figure here comes from OONI’s aggregation API for the web_connectivity test, one request per target domain returning all countries at once, for the window 2026-08-01 to 2026-09-01 (end exclusive). The target list is 25 domains across six categories, and the exact hostname for each was probed individually — nordvpn.com carries measurements where www.nordvpn.com carries none, and for other providers it is the other way round, so a “tidied” hostname would silently produce a zero that reads as “nowhere blocked”.

Repeatability: the collector is a script in our repository, the thresholds are constants inside it, and every threshold is written into the output file alongside the results. Re-running it against a later month regenerates this page in full. It is pulled monthly.

Limitations we are aware of. OONI probes are volunteer-run, so coverage is uneven and skews toward networks where volunteers install the app — which includes filtered corporate and campus networks, the source of the Germany and Mexico artefact. Measurements are of domain reachability, not of service usability. A month-long window smooths over short blocking events, which will under-report protest-driven shutdowns. And the target list is 25 platforms chosen by us; a different list would produce different country totals, which is why the totals are always reported as “of 25 platforms measured” rather than as an absolute score.

Source, licence and how to cite this

The measurements are collected and published by OONI, the Open Observatory of Network Interference, a free software project that has been running network measurement tests through volunteer probes since 2012. OONI publishes its data openly under a Creative Commons licence; the analysis, thresholds, category choices and every judgement about what the numbers do and do not support are ours, and OONI has not reviewed or endorsed them.