One of those numbers describes what got made. The other describes what people actually chose. They have been drifting apart for years.

 

English is used by 49.5% of all websites whose content language can be identified, according to W3Techs, which surveys the web daily.

 

Second place is Spanish, at 6.0%.

 

That is not a ranking so much as a cliff. German takes 5.9%, Japanese 4.9%, French 4.5%. Chinese — spoken by well over a billion people — accounts for 1.3% of website content. Arabic gets 0.6%.

 

Now put a demand figure next to it.

 

More than one-third of all viewing on Netflix in the first half of 2025 came from non-English language titles, and 10 of the 25 most-watched series were non-English, the company reported in its Engagement Report covering January through June. The report captures roughly 99% of viewing on the platform — over 95 billion hours. "Squid Game" alone drew 231 million views across three seasons.

 

Supply sits at half English. Demand, on the largest measured catalogue in the world, has already moved a third of the way out.

The gap is infrastructure, not taste

The obvious explanation — people prefer content in their own language — is true and not very interesting. The harder question is why the supply side took so long to notice.

 

Part of the answer is that a great deal of the world never had the option. As of 2023, 54% of the world's population had limited or no access to digital language technologies, according to Ethnologue, which classifies languages on a six-point scale from Supported to Unsupported.

 

A Supported language gets voice recognition, machine translation, spell-checking, working keyboards. An Unsupported one gets nothing: no software, no fonts, no digital presence at all. Speakers of those languages face a choice Ethnologue puts plainly — struggle through content in a language they may not fully understand, or be shut out of it entirely.

 

That is not a preference. It is a filter applied before anyone gets to express a preference, and it means the low share of non-English content online is partly a measure of who was able to publish, not of who wanted to read.

Streaming ran the experiment nobody else did

What makes the Netflix figure valuable is that it removed the filter and watched what happened.

 

Subtitles and dubbing turned a Korean series into something a viewer in Ohio could watch without learning Korean. The infrastructure problem got solved at the platform level, once, and then the demand appeared — not gradually, but at a scale that now includes ten of the platform's twenty-five biggest shows.

 

The lesson generalises badly to anyone who cannot afford a localisation department, which is nearly everyone. A newspaper cannot dub itself. Neither can an independent creator. What they can do is be findable by people searching in their own language — which is a much smaller problem, and still mostly unsolved.

Where this lands for creator platforms

Subscription platforms inherited the English-first assumption wholesale, and their search boxes show it.

 

A creator's page is written in whatever language they speak, but discovery on OnlyFans runs through the account handle, which means the language of the page is invisible until you are already on it. There is no way to filter by it, no browse view that groups it. So somebody looking for free pages on OnlyFans works from outside indexes built on what creators state publicly — and those indexes carry the same limitation as everything above: they list what was published in a form somebody could catalogue. A creator writing in a language with weak digital support is under-represented there for exactly the reason they are under-represented everywhere else.

What the numbers do not say

W3Techs measures websites, not people and not usage. A language with few sites can have enormous traffic to them, and the survey cannot see that.

 

Netflix's figure is one company's catalogue, shaped by what Netflix chose to commission and license. It is evidence that demand exists when supply does, not a measurement of global appetite.

 

And Ethnologue's 54% is a 2023 derivation covering access to language technology, not internet access as such. The two overlap without being the same thing.

 

What survives all three is narrow and worth stating anyway: the web publishes in English at a rate that has nothing to do with how the world reads, and every platform that removed the language barrier discovered an audience that was there the whole time.