23 of 34 Nepali Websites Contain No Nepali At All

I have written before about Nepali websites failing to declare Nepali — one of 34 sites in my sample set lang="ne". That is a markup problem and a small one.

This is a different question, and I did not expect the answer. Forget what the pages declare. How much Nepali text is actually on them?

I fetched all 34 homepages again on 5 September 2026, stripped the scripts and styles, and counted characters in the Devanagari Unicode block against total text length. No interpretation, just arithmetic.

The result

Twenty-three of thirty-four homepages contain zero Devanagari characters. Not a heading, not a menu item, not a word.

Eleven contain some. Ten contain enough to matter. And they are not evenly spread — they are almost entirely one sector:

Three news sites publish in Nepali. Everything below them is a rounding error — a stray word, a tagline, a menu item in Devanagari sitting in an otherwise English page.

Every trekking company, every clinic, every tourism body in the sample: zero. Including ntb.gov.np, the Nepal Tourism Board, whose homepage does not contain a single Nepali character.

The declared languages are worse than the content

Alongside the count, the lang attribute across the same 34:

en on 19 sites, en-US on 10, en-us on one, en-gb on one, ne on one, missing on one — and on one site, zxx.

zxx is the ISO code for no linguistic content. It exists for things like instrumental music. A Nepali website is declaring to every browser and crawler that it contains no language at all. That is almost certainly a template default nobody looked at, which is rather the point.

Why I think this happens, and why the obvious explanation is incomplete

The obvious answer is tourism. A trekking company sells to foreigners, so it writes in English. Fair enough, and for that sector I would not argue.

But the sample is not all trekking companies. It includes hospitals, colleges, banks, government bodies and marketplaces whose customers are overwhelmingly Nepali, reading Nepali, searching in Nepali. Those sites are also English-only.

What I think is actually going on is that English is the default of the tooling and the profession rather than a decision anyone made. Themes ship in English. Agencies write in English. The people commissioning the site are comfortable in English. Nobody chooses to exclude Nepali; it simply never comes up, and the result is a hospital website that a large share of its own patients cannot read comfortably.

There is a plainer commercial version of the argument too. If a customer types a query in Nepali — and a great many do, in Devanagari or in Romanised Nepali — a page with no Nepali on it has almost nothing to match against. I have written about that mechanism in Nepali language search. This study is the supply side of the same problem: the demand exists, and there is essentially nothing to serve it.

What this is not an argument for

Not translation. A machine-translated Nepali version of an English site is usually worse than no Nepali version, because it reads as translated, and because it doubles your pages without doubling your meaning. I would rather read a company's English than its Google Translate Nepali.

Not bilingual everything. Maintaining two full versions of a site is real ongoing work and most small businesses will not sustain it.

What I would actually do is narrower: identify the two or three pages where a Nepali-speaking customer makes a decision — services and prices, how to reach you, how to book — and write those properly in Nepali. Written, not translated. That is a day of work, it is maintainable, and it covers the pages that matter.

And if you do it, declare it: lang="ne" on the Nepali pages, correct hreflang between the versions. The markup is the cheap part, but it is the part that tells a search engine the Nepali page exists.

Method

Thirty-four live Nepali websites across seven sectors — the same sample as my earlier technical audit, re-fetched on 5 September 2026. Homepage only. Scripts, styles and tags stripped, entities decoded, then characters in the Unicode range U+0900–U+097F counted against total text length. All 34 fetched successfully.

Two honest limitations. Homepages only — a site could have Nepali on interior pages that this does not see, though a homepage with no Nepali at all is a strong signal. And the Devanagari range covers Nepali and Hindi alike, so this measures script rather than language; in this sample the distinction is not material.

The aggregate counts, method and limits are published in the research section. Point the same method at your own sector and tell me what you find — I would be glad to be shown a sector where this is not true.