When I audited 34 live Nepali websites in August 2026, I was measuring the usual things. Headings, canonicals, schema, image attributes. One number stopped me.
Thirty-three of the 34 sites declared a language on the <html> element. Here is what they said they were:
en— 19 sitesen-US— 10 sitesen-usanden-gb— one eachzxx— one sitene— one site
One out of 34. Ten Nepali organisations told the web they publish American English. One declared zxx, which is the ISO 639 code for "no linguistic content" — almost certainly a plugin trying to stop browsers offering to translate the page.
To be fair, most of those sites genuinely are written in English, and lang="en" is then correct. That is not the interesting part. The interesting part is what it reveals about who the sites think they are talking to.
How people in Nepal actually search
A customer looking for a dentist in Kathmandu might type any of these:
dentist in kathmandukathmandu ma ramro dentistकाठमाडौंमा राम्रो दन्त चिकित्सक
Three scripts, one intent. English, Romanised Nepali typed on a QWERTY keyboard, and Devanagari. The same person may use different ones depending on whether they are on a phone or a laptop, whether they are typing or using voice, and how quickly they need an answer.
Romanised Nepali is the one that gets ignored, and it may be the largest of the three for commercial queries. It is what you get when someone thinks in Nepali and types on a keyboard that was not designed for it. There is no standard spelling. Ramro, raamro. Kaha, kahaa. Keyword tools handle this badly because they expect a canonical form and there isn't one.
So it disappears from keyword research, and therefore disappears from content plans, and therefore nobody writes for it.
What the language attribute actually does
Let me be precise, because there is a lot of overstatement about this.
The lang attribute is not a ranking factor. Setting lang="ne" on an English page will not make you rank for Nepali queries, and if you set it wrongly you will make things worse. Google mostly determines language from the content itself and treats the attribute as a weak hint.
What it does affect is real, just smaller than people claim:
- Screen readers use it to pick pronunciation rules. A Nepali page announced with an American English voice is close to unusable.
- Browser translation uses it to decide whether to offer translation, and what from.
- Font and typographic rules — line breaking and default font selection differ by language.
- hreflang depends on it being right if you serve more than one language version.
So: fix it because it is correct and because it affects accessibility, not because it will move you up the results. The zxx case is the one genuine bug — it tells assistive technology the page contains no language at all.
The real opportunity is the content, not the tag
Here is what I think the audit number is actually a symptom of. Almost every Nepali business website is written as though its audience is a foreign visitor. Formal English, brochure phrasing, "Committed to Excellence in Patient Care".
Meanwhile the person who will actually walk into the clinic is typing kathmandu ma dentist ko price into a phone.
That gap is the opportunity, and it is wide open. In the SERPs I looked at while researching this cluster, the competition for Romanised Nepali commercial queries is thin. Everyone is fighting over best dentist in kathmandu. Almost nobody has written the page that answers dentist ko fee kati parcha.
How to find these queries when tools will not tell you
Keyword tools have poor coverage of Romanised Nepali, so you have to go to the source:
- Google autocomplete on a phone, in Nepal. Start typing a Romanised Nepali phrase and read the suggestions. Those are real queries. Do it on mobile, because suggestions differ.
- People Also Ask. Run a query in English, then look at what Google offers as related questions. Then run the Romanised version and compare — the sets are often different.
- Your own inbox and messages. The questions customers send you on Facebook and Viber are the queries, already written in your customers' words. This is the single best keyword source most businesses have and never look at.
- Search Console. Look at the queries you already receive impressions for. You will usually find Romanised Nepali terms you never targeted, where you rank on page three by accident. Those are your fastest wins.
That last one is worth emphasising. If you are already getting impressions for a query you never planned, Google has told you it thinks you are relevant. A page that deliberately answers it will usually outrank the one that stumbled into it.
What to do, in order
Set the language attribute honestly. If the page is English, lang="en" is right. If it is Nepali, lang="ne". If you have both versions, set each correctly and connect them with hreflang. If your site says zxx, fix it today.
Do not machine-translate your site into Nepali. This is the most common bad advice in this area. A poorly translated duplicate of every page gives you twice the pages and half the quality, and Nepali readers can tell instantly. It reads as though you did not care enough to write it.
Write specific pages for specific queries. Better to have three genuinely useful Nepali pages answering the three questions customers actually ask than forty translated ones nobody finishes reading.
Use the words your customers use, in the language they use them. If everyone says chek-up, write chek-up. Matching real language is not unprofessional. It is the entire job.
Why this matters more every year
AI-generated answers are increasingly what people see first. When a system assembles an answer rather than returning ten links, it is drawing on whatever content exists that addresses the question. If nobody has written a good Nepali-language answer to a common Nepali question, the answer gets assembled from whatever thin material is available — or the question gets answered in English, badly.
Whoever writes the genuinely good Nepali-language answer to a common question now is likely to be the source those systems reach for. That is a narrow window and it is open right now.
There is more on how that side of search is changing in Google Search vs AI Search.
I write in both languages — my poems and stories are in Nepali and my client work is mostly in English — and the thing I keep noticing is that the two audiences are not separate people. They are the same person, choosing a keyboard.