Information-gain assets to add before publish:
- Side-by-side screenshots of the same intent searched three ways, with the SERP for each.
- Your Search Console query export showing which script real Nepali visitors actually used to reach your pages.
- A tool screenshot showing reported volume for a Devanagari term versus its romanised equivalent.
Keyword research built for a single-language market breaks in Nepal. The same person, looking for the same thing, will type it in English, in romanised Nepali, or in Devanagari depending on their device, their keyboard, and how formal the search feels. Keyword tools treat those as three unrelated terms and understate all of them.
The short answer
Research intent in all three scripts, not just English. Use Search Console and Google autocomplete as your primary sources rather than volume estimates from Western keyword tools, because those tools have thin data for Nepal and will show zero volume for terms that get real traffic.
The three-script problem
One intent, three surface forms. A Kathmandu resident looking for a dentist might type any of these:
| Form | Example | Who tends to type it |
|---|---|---|
| English | dentist in Kathmandu | Urban, English-comfortable, often on desktop or searching for a professional service |
| Romanised Nepali | kathmandu ma dentist | Mobile users typing quickly on a QWERTY keyboard without switching input method |
| Devanagari | Nepali script equivalent | Users with the Nepali keyboard enabled, or using voice search in Nepali |
All three are the same commercial intent. A keyword tool will show them as three rows, two of which report no volume, and most people building a content plan will drop those two.
Why keyword tools understate Nepal
Global keyword tools build volume estimates from clickstream panels and third-party data partnerships. Both are thin for a market Nepal’s size. Three consequences follow:
- Reported zero does not mean zero. A term can drive real traffic and still show no volume, because the tool has no panel data from Nepal for it.
- Romanised Nepali is invisible. Tools built on English-language corpora do not recognise “ma”, “ko” or “kasari” as meaningful tokens and cluster them badly.
- Devanagari volume is fragmented. Spelling variation in Nepali typing is wide, so a single intent splits across several written forms, each with a small reported number.
The practical response is not to abandon tools. It is to demote them from primary source to sanity check.
Where the real data is
1. Search Console, filtered to Nepal
This is your only first-party source and by far the most reliable. Filter the performance report by country and read the actual queries. You will see romanised and Devanagari terms bringing impressions to pages you never optimised for them. That list is your keyword research.
If your site is new and has no data, borrow it: the same report on any Nepali site you have access to will show the same pattern.
2. Google autocomplete, with location set correctly
Type the first two words of your intent in each script and record what Google suggests. Do this in an incognito window with location set to Nepal, otherwise you are researching a different market. Autocomplete reflects real query volume in a way no third-party estimate does.
3. People Also Ask
Expand every question, then expand the questions that appear underneath. This is the fastest way to map the sub-questions that matter, and it works identically in all three scripts.
4. YouTube search suggestions
Video is a dominant format for Nepali-language search, and YouTube’s suggestion engine is more forgiving of romanised Nepali than Google Ads’ keyword planner. Terms that show nothing in a keyword tool will often autocomplete readily on YouTube.
5. Your own inbox and messages
The exact phrasing customers use when they contact you is the phrasing they used to search. Read your last fifty enquiries and note the words, not your words for the same thing.
Which script should you actually write in?
This is the question everyone asks and the answer is not “all of them.”
Google’s guidance on multilingual sites is explicit: use different URLs for each language version, and use hreflang annotations to connect them. What Google warns against is adapting content based on IP analysis, because Googlebot does not vary its crawl location and will simply never see your alternate versions.
So the decision is structural, not stylistic:
- One language per URL. Do not mix Nepali and English versions of the same content on one page and hope Google sorts it out.
- Only build a Nepali-language section if you will maintain it. A half-translated site is worse than a well-made English one.
- Romanised Nepali is not a separate language version. It is a query form. You cover it by using the natural phrasing inside otherwise English or Nepali content, not by building a third site.
A workable research process
- Write down the intent in plain language, before touching a tool. “Someone wants to know what a website costs in Nepal.”
- Search that intent in English, in romanised Nepali and in Devanagari, from a Nepal location. Screenshot each SERP.
- Record the autocomplete suggestions and the People Also Ask questions from all three.
- Pull Search Console queries for Nepal and mark which of your recorded terms already bring impressions.
- Check a keyword tool last, purely to catch high-volume English terms you missed. Ignore its zeros.
- Cluster by intent, not by string. All three script forms of one intent belong to one page.
- Choose the primary phrasing for headings based on which form dominates your Search Console data, not on which has the highest tool volume.
Mixed-language queries
Nepali search is frequently code-mixed: an English noun inside a Nepali sentence structure, or the reverse. “Website banauna kati kharcha lagcha” carries an English loanword inside a Nepali query. These behave as their own terms and are almost never in keyword tools.
You cannot pre-plan for every mixed form. What you can do is write naturally about the topic using the words your customers actually use, including the English loanwords they use, rather than translating everything into formal Nepali that nobody types.
What this changes about content planning
If you accept that tool volume is unreliable for Nepal, the ranking criterion for what to write next changes. Instead of sorting a list by monthly searches, sort by:
- Whether the intent leads to an enquiry.
- Whether Search Console already shows impressions you are not converting.
- Whether the existing results genuinely answer the question or just occupy the SERP.
That is a better prioritisation in any market. In Nepal it is the only one available.
Frequently asked questions
Should I write my website in Nepali or English?
Write in the language your customers read for that decision. Professional services, B2B and higher-value purchases skew English in Nepal; consumer, local and informational content skews Nepali. If you build both, give each language its own URL and connect them with hreflang.
Why does my keyword tool show zero volume for Nepali terms?
Because the tool has thin panel data for Nepal, not because nobody searches the term. Treat zero as no data rather than no demand, and verify against Google autocomplete and your own Search Console query report.
Do I need hreflang for a Nepali and English site?
Yes, if you publish the same content in both languages on different URLs. Hreflang tells Google which version to show which user. Without it, the two versions compete with each other in the same results.
Does Google understand romanised Nepali?
It handles it well enough to serve relevant results, which is why these queries appear in Search Console at all. What it does not do is give you a tidy volume estimate, so plan for the intent and let the phrasing appear naturally in your content.
Is machine-translating my English site into Nepali a good idea?
Only if a Nepali speaker edits every page afterwards. Unedited machine translation produces content that reads as low effort to both readers and quality raters, and it is one of the clearer routes to being treated as low-value.
Sources
- Managing multi-regional and multilingual sites – Google Search Central
- Tell Google about localized versions of your page – Google Search Central