Half of Nepali Websites Have No Structured Data — Which Now Costs More Than It Used To

When I audited 34 Nepali websites earlier this year I recorded their structured data along with everything else, and then wrote about other things. Coming back to that column now, it says something I did not expect.

Sixteen of the 34 carry no structured data whatsoever. Zero of the 34 carry broken structured data.

That combination is worth pausing on, because it is the opposite of what audit tools usually report. There is no validation problem here. Every site that implemented schema implemented it correctly. The gap is binary: half the sample did it and half never started.

What the half that did it are using

Across the 18 sites with markup, the types that appeared:

This is the plugin footprint. WebSite plus SearchAction plus Organization is what an SEO plugin emits when you install it and fill in two fields. It is not wrong and it is not nothing — but almost none of it describes what the business actually is or does.

What is missing is the specific stuff. In a sample containing trekking companies, clinics, colleges and banks, I would expect LocalBusiness, MedicalOrganization, CollegeOrUniversity, TouristTrip, Service, Offer, FAQPage, Event. Barely any of it appears.

So even the half that has structured data has mostly told search engines a website exists here rather than this is a hospital in Kathmandu with these departments and these hours.

Why the cost of this changed

For most of the last decade the honest answer about schema was: it is worth doing, it might get you a rich result, it is not a ranking factor, do not overthink it. I have made that argument myself, and for classic blue-link search it still broadly holds — I set out which types actually earn their place in a separate piece.

What changed is that a growing share of queries are now answered rather than listed. When a system composes an answer, it has to decide which facts it trusts enough to state. Prose has to be parsed and inferred. JSON-LD does not — it is already a set of typed assertions with an author attached.

That does not mean schema buys you a citation. It means a page carrying it has given the machine something to verify against, and a page without it has given it a wall of text to guess from. Between two otherwise similar sources, that is not nothing.

There is a Nepal-specific edge here too. For a lot of local queries there simply are not many authoritative pages competing — I found the same shape when I looked at AI crawler policy across the same sample, where 27 of 34 had never made a decision either way. A shallow pool means the few legible sources get drawn on disproportionately. That advantage exists precisely because most sites have not bothered, and it will close.

What I would add, and in what order

For the 16 with nothing, the whole useful set is about two hours of work:

1. Organization or LocalBusiness — name, address, phone, opening hours, and sameAs pointing at your social profiles. If you have a physical address this is the single highest-value item, because it is what confirms the entity on your website and the entity on your Business Profile are the same one.
2. WebSite — the identity of the site itself.
3. Article or BlogPosting on anything dated, with a real named author rather than "Admin".
4. The type that actually describes you — MedicalClinic, CollegeOrUniversity, TravelAgency. This is the step almost everybody skips and it is where the specificity lives.
5. BreadcrumbList — small, reliable, and one of the few that nearly always displays.

Two things not to do. Do not invent FAQ blocks to have something to mark up; Google reduced FAQ rich results substantially and the version that remains rewards pages that genuinely are questions and answers. And do not mark up review ratings on your own site — that is discounted for most business types and is a straightforward way to earn a manual action.

Method and limits

Thirty-four live Nepali websites across seven sectors, homepage only, audited on 23 August 2026. Every application/ld+json block was extracted and parsed; a block that failed to parse counted as invalid. None did. Types were read from the parsed graph.

Three limits worth stating. Homepage only — a site may carry richer markup on interior pages, and for the news sites in particular that is likely. Microdata and RDFa were not counted, only JSON-LD, so a site using the older formats would read as zero here. And parsing successfully is not the same as being semantically correct: markup can be valid JSON, valid Schema.org, and still describe the business inaccurately. This measures presence, not quality.

The method and the aggregate results of the audit are published, and the wider technical picture from the same sample is in the audit writeup.