SEO & AI

Google Search vs AI Search: What Should Nepali Businesses Optimise For?

Every few months someone announces that SEO is dead and something else has replaced it. This time the claim has more behind it than usual, so it is worth looking at the actual numbers before deciding what to do.

What the 2026 data says

The measurements are consistent enough across independent sources to take seriously:

So the pessimists are right about the mechanism. When Google answers the question itself, far fewer people click anything.

But there are two numbers that complicate the panic, and they get much less attention.

Seer Interactive tracked 53 brands across 5.47 million queries from January 2025 to February 2026. Click-through rate on AI Overview queries bottomed out at 1.3 per cent in December 2025 — and then rose to 2.4 per cent by February 2026. Nearly double, in two months.

And being cited inside an AI Overview delivers around 35 per cent more organic clicks than not being cited.

Those two together change the strategy completely. The pie is smaller. But there is a specific, identifiable position within the smaller pie that is worth more than the old position one, and it is being distributed to whoever the system decides is a credible source.

An important caveat about Nepal

Every figure above comes from studies of predominantly US, English-language search. I have not found equivalent measurement for Nepal, and I am not going to pretend the numbers transfer cleanly.

Some reasons they might differ here: AI Overview coverage varies by query language and by country, and rollout has not been uniform. Query mix matters — a market with a high share of local and transactional searches sees fewer AI Overviews than one dominated by informational questions, because Google shows them less for "dentist near me" than for "what causes tooth sensitivity".

What I would say with confidence: the direction is the same everywhere, the timing differs, and informational content is affected first and hardest.

What actually changes for a Nepali business

Sort your pages into two piles.

Pile one: pages that answer a question. "What causes tooth sensitivity." "Best time to trek Annapurna." "How much does a root canal cost." These are exposed. An AI answer can satisfy the question without a click, and increasingly does.

Pile two: pages someone needs to actually reach. Booking a trek. Finding your opening hours. Comparing your prices. Contacting you. An AI summary cannot complete a booking. These are far more durable.

Most Nepali business websites are almost entirely pile two, and are therefore less exposed to this than a content publisher would be. If your traffic comes from people looking for a service in Kathmandu, AI Overviews are a smaller threat to you than the headlines suggest.

The businesses that should be worried are the ones whose traffic strategy was to write informational blog posts and hope some of those readers convert. That model was already weak. It is now much weaker.

What earns a citation

Here is the part that makes this less frightening than it sounds. The things that get content cited in an AI answer are mostly the things that got it ranked in the first place.

The differences that seem to matter:

Notice what is absent from that list: keyword density, word count, publishing frequency. The tactics that were already hollow are now completely worthless.

What you have to change about measurement

This is the real disruption, and almost nobody talks about it.

If your reporting is "organic sessions, month over month", it is going to show decline even when your visibility improves. Impressions can rise while clicks fall. You can win the citation and lose the session.

What to track instead, all available in Search Console:

I would go further: for most small businesses, sessions were never the right number anyway. Traffic that does not turn into customers was always a vanity metric. AI search is just making that obvious faster.

So what do you optimise for?

You do not choose. The question is framed wrongly.

There is no separate discipline where you optimise for AI systems instead of search engines. The AI answer is assembled largely from pages the search index already considers good. The overlap between "ranks well" and "gets cited" is very high. Anyone selling you a distinct AI-optimisation service is mostly selling you SEO with a new invoice.

What genuinely changes:

  1. Informational content is worth less than it was. Stop treating volume of blog posts as a strategy.
  2. Content that only restates what already exists is worth close to nothing. It cannot be cited over its own sources.
  3. Being a primary source — original data, first-hand knowledge, genuine local specifics — is worth much more than it was.
  4. Your measurement has to separate visibility from clicks or it will mislead you.

That is a real shift, and it happens to reward the same thing good SEO always rewarded: knowing something worth knowing and saying it clearly. The people hurt hardest are the ones who built on thin content, and they were living on borrowed time regardless.

If you want a view on which of your pages sit in which pile, send me a message. It is usually a short conversation and it tends to be reassuring.