How to Show Up in AI Search Answers, Not Just Google

A school administrator in Lalitpur needs student management software. Two years ago she would have typed "school software Nepal" into Google and worked down the results. Now there is a fair chance she opens ChatGPT or Gemini and asks which companies in Nepal build it, then contacts the two or three that get named.

There is no page two in that conversation. You are either in the answer or you are invisible.

This is what people mean by GEO (generative engine optimization) and AEO (answer engine optimization). Both are attempts to name the same problem: search results are being replaced by summaries, and the summary only mentions a handful of businesses.

How an AI answer actually gets assembled

The mechanics matter, because they explain why the usual SEO advice only half works.

When someone asks a question about Nepali software companies, the assistant runs a search behind the scenes, pulls a set of pages, reads them, and writes an answer from what it found. It is not consulting a ranking. It is reading text and deciding which sentences answer the question.

Two things follow from that. First, position one is worth less than it used to be, because a page ranked seventh gets read alongside the rest. Second, the thing being extracted is a sentence, not a page. If your site has no sentence that directly answers the question, the model has nothing to lift, and it takes the answer from a competitor who wrote one.

Write the answer, not the pitch

Most homepages in Nepal open with something like "your trusted technology partner for digital transformation". Nothing in that sentence can be extracted. It contains no company, no service, no place, no price.

Compare it with a sentence that names things: "Antbytelabs Nepal builds custom ERP, school management systems and e-commerce sites in Kathmandu, starting at NPR 25,000 for a business website." That version survives being summarised, because every part of it is a fact a model can repeat and attribute.

Apply the same test page by page. Put the direct answer in the first two sentences under each heading, then explain underneath. If a reader has to get to paragraph four to learn what you charge or where you are, a model reading the page will usually stop before it gets there too.

Question-shaped headings help for the same reason. "How much does a website cost in Nepal?" matches what someone typed far more closely than "Pricing".

Facts survive summarisation, adjectives do not

Watch what happens when a model compresses a page. Numbers, dates, names and prices come through. "Leading", "innovative" and "world-class" get dropped, because every competitor's page says the same thing and none of it distinguishes anyone.

So publish the specifics most agencies hide. Real price ranges. Your founding year. The number of projects delivered. Which payment gateways you have integrated, by name. How long a typical build takes. Whether the client owns the source code. Our own website cost breakdown for Nepal exists for exactly this reason.

Published prices feel risky. In practice they filter out the enquiries you were going to lose anyway, and they give an AI something concrete to say about you when a buyer asks who is affordable.

Be the same business everywhere

Models build a picture of your company from every source they can find, then look for agreement between them. Where sources disagree, confidence drops, and a business they are unsure about is a business they leave out.

Your name, address and phone number should match exactly across your website, Google Business Profile, LinkedIn, GoodFirms, Clutch and any local directory you are listed in. Not "roughly match". The same spelling, the same phone number, the same city.

This is where most Nepali companies lose. Three different phone numbers across four profiles, a LinkedIn page under a slightly different name, an address that says Kathmandu on the website and Bhaktapur on a directory. Each mismatch chips away at the confidence a model has in describing you.

Being listed somewhere you do not control matters too. When a model looks for software companies in Nepal, it reads directories, comparison articles and forum threads alongside your own site. A company that appears in three of those has an easier time getting named than one that only appears on its own domain, however good that domain is.

Structured data does the reading for you

Schema markup gives a machine your facts without asking it to infer them from your layout. Three types earn their place on most business sites:

• Organization on every page, with your legal name, logo, address, phone and links to your profiles elsewhere. This is what ties your site to the entity a model has built from other sources.
• FAQPage on any page with a genuine question-and-answer block. Each pair becomes a clean question with a clean answer, which is close to the exact shape an assistant wants.
• Article on blog posts, with the author and the date it was updated. Assistants prefer recent sources, and an unmarked page from 2022 looks the same age as one from last week.

Add the FAQ block for readers who will actually use it, then mark it up. Schema for questions nobody asks is wasted work.

Check whether any of it worked

You cannot see rankings here, but you can see three other things.

Ask the assistants directly. Open ChatGPT, Gemini, Perplexity and Claude, and ask the questions your buyers ask: who builds e-commerce sites in Kathmandu, which company integrates eSewa, what does a mobile app cost in Nepal. Write down who gets named. Repeat it monthly. Slow, manual and still the most honest signal available.

Read your server logs. AI crawlers identify themselves in the user agent, including GPTBot, ClaudeBot, PerplexityBot and Google-Extended. If none of them have touched your site, nothing else you do here matters yet, and the first thing to check is whether your robots.txt is blocking them.

Watch referrals in analytics. Traffic from chatgpt.com and perplexity.ai now shows up as its own source. The volume is small next to Google. The intent behind it is much stronger, because the person arrives having already been told you are worth contacting.

What this does not replace

Every assistant mentioned above finds pages through a conventional search index. A site that ranks nowhere gets retrieved by nobody. Page speed, crawlability, internal links and pages worth linking to still decide whether you are in the pool of documents that gets read at all. That groundwork is ordinary SEO and performance work, and it comes first.

GEO is what you do once that groundwork holds: write pages that answer questions in their first sentence, publish facts specific enough to be quoted, and make sure every profile with your name on it says the same thing. If you want to see the pattern applied to a real page, our Kathmandu software company page is built this way, and the FAQ answers each question in its first line.

Most competitors in Nepal have not started. The buyer asking an assistant which company to call is getting an answer today, from whatever pages happen to be readable.