How is AI search changing SEO, and what should businesses do about it?

Somebody deciding which clinic software to buy used to open Google, type a query, and work down a page of ten blue links. Increasingly they open ChatGPT, Claude, Gemini or Perplexity and ask a question in a sentence — and they get an answer, with two or three businesses named in it.

Nobody scrolls an answer. Being the fourth-best result on a page nobody loads is worth nothing. That single change is what is reshaping search work right now, and most of the SEO industry has not caught up with it because the old scoreboard still reports a number.

What has actually changed

Three things happened at once, and it is worth separating them because they need different responses.

Search engines started answering rather than listing

Google now generates an answer above the results for a large share of informational queries. The links are still there, below, but the question has often already been resolved by the time a reader reaches them.

Assistants became a starting point, not a destination

ChatGPT, Claude, Gemini and Perplexity are now where a lot of commercial research begins — comparisons, costs, shortlists, "who should I use for this". These tools browse the live web and cite sources, which means they can name your business or somebody else's.

The unit of success moved from a position to a mention

You can no longer fully measure visibility with a rank tracker, because a rank tracker cannot see an answer in which you were named and not clicked. A customer can arrive already knowing your price and your turnaround without your analytics ever recording the research.

Is SEO dead?

No, and anybody telling you otherwise is selling something. An AI system cannot cite a page it cannot crawl, cannot parse or does not trust, so the technical foundation of SEO matters more than it did, not less. Crawlability, speed, clean HTML, correct structured data and genuine authority are the entry ticket.

What is changing is the layer above that. The page that wins a citation is not the comprehensive 3,000-word guide that covers a topic exhaustively. It is the page that answers one specific question completely, early, and in a structure a machine can lift.

What is the difference between SEO, AEO and GEO?

The three terms overlap and are often used loosely. The distinction that actually matters in practice is what each one is trying to win.

You do not choose between them. The same page can do all three, and the work that serves the newest one — clarity, specificity, structure — happens to improve the oldest one too. That is the useful part: none of this asks you to abandon what already works.

Why do AI systems cite some businesses and not others?

From the outside it looks arbitrary. It is not. A model assembling an answer needs statements it can repeat without being wrong, attributed to a source it can name. Several things make that easy or impossible.

1. The answer is in the first paragraph. A model reading your page for a cost question and finding four paragraphs of preamble will take the number from a competitor who led with it.
2. The page contains real facts, not adjectives. "Affordable, world-class service" is unquotable. "Hand-written in Next.js, with the source code handed over at the end" is a sentence an assistant can repeat and attribute.
3. The structure is real HTML. A genuine table, a genuine ordered list, genuine headings. A price grid built from styled divs is invisible to an extractor; the same data in a `<table>` is not.
4. The business resolves to one entity. Consistent name, address and phone across your site, your structured data and your business listings. A business that looks like three slightly different businesses is one a model cannot confidently name.
5. The claims are checkable. Specific, dated, sourced content is safer for a model to cite than sweeping superlatives, because repeating an unverifiable claim is a risk the system avoids.
6. The site is machine-readable at all. Content rendered only by client-side JavaScript, hidden behind interactions, or locked in images cannot be read, and therefore cannot be cited.

What does this mean for an Indian business specifically?

Two things, and they cut in opposite directions.

The first is that the competition in generated answers is thinner than in search results. Ranking first for a competitive commercial term in India can take a year and a budget. Being the source an assistant cites for a specific question — what a procedure costs, what a licence requires, what a state's rules actually say — is frequently available to whoever writes the best answer, because almost nobody has written one.

The second is that a large share of Indian search happens in languages and scripts that most content strategies ignore. Queries typed in transliterated Hindi, or in Tamil and Bengali script, are real volume with very little competing content. A business willing to publish properly written material in the language its customers actually use is competing in a much emptier room.

How do you build a website AI tools can recommend?

This is the part we do, so treat the rest of this section as what we think and act on rather than as neutral advice. It is also all checkable against this site, which is built exactly this way.

Answer-first pages

Every substantive page opens with a direct, self-contained answer to the question it exists for — written so it survives being quoted with nothing around it. If the first paragraph needs the second to make sense, it is not an answer.

Real semantic structure

Tables as tables, lists as lists, headings in order, definitions as definition lists. Not because it validates, but because an extractor reads the tag and not the styling.

Structured data generated from the page

Schema.org markup describing the organisation, its services, its prices and its questions — generated from the same content the reader sees, so the two cannot drift apart. Structured data that contradicts the visible page is worse than none.

One resolvable entity

A single consistent business name, a single address string, one Organization node with a stable identifier. This is the least glamorous item on the list and among the most decisive.

Published specifics

Real price bands, real timelines, real constraints, and the things that go wrong. Specifics are what get quoted; they are also what a serious buyer is looking for.

A clean machine-readable mirror

We publish a plain-markdown version of our generated pages, and an llms.txt describing the business, because a crawler handed a clean heading tree keeps the structure a styled page makes it reconstruct.

Speed on the device people actually use

Hand-written code with no page builder, measured on a mid-range Android over a throttled connection. A site that takes four seconds to render is one a crawler samples less and a customer leaves.

How do you measure something that produces no click?

Imperfectly, and honestly. There is no equivalent of a rank tracker for citations yet, and anyone claiming a precise number is overstating what is knowable. What can be done:

• Ask the assistants directly, on a schedule. Put the questions your customers ask to ChatGPT, Claude, Gemini and Perplexity, record whether you are named and what is said about you, and watch the change over months.
• Watch for the arrival pattern rather than the source. Enquiries that begin already knowing your price, your process or your turnaround are usually people who researched you somewhere your analytics cannot see.
• Track branded search volume. A rise in people searching your name without a matching rise in campaign spend is often the visible shadow of invisible recommendations.
• Keep measuring ordinary SEO. It still works, it is still measurable, and the technical work underneath both is the same.

What should you do first?

1. Find out what the assistants currently say about you. Ask them. It takes twenty minutes and it is frequently the most uncomfortable and most useful thing a business owner does all quarter.
2. Fix the entity basics. One name, one address string, consistent across the site, the structured data and your business listings.
3. Rewrite your three most commercial pages answer-first. Lead with the answer, publish the real numbers, structure the content properly.
4. Audit what a machine can actually read. If the content only appears after JavaScript runs, or lives inside an image, it does not exist to an extractor.
5. Then keep writing. The durable position is owning the specific questions your buyers ask, and that is a body of work rather than a fix.

Where MyFloww fits

MyFloww is a software studio in Bengaluru. We build hand-coded Next.js websites for businesses across India, and we build them to be found by search engines and quoted by AI answer engines — because those are now two different jobs and most sites are built for neither.

If you already have a site, our website audit tells you where it stands: technical health, structured data, keyword position, and specifically what ChatGPT, Claude and Gemini can read and currently say about you. You get a prioritised written plan and you can implement it with us, with your own developer, or alone.

If you want the work done continuously, that is SEO and GEO management. If the site itself is the problem, that is Next.js website development. Everything is quoted per project as a fixed scope for a fixed price, and the bands are published rather than hidden behind a call.