What is generative engine optimisation?

Generative engine optimisation is the work of getting your business named inside the answer an AI assistant gives, rather than getting a link ranked on a page of results. When somebody asks ChatGPT for the best accounting software in India, they get a paragraph naming three or four companies. If you are not one of them, you were not in the running, and there is no second page to be on.

Why this is a different job from SEO

Search engines return a list and let the person choose. An assistant returns a decision and cites a handful of sources it used to make it. Those are different outcomes, so the work that produces them is different in three concrete ways.

The unit of winning changed. A search engine can show ten results. An assistant usually names three. There is no position eleven, and being fourth is closer to being invisible than it is to being third.

The source set is narrower. When an assistant answers a buying question it tends to read a small number of pages: category listings, comparison articles, forum threads, and the vendors' own pages when those pages carry something specific enough to quote. You are not competing with every page on the web. You are competing for a place in a short list of sources.

Specificity beats optimisation. A page written to rank often hedges. A page that gets quoted states a number, names the thing, and gives a date. An assistant asked to recommend something will reach for the page that lets it be concrete, because a vague page gives it nothing to say.

What actually moves it

Three things, in the order they usually matter.

Pages on your own site that answer one buyer question each. Not a services page. A page that answers the exact question somebody types, with the answer in the first paragraph and the evidence underneath.

Being present on the sites the assistants already read. Category directories, review sites, and the comparison pages in your sector. An assistant that has never seen your name anywhere except your own site has one source for you and several for your competitor.

Facts that can be checked. Coverage stated as a count rather than "comprehensive". Pricing with a date rather than "contact us". A method described well enough that a reader could repeat it.

The six kinds of page that actually get cited

From our own runs across four companies in four sectors, the pages that earned citations fell into six shapes. This is the most directly useful list we have, because it is what we would write first for a new client.

• A price page with real numbers and a date. Even a range, even with conditions attached. Assistants prefer a business they can describe in money terms over one they cannot.
• A named coverage or scope list. Which regions, which integrations, which certifications, which courts. Named, counted and dated, never "comprehensive".
• A "how to choose" page that does not conclude with you. Written to help somebody decide, including when the right answer is a competitor or no purchase at all.
• A limits page. What you do not do and who should buy something else. Quoted far more often than most businesses expect, because a careful recommendation wants exactly this.
• A documented method or process page. Step by step, including what goes wrong. Assistants prefer a source that shows its working.
• A page answering the awkward commercial question your sector avoids in public: the total cost, what the guarantee excludes, why a quote varies so much.

How you know whether it is working

You ask the assistants the questions your buyers ask, without naming yourself, and count how often you are named. Naming yourself in the question is the most common mistake and it ruins the measurement: an assistant asked "is Acme good?" will say something polite about Acme. An assistant asked "what is the best X for Y?" will tell you the truth about whether Acme came to mind.

Two things make that count honest. Ask more than one assistant, because they answer the same question differently on the same day from different sources. And put a range on the number, because a handful of questions cannot produce a percentage you should believe.

There is a third thing that matters more than most people realise: ask the assistant whether it actually ran a live web search. Some answer from memory, which is months old and cannot reflect anything you published recently. A reading taken from memory and a reading taken from live search are two different measurements, and averaging them produces a number that means nothing.

What specificity looked like in our own run

On 17 September 2026, on our sister company clawlaw.in, Claude was asked how to find every case filed against a company in India. Two enterprise vendors were placed above clawlaw.in, and Claude said why: they publish explicit court and tribunal coverage lists. It added that its ordering reflected source authority and price transparency rather than product quality, and that a tool could be better than what it ranked above and still lose the position by publishing fewer verifiable numbers.

That is the whole argument for specificity, made by the engine rather than by us. The competitors did not win on quality. They won on having published a list.

The same run produced the reverse case. Two clawlaw.in comparison pages stated competitors' prices with no link, no date and no source. Claude checked those figures independently and found them correct, and it still treated the pages as advocacy and used official sources instead, because at the moment of reading it had no way to know they were right. Unsourced competitor pricing cost more citations in that run than anything else.

What it cannot do

It cannot make an assistant recommend a business that does not fit the question. If you sell to enterprises and the question is about free tools for freelancers, the honest answer is that this is not your question, and no amount of content changes that.

It also cannot be guaranteed. Anyone promising you a position in an AI answer is promising something they do not control, and the assistants change what they do without telling anybody.

It will not win the questions that belong to official sources. A question about how to use a government portal should be answered by that portal, and it will be. Trying to outrank an official source on its own subject is the least productive work available in this field.

And it is slower to read than search. A rank tracker gives you a number every day. An assistant's answer varies between runs, so a single check tells you very little and you need several readings before a change means anything.

Common questions

Is this just SEO with a new name?

No, but it overlaps more than the marketing around it suggests. Being crawlable, having real depth on a subject, and being mentioned on other sites all matter to both. What changes is the target: ranking puts you on a list somebody chooses from, and being cited puts you inside the answer they were given. The work that produces the second is more specific and less keyword shaped.

How long does it take?

In our own runs the first citation of a new page arrived within a few weeks of publishing, which is faster than the same pages took to rank in ordinary search. A change in your overall mention rate takes longer, because a rate only moves when many questions move. Expect weeks for the first signal and months for the rate.

Do I need to do this if I already rank well on Google?

Ranking well is a good starting position and it is not the same as being named. Every business in our runs ranked respectably for its own category and was close to absent when an assistant was asked the same thing in plain words. Nothing in their existing reporting would have shown it, which is the uncomfortable part.

Can I pay to be included in AI answers?

Not in the way you can buy an advertisement, and you should treat anybody who says otherwise with suspicion. What you can do is make yourself the easiest business in your category to describe accurately, which is a content and evidence problem rather than a media buying one.

How many pages does this take?

Fewer than most agencies will tell you, if they are the right pages. Fifty pages that each answer a real buyer question beat five hundred that do not. The count matters much less than whether each page answers something somebody actually types.

What is the single cheapest thing to fix first?

Check that your pricing and product pages can be read without JavaScript running. In more than one of our runs, the prices an assistant quoted came from a third party listing rather than the company's own site, because the company's own prices only appeared after scripts ran and a crawler received an effectively empty page.

What to do first

Write down the ten questions your buyers actually ask, in their words, and ask them of two assistants without naming your company. Record who gets named and, more importantly, which pages get cited. That list of cited sources is your real competitive set, and it almost always contains one source you had not thought about and are not on.