How to rank in ChatGPT and AI search

There is no setting for this. No file you upload, no schema you bolt on, no agency package called AI SEO. You get quoted by ChatGPT and Google's AI answers for three unglamorous reasons: a machine can read your page, your page answers the question in a chunk worth lifting, and other websites already say you exist. That is most of it. The rest of this piece is the detail.

Does AI search matter yet, honestly?

For most businesses in India, Google is still where the customers come from. Do not let anyone tell you to abandon it. But two things have changed enough to be worth your attention.

The first is scale. Sam Altman said in October 2025 that ChatGPT had passed 800 million weekly users. Whatever you think of the company, a meaningful number of people now start a question there instead of in a search box.

The second is the clicking. The Pew Research Center tracked the browsing of 900 American adults across roughly 69,000 Google searches in March 2025. When an AI summary appeared, people clicked through to a website on 8 per cent of visits. When it did not, they clicked on 15 per cent. Only 1 per cent clicked a link inside the summary.

Read that carefully, because it gets misquoted constantly. It is a study of US browsing habits over one month, not a law of physics, and it does not mean your traffic halves. What it does mean is that a slice of the easy traffic — the people who only wanted a fact — has gone, and is not coming back. The visits you keep are the ones with intent behind them.

What Google actually says about AI Overviews

This is the part the AI SEO industry would rather you did not read. Google's own documentation for site owners, updated in December 2025, states plainly that "there are no additional requirements to appear in AI Overviews or AI Mode". It goes further and says you do not need to create AI text files or special structured data for it.

Their stated requirements are the ordinary ones. The page must be indexed and eligible to show a normal snippet. Crawling must be allowed, by your robots.txt and by whatever CDN sits in front of you. Important content must exist as text. Internal links must be able to find the page. Structured data must match what a human sees. Keep your Business Profile current.

The three things that decide whether you get quoted

1. A machine has to be able to read it

Half the sites we audit fail here before anything else. Prices rendered inside a JPEG. A services list that only exists after a script runs. A hero headline that is an image of text. You see a full page; a crawler sees a blank one.

Then there is robots.txt. OpenAI runs separate crawlers and they do separate jobs, which is useful to know before you block anything:

Plenty of sites have blocked GPTBot on principle and accidentally assumed they were safe from everything, or blocked the lot and wondered why they vanished from ChatGPT. Decide deliberately. OpenAI notes that robots.txt changes can take around a day to register.

2. It has to be quotable

A language model building an answer is looking for a passage it can lift more or less whole. So write passages that survive being lifted.

Put the answer in the first two sentences, before the context and the caveats. Use the actual question as your heading, phrased the way a person says it out loud. Keep one idea under each heading, so a section can be understood without the one above it. Put comparisons in a table and specifics in numbers — models reach for a table far more readily than for a paragraph that circles the point.

The test we use internally: take any 80 words from the middle of your page and read them cold. If they still make sense and still say something a person could act on, that passage can be quoted. If they only make sense after the three paragraphs above, it cannot.

3. Other people have to say you exist

This is the one nobody wants, because you cannot do it in an afternoon. A model answering "best branding studio in Udaipur" is not ranking websites. It is looking for agreement across the sources it can see. One website claiming to be the best is a claim. Four sources naming the same studio is a fact, as far as the machine is concerned.

Which in practice means: a Google Business Profile that is filled in and has real reviews. Credits on the client sites you built. A LinkedIn page that exists. Listings in the directories that cover your city. Being mentioned in a local news piece or a roundup. None of it is exciting. All of it is what the answer gets assembled from.

What about llms.txt?

You will be sold this one, so here is the position. llms.txt is a markdown file at the root of your site that describes your pages for AI agents. It was proposed by Jeremy Howard in September 2024. It is a community proposal, openly described as such, and it is not a ratified standard.

No AI company has announced that it uses the file to decide what to quote. OpenAI, Anthropic and Google publish llms.txt files for their own developer documentation, which gets cited as evidence of adoption and is not the same thing at all. Google's documentation explicitly says you do not need to create AI text files to appear in its AI features.

We publish one at howdy-digital.com/llms.txt anyway. It takes ten minutes, it breaks nothing, and if the thing is adopted we are already there. That is the whole argument for it. Anyone charging you for an AI visibility package whose centrepiece is this file is charging you for a bet.

What we would fix first

In this order, because the order is most of the value:

1. Make every important page readable as plain text. View your page with JavaScript disabled. Whatever disappears, a crawler may never have seen.
2. Check robots.txt deliberately. Know which AI crawlers you allow and why, rather than inheriting whatever your theme shipped with.
3. Rewrite your headings as the questions people ask, and put the answer directly underneath. This single change does more than everything else on this list.
4. Add FAQ sections with real answers — the specific ones customers ask on the phone, not the generic set. Mark them up with FAQ structured data that matches the visible text.
5. Finish your Google Business Profile and get reviews. For any "in [your city]" question, this is the strongest single signal you control.
6. Earn four or five honest mentions elsewhere. Credits, directories, a client case study on their site.
7. Then test monthly by asking ChatGPT, Gemini and Perplexity the question your customer would ask, and noting who gets named.

How to tell whether it is working

There is no Search Console for ChatGPT yet. The measurements available are rough, and anyone presenting you a precise AI visibility score is presenting you an estimate wearing a suit.

Three things genuinely tell you something. Ask the AI tools your customer's question each month and record the answer. Check your server or CDN logs for visits from OAI-SearchBot and ChatGPT-User, which prove you are being read. And watch for referral traffic from chatgpt.com and perplexity.ai in your analytics — small numbers, usually, but those visitors arrive already briefed, and they convert accordingly.

Frequently asked questions