How to get cited by ChatGPT and AI Overviews

A client asked us last month why their traffic was flat while their enquiries were up. The answer was in their referrer data: a slow trickle of visits from AI assistants, each one converting at roughly six times the rate of an organic search click.

That is the shape of the shift. Fewer clicks, better clicks. Search increasingly answers the question directly and names a couple of sources. You either get named or you do not exist in that conversation.

Here is the work that gets you named, ordered by return on effort. This is the companion to our SEO playbook for Indian D2C stores — the fundamentals there still apply, this layers on top.

First, be readable by a machine

None of the clever stuff matters if the crawler cannot read the page.

Server-render your content. If the answer only appears after JavaScript executes, assume it is invisible. Many retrieval crawlers do not run JavaScript at all. This is the most common reason a well-written page never gets quoted. Static generation or server rendering fixes it permanently.

Do not gate the answer. Content behind a signup, a cookie wall or an interstitial cannot be cited. Put the substance in the HTML and keep the conversion offer alongside it, not in front of it.

Keep the HTML clean. Semantic headings, real paragraphs, actual lists. A page built from nested divs with visual styling for hierarchy reads as mush to a parser.

Check your robots rules. Plenty of sites blocked every unfamiliar user agent during the first AI panic and never revisited the decision. Retrieval crawlers are how you get cited. Look at what you are refusing.

Second, write passages that can be lifted

This is the actual craft, and where most content fails.

An AI answer is assembled from quotable fragments. A fragment is quotable when it is self-contained: it states the claim, it does not depend on the previous three paragraphs, and it fits in a couple of sentences.

Lead with the answer. Ask the question as a heading, then answer it in the first sentence underneath. Reasoning, nuance and caveats come after. The inverted pyramid is back, and this time a machine enforces it.

One idea per heading. Headings phrased as real questions do disproportionately well because they match the retrieved query almost exactly.

Put numbers in the sentence. “Most Indian D2C stores load in 6 to 8 seconds on 4G” is quotable. “Load times are often poor” is not. Specifics get cited, vagueness gets skipped.

Give it something to attribute. First-hand data, a price band, a timeline from a real project, a small original study. Assistants preferentially quote sources that say something the other twenty pages did not.

Date it and update it. Freshness is weighted heavily. A visible published date, an updated date, and genuinely revised content beat a stale page with better prose.

Third, make yourself an entity, not a string

Retrieval systems reason about entities — a company, a person, a product — not just keywords on a page.

Publish consistent structured data. Organization markup on the site, with the same legal name, address, email and social profiles everywhere. Article or BlogPosting on posts, with author and dates. FAQPage where you genuinely answer questions. BreadcrumbList so hierarchy is explicit.

Keep your name identical everywhere. Google Business Profile, LinkedIn, directories, your own footer. Inconsistent naming splits your identity into several weak entities instead of one strong one.

Give the author a real identity. Posts written by a named person with a bio page and a track record carry more weight than posts by “Admin”. Link the author to their page in the markup, not just visually.

Get mentioned elsewhere. Citations in AI answers correlate strongly with being discussed on sites the model already trusts — trade publications, community forums, comparison pages, and yes, Reddit and Quora threads. One well-received answer in a genuine community discussion often outperforms months of link building.

Fourth, cover the question, not the keyword

Keyword-per-page thinking produces thin pages no assistant needs. Topic coverage produces the page that gets quoted.

Take one buyer question — “should I build an app or a website” — and answer it completely: the decision framework, the cost bands, the timelines, the failure cases, the counter-argument. One thorough page beats six shallow ones, and it is the page an assistant can rely on for a full answer. We did exactly this with our own app-versus-website post, and it remains one of our most-quoted pages.

Then link related pages together deliberately. Internal links tell a retrieval system which pages belong to the same topic cluster and which page is the authority in it.

The housekeeping, briefly

An llms.txt file listing your key pages in plain text takes an hour. Support is inconsistent and it has no bearing on Google ranking, so publish it if you like tidiness, but do not mistake it for strategy. Do it after everything above.

Measuring it

You cannot pull a rank report for AI answers, so build a crude one.

Run a monthly prompt panel. Write down the twelve questions your buyers actually ask. Ask each assistant every question on the first of the month. Log three columns: were you mentioned, were you quoted, what was quoted. Three months of that data tells you more than any tool.

Track assistant referrals separately. Set up a segment for assistant domains in your analytics. Watch conversion rate, not volume — volume will look disappointing and conversion will not.

Watch branded search. AI mentions often show up as a lift in people searching your name directly a few weeks later. That is the channel working even when the referral number stays small.

What we would do first

If you have one week:

1. Confirm your key pages render without JavaScript.
2. Add FAQ blocks with real questions to your five most commercial pages, marked up with structured data.
3. Rewrite the opening paragraph of those pages so the first sentence answers the question.
4. Fix Organization markup and make your business name identical across every profile.
5. Publish one page with original data nobody else has.

That is the whole game right now. It is unglamorous, it is mostly structure, and it compounds — which has been true of search for twenty years.

Want the audit done for you, with the prompt panel set up and tracked? We run that as a fixed-scope engagement.