How to Get Your Business Cited by ChatGPT, Perplexity & AI Overviews

Five levers, and they build on each other. You cannot be cited if the engine can't read you, so machine-readability comes first; you won't be cited if you're not the clearest answer, so structure comes next; and you won't be cited widely unless people beyond your own site are talking about you, which is where the biggest signal lives.

It helps to know this is measurable, not mystical. The Princeton and IIT-Delhi GEO study tested tactics across real generative engines and found concrete winners: adding relevant statistics, direct quotations, and cited sources to a page raised its visibility in AI answers by up to 40% — and the biggest gains, over 100%, went to content that wasn't already ranked first. Keyword stuffing, by contrast, made things slightly worse. The levers below are ordered to put that evidence to work.

This is first because it is a hard gate. AI crawlers are not browsers. The Vercel and MERJ crawler study found that GPTBot, ClaudeBot, and PerplexityBot fetch your pages but execute no JavaScript whatsoever. If your content is rendered client-side — painted in by a JavaScript framework after the page loads — the engine sees an empty shell and cites someone whose content was actually there.

So the answer must live in the HTML your server sends. That means server-side rendering or static generation for anything you want cited, a clean and crawlable site structure, fast pages, and no critical content trapped behind clicks, tabs, or scripts. Valid structured data (schema.org) is worth having — it helps engines understand what a page is about — though be clear-eyed that Google says it is part of normal SEO, not a special AI lever.

This is unglamorous and decisive. A brilliant answer the crawler can't read is worth exactly nothing to it.

Engines lift answers, so give them one to lift. The pattern that works is answer-first: a heading that mirrors the exact question a person would ask, then a complete, self-contained answer in the first 40–60 words, then the depth and nuance underneath for the reader who wants it. That single block can win a featured snippet, a People-Also-Ask slot, and an AI citation at once, because each of those surfaces wants the same thing — a clean, extractable answer.

This is the opposite of the classic SEO essay that buries the answer 800 words down to keep people scrolling. An answer engine will simply skip that page for one that gets to the point. Lead with the answer; earn the read with what follows it.

Match the shape to the question, too: a paragraph answer for a "what is" query, a numbered list for a "how to", a comparison table for an "X vs Y". You are not writing to fill a page; you are writing the thing a machine will quote.

This is the single highest-leverage content move, and it is backed by the research directly. In the GEO study, the tactics that raised visibility most were adding statistics, direct quotations, and cited sources to a page — the additions that make content feel authoritative and quotable to a model. An answer engine reaching for a number, a benchmark, or an expert line will cite the page that has one.

So become the source of the number. Publish your own data — benchmarks, survey results, real pricing, before-and-after metrics, a genuine breakdown nobody else has bothered to write. Quote named experts. Cite your own primary sources inline (as this article does). Generic content that merely restates what is already everywhere gives an engine no reason to pick you; a page with a fact that exists nowhere else is exactly what it needs.

There is a strategic bonus here. The study found the biggest visibility gains — over 100% — went to content that wasn't already the top result. If you are not yet the incumbent authority in your space, original data is disproportionately how you leapfrog one.

This is where most of the actual citations come from, and where most businesses aren't looking. Remember the number: 84% of the sources AI answers cite are third-party — other people's editorial, reviews, and community discussion — not a brand's own website. You can perfect your own pages and still lose to a competitor who is simply talked about more, in more trusted places.

So the work extends past your domain. Wikipedia is consistently one of the most-cited sources across engines and ChatGPT's single most-referenced domain; Reddit and community forums feature heavily, with some analyses finding Reddit in close to half of Perplexity's citations (a volatile figure, but directionally clear). Models trust consensus across many independent sources far more than one company's self-description.

Concretely: earn real reviews on the platforms your industry trusts (for software, that is Clutch, G2, and the like); participate genuinely and helpfully in the communities where your buyers actually are, rather than spamming them; get mentioned in real publications through useful contributions and expert commentary; keep your name, category, and facts consistent everywhere so models see one coherent entity; and seed the structured, factual sources like Wikidata that feed the knowledge graphs these engines lean on. This is slower than editing your own page — and it is the dominant lever.

You cannot improve what you cannot see, and there is no Search Console for AI answers yet — so you build the measurement yourself. It is simpler than it sounds.

Run a monthly prompt panel: take the 20–50 questions your buyers actually ask an AI ("best providers of X", "how much does Y cost", "is Z worth it") and put them to ChatGPT, Perplexity, Gemini, and Google's AI Overviews. Log whether you appear, who does when you don't, and what sources the answer cited. That single habit turns AI visibility into a trend you can watch move. Pair it with your analytics — filter for referral traffic from chatgpt.com, perplexity.ai, and gemini — to see AI sending real visitors, who tend to arrive well-informed and convert above average.

Do this before you start changing things, so you have a baseline, and you will actually be able to prove what worked.

A few things look like AEO/GEO work and aren't, and knowing them saves real effort.

Skip the llms.txt file as a strategy — Google has confirmed it does not use it, and in real server logs 97% of llms.txt files receive zero AI-bot requests. It is a one-day nicety at most, never a substitute for readable HTML. Don't chase FAQ rich results either: Google fully retired them in May 2026, so while the visible Q&A on your page is valuable, the rich-result markup no longer earns you anything. Don't hunt for a special "AI schema" or meta tag — Google has stated plainly that none exists. And don't optimise for a single engine: because ChatGPT, Perplexity, and AI Overviews cite such different source pools, tuning for one leaves you invisible in the others.