How to Get Found by ChatGPT and Google AI (GEO in 2026)

To get found by ChatGPT, Perplexity and Google AI Overviews, write content that AI engines like to quote: clear answers up front, real statistics with named sources, and expert quotes. In testing, content built that way earned up to 40% more visibility in AI answers (Aggarwal et al., GEO, KDD 2024). Old-school keyword stuffing does the opposite — it performs worse than doing nothing.

That's the short version. Here's how it actually works, and what to do about it.

What changed about getting found?

For twenty years, "getting found" meant ranking on Google and earning a click. That's shifting. AI Overviews — the AI summary at the top of Google results — appeared on roughly 15–18% of searches in late 2025 (Semrush, 2025). When that summary shows up, fewer people click through. Ahrefs measured a 34.5% drop in clicks for the top organic result when an AI Overview is present (Ahrefs, 2025). Seer Interactive measured a 54.6% year-on-year drop across all positions (Seer Interactive, 2025).

So the goal is no longer only "rank first." It's also "be the source the AI quotes." That's what people mean by GEO — Generative Engine Optimization.

How do you get cited by AI engines?

The most useful evidence here is a study from Princeton, Georgia Tech, Allen AI and IIT Delhi, presented at KDD 2024. The researchers tested nine ways of writing across 10,000 queries. Three things stood out (Aggarwal et al., KDD 2024):

Cite your sources — link to credible references inside the text: +30–40% visibility
Add quotations — quotable lines an AI can lift verbatim: +30–40% visibility
Add statistics — specific numbers, not vague claims: +30–40% visibility

They validated this on the live Perplexity.ai engine, where the same techniques delivered a 22–37% real-world improvement (Aggarwal et al., KDD 2024). One caveat worth being honest about: the headline 40% came from a lab engine, and it was only confirmed on Perplexity, not on Google AI Overviews or ChatGPT. But the mechanism is consistent — AI models favour well-sourced, specific writing — so it's a sound bet across engines.

Content with citations, statistics and quotations earned up to 40% more visibility in AI answers — and keyword stuffing scored about 10% worse than doing nothing (Aggarwal et al., KDD 2024).

This is exactly the kind of finding we design around. Every page we build leads with a direct answer, names its sources, and uses real numbers — because that's what the research rewards.

What does "answer first" actually mean?

AI engines quote the part of your page that answers the question fastest. So put the answer in the first sentence, then explain underneath. This is the old journalist's "inverted pyramid," and it maps neatly to how these systems extract text.

In practice that means:

A one- or two-sentence answer at the very top of the page
Question-style headings that match how people actually ask
Short paragraphs, lists and tables instead of dense walls of text
Specific figures — "34.5% fewer clicks" beats "significantly fewer clicks"

It's worth knowing who already wins here. In Google AI Overviews (US, March 2025), the three most-cited sources were Wikipedia, YouTube and Reddit, and government sites were cited about 3× more than you'd expect from their share of results (Pew Research Center, 2025). You won't out-Wikipedia Wikipedia — but you can be the clearest, best-sourced answer in your niche.

Does speed still matter for AI and Google?

Yes, though less as a ranking lever and more as a trust-and-conversion one. Core Web Vitals — Google's speed and stability scores — are a confirmed ranking signal, but they act as a tiebreaker rather than a dominant factor (Search Engine Journal, 2025). The targets to clear are an LCP of 2.5 seconds or less, an INP of 200 milliseconds or less, and a CLS of 0.1 or less (Google, 2025).

The bigger payoff is business outcomes. In Google's own case studies, Vodafone saw an 8% sales lift after a 31% improvement in load time, and Redbus saw mobile conversions rise 80–100% (web.dev, 2024). These are before-and-after studies, not controlled experiments, so treat them as strong signals rather than guarantees. We treat fast load times as a build rule, not a nice-to-have — partly for ranking, mostly because slow pages quietly cost sales. (Our the way we build websites page covers the how.)

What about schema and llms.txt?

A few technical pieces help AI and search engines understand your site:

Schema markup (structured data) feeds Google's Knowledge Graph and rich results. Organization, Article, BreadcrumbList and Product are the types still worth doing (Google, 2025). One catch: AI crawlers don't run JavaScript, so schema must sit in the static HTML, not be injected by a tag manager (Ahrefs, 2025).
llms.txt is a simple file that lists your key pages in plain Markdown for AI engines. Over 1,000 sites use it, including Stripe and Cloudflare (llmstxt.site, 2025). There's no proof yet that it lifts citations — but it's cheap to add, so the risk/reward favours doing it.

FAQ

What is GEO?

GEO stands for Generative Engine Optimization — shaping your website so AI engines like ChatGPT, Perplexity and Google AI Overviews quote it in their answers. The proven levers are citations, statistics and clear quotable writing (Aggarwal et al., KDD 2024).

Is SEO dead now that AI answers questions?

No. Traditional Google search still drives most organic traffic. But AI Overviews already reduce clicks notably when they appear — up to 34.5% for the top result (Ahrefs, 2025) — so being the cited source matters more each year.

Should I block ChatGPT and other AI bots?

For visibility, allow the search bots (OAI-SearchBot, PerplexityBot) so you appear in AI answers. Blocking training-only crawlers like GPTBot has no effect on AI search visibility (OpenAI docs, 2025) — that's a separate, philosophical choice about your data.

How long should a page be?

Long enough to answer the question, no longer. Google says it has no preferred word count (Google, 2025), and over a billion pages with more than 1,000 words get zero traffic (Ahrefs, 2025). Match length to intent and structure it clearly.

The pattern across all of this is the same: AI engines reward pages that are fast, clearly structured, and honestly sourced. That's how we build — calm, hand-coded websites tuned to exactly this research, not guesswork. If you'd like a site built to these numbers, book a call and we'll talk it through.