Generative Engine Optimization (GEO): How to Get Cited by ChatGPT and Perplexity

Why care now?

Because the AI answer layer is already taking the clicks. Pew Research analyzed 68,879 real Google queries in July 2025: when an AI Overview appears, 8% of users click an organic result, versus 15% without one. Ahrefs, across 300,000 keywords, measured position-1 CTR down 58% on affected queries between late 2023 and late 2025.

The flip side gets less attention: according to Seer Interactive, brands cited inside an AI Overview earn roughly 35% more organic clicks than brands absent from the same results page. The traffic doesn’t vanish entirely; it gets redistributed toward whoever the answer cites. GEO is the work of being on the right side of that redistribution.

Lever 1: let the right robots in

This is the absolute precondition, and it’s a plain text file.

A site that blocks GPTBot in robots.txt will almost never be cited by ChatGPT, however good its content. The correlations published in 2026 are blunt: domains blocking GPTBot get about 0.003 ChatGPT citations per Google ranking held, versus 0.417 for domains that allow it. Blocking a major AI crawler amounts to erasing yourself from the answer layer.

The nuance that matters: AI bots don’t all do the same job. Search and citation bots (OAI-SearchBot, ChatGPT-User, PerplexityBot, Claude-User) fetch content to answer a user, citing you in the process. Training bots (GPTBot for OpenAI’s models, CCBot, Google-Extended for Gemini) collect data to train future models. Blocking training crawlers is a defensible editorial stance; blocking search crawlers, for a business that wants customers, is self-sabotage.

Go check your robots.txt now. No GEO fix has a better effort-to-impact ratio.

Lever 2: write to be quoted, not just to rank

Answer engines don’t lift pages, they lift passages. A quotable passage is self-contained: it can be extracted verbatim and still be true, dated and understandable.

What works, based on observed citations:

• Answer first. Open each section with the answer in 40 to 60 words, then elaborate. Models favor passages that address the question without a warm-up.
• Headings phrased as questions. “How much does an SEO audit cost?” matches a real user query far better than “Our pricing”.
• Dated, sourced figures. A claim with a number, a date and a source is vastly more quotable than a generality. Models look for facts they can attribute.
• Visible freshness. Perplexity clearly favors recent content; showing the year in your title and an honest updated date helps. Google’s AI Overviews, meanwhile, draw 97% of their citations from the existing top 20 organic results: classic SEO remains the price of entry.

A word on llms.txt, since it comes up in every audit we run: the file is a useful convention for developer tools, but the major AI crawlers currently ignore it, and Google confirmed in May 2026 that it isn’t needed for any of its AI features. Publish one if it costs you little (we did), just don’t expect miracles from it. We wrote a separate piece on it.

Lever 3: exist beyond your own site

Language models build trust by cross-checking. A brand mentioned consistently across independent sources (trade press, Reddit, category directories, YouTube, review sites) is far more likely to be offered as an answer than a brand that only exists on its own domain.

The 2026 studies point the same way: brand mentions correlate with AI visibility about three times more strongly than classic backlinks do. That’s a quiet but real reversal of the SEO hierarchy: links passed authority to Google; mentions pass existence to the models.

In practice, for a small business or SaaS: an up-to-date listing in your category’s directories, genuinely useful (non-promotional) answers in the forums where your audience asks questions, and ideally one piece of proprietary data published yearly that others cite back to you.

Where does E-E-A-T fit?

The signals that reassure Google reassure the models too: a named author, with a page saying who they are and a Person schema to disambiguate them; publication and update dates; a clear publishing entity (legal notice, address, history). AI engines happily write “according to [name], [role] at [company]”. Give them the parts of that sentence.

How to measure your GEO today

Honestly, measurement is still artisanal in 2026. Three useful habits:

1. Ask ChatGPT, Perplexity and Gemini the five questions your customers ask you, and note who gets cited. Repeat monthly.
2. Watch your analytics for AI user agents and referrers (ChatGPT-User visits, perplexity.ai referrals): low volume, highly qualified.
3. Audit the fundamentals: crawler access, structure, E-E-A-T. That’s what the GEO dimension of the PimpSEO report checks, scored out of 100 like the other eight. Paste your URL and you’ll know within three minutes whether answer engines can even read your site.

GEO is not a separate discipline from SEO. It’s 2026’s SEO, extended to engines that answer instead of listing. The sites that win apply the same virtues everywhere: technically open, factually precise, humanly identifiable.