SEO and GEO

Search engine optimisation predates generative answers by decades. Its generative counterpart was formalised in 2023 by Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande, in a paper accepted to KDD 2024 that found black-box content changes could lift a source's visibility in generative-engine responses by up to 40%, with the effective strategy varying by domain^[6]. The same paper describes generative engines as rapidly replacing traditional search engines including Google and Bing^[7], a framing of the shift rather than a measurement of it. See generative engine optimization for the paper itself.

The two overlap most where the answer engine is Google's own. BrightEdge tracking found the share of AI Overview citations that also rank organically rose from 32.3% in May 2024 to 54.5% by September 2025^[8]. Semrush, across 5,000 keywords, found the relationship is engine-specific: Perplexity overlaps Google's top 10 on more than 91% of domains, AI Overviews on around 86% and AI Mode on around 54%, while ChatGPT overlaps least and, on Semrush's earlier research, tracks Bing more closely than Google^[5]. Ahrefs, sampling 15,000 long-tail prompts, measured only 12% overlap between the links cited by ChatGPT, Gemini and Copilot and Google's top 10 for the same prompt^[9]. The query sets and the definition of overlap differ between those studies, so they are not one contested number.

Two behaviours have no strong analogue in classic ranking. Around 60% of commercial queries are reported to cite content refreshed within the previous six months, and pages updated within twelve months to be twice as likely to keep a citation they already hold^[3]. The weight placed on a brand's own pages also varies by engine: Perplexity is reported to emphasise industry expertise, customer reviews and earned media over a brand's own site^[4], ChatGPT to take 48.73% of its citations from third-party directories, and Gemini 52.15% of its citations from brand-owned sites^[10]. Those figures come from vendor panels rather than from the engines. That makes brand mentions on other people's domains a GEO asset in a way backlinks alone never were.

Not every SEO lever moves an AI citation. Ahrefs tracked 1,885 pages that added JSON-LD schema and found no meaningful citation gain on any platform^[11], which matches John Mueller's April 2025 statement that structured data is used to display search features rather than to improve ranking^[12]. Content structure and E-E-A-T cover the content-side signals that both surfaces read.

Traditional search results tend to be predictable and stable, so one observation is a fair sample. AI answers vary across runs, prompts and time, which is why GEO research reports visibility as a distribution over repeated queries rather than as a single reading^[13]. AI visibility measurement covers the sampling methods and how wide the error bars are, and zero-click search covers what happens to the traffic either way.