Generative Engine Optimization (GEO) is the practice of structuring your web content so that AI systems — ChatGPT, Perplexity, Google AI Overviews, Claude — select it as a cited source in their answers. Classic SEO gets you ranked on page one of a search results page. GEO gets you quoted when there is no page one anymore.
The shift is already measurable. According to a 2024 study by SparkToro and Datos, zero-click searches on Google now account for more than 60% of all queries in the US and EU — meaning users read the AI-generated answer and never visit any website. For branded queries and informational intent keywords, the number is even higher. If your content strategy still optimizes only for ranked position, you are optimizing for a surface that a growing share of users never see.
What GEO actually is
The term was formally defined in a 2023 research paper by Aggarwal et al. from Princeton, IIT Delhi, Georgia Tech, and Microsoft Research titled "GEO: Generative Engine Optimization." The researchers tested nine content modification strategies — adding citations, including statistics, using authoritative language, improving fluency — and measured which ones caused large language models to increase a source's visibility in generated responses. Their finding: citations and statistics had the strongest positive effect, increasing AI-sourced visibility by up to 40% in their test set.
Where classic SEO is about signal density (keywords, backlinks, page authority), GEO is about answer quality. AI systems are trained to produce accurate, trustworthy, citable answers. Content that reads like a primary source — specific, structured, factual, attributed — is more likely to be surfaced and quoted than content that reads like a blog post optimised for a keyword density ratio.
The three AI surfaces you need to optimise for
1. Google AI Overviews (formerly SGE)
Google's AI Overviews appear above organic results for the majority of informational queries. They pull from indexed pages and prioritise content that directly answers the query in the first 100–150 words, uses proper heading structure (H2, H3), and is supported by structured data markup. Google has confirmed that E-E-A-T signals (Experience, Expertise, Authoritativeness, Trustworthiness) directly influence which sources are surfaced in Overviews — the same signals that affect organic ranking, but weighted differently and applied at the paragraph level rather than the page level.
2. Perplexity
Perplexity uses a retrieval-augmented generation (RAG) architecture: it runs a live web search, retrieves the top sources, and synthesises an answer with inline citations. To appear as a Perplexity source, your page needs to rank in the top results for the query (standard SEO still applies), and your page content needs to contain the answer in a form the model can extract cleanly — short paragraphs, factual assertions, minimal filler. Pages with high information density per sentence outperform long-form content that buries the key point.
3. ChatGPT (with browsing)
When ChatGPT's browsing tool is active, it fetches pages directly. OpenAI's web crawler, OAI-SearchBot, respects robots.txt and llms.txt directives. Pages that load quickly (under 2s TTFB), return clean HTML, and use semantic markup are easier for the model to parse and cite. Pages blocked behind JavaScript-only rendering or cookie consent walls are effectively invisible to AI crawlers.
The GEO signals that actually move the needle
Based on the research and our own client projects, these are the modifications with the strongest and most consistent impact:
Cite your sources inline
AI models are trained on text that includes attribution. Content that cites a study, a report, or a named expert reads as more authoritative to the model. Include the source name and a hyperlink — not as a reference section at the bottom, but inline in the paragraph where the claim appears.
Lead with the answer
Every section should open with a direct, assertable statement. "GEO increases AI citation rates by up to 40% (Aggarwal et al., 2023)" is better than "In this section, we will explore how GEO affects visibility in AI-generated responses." AI models extract the first clean statement that answers the implied question.
Use precise statistics
Quantified claims are preferred over qualitative ones. "60% of searches are zero-click" outperforms "most searches don't result in a website visit." Specificity signals primary research.
Structure with semantic HTML
H2 for major sections, H3 for subsections, <article> wrapping the body, <time> for publish dates. AI crawlers use heading structure to chunk content into segments they can evaluate and cite independently.
Write FAQ blocks for conversational queries
Conversational AI surfaces handle question-phrased queries by extracting the nearest matching Q&A pair. A dedicated FAQ section using Schema.org FAQPage markup gives both AI and Google's rich results an extractable answer block.
Publish authoritative long-form content
Thin pages under 600 words rarely appear as AI sources for non-trivial queries. Comprehensive pages that cover a topic at depth (1,500–3,000 words) with multiple sections and sub-topics give AI systems more surface area to match against varied query phrasings.
What classic SEO still does
GEO does not replace SEO — it extends it. Google AI Overviews pull from indexed pages. To be indexed, you need crawlability, page speed, domain authority, and backlinks. Perplexity's RAG pipeline still runs a search before it retrieves. If your page ranks nowhere, it is not retrieved. Technical SEO — Core Web Vitals, canonical tags, structured data, mobile responsiveness — remains the foundation on which GEO optimisations are applied.
Think of it as a two-layer stack: SEO determines whether you are in the pool of candidates that AI systems consider. GEO determines whether, among those candidates, your content is the one that gets cited.
How we implement GEO at Nous Frame
On every project we ship, GEO optimisation runs in parallel with technical SEO. Concretely, this means:
- An llms.txt file at the domain root directing AI crawlers to key pages
- Schema.org Article or FAQPage markup on all long-form content
- Opening paragraphs rewritten to answer the implied search query directly in the first two sentences
- All factual claims attributed inline to a named source with a hyperlink
- Headings framed as questions when targeting conversational AI queries
- Core Web Vitals score of 90+ on mobile to ensure AI crawlers receive clean HTML
Sources
- 1. Aggarwal et al. (2023). "GEO: Generative Engine Optimization." Princeton University, IIT Delhi, Georgia Tech, Microsoft Research. arXiv:2311.09735
- 2. SparkToro & Datos (2024). Zero-click search statistics for US & EU. SparkToro Blog.
- 3. Google Search Central. "Creating helpful, reliable, people-first content." Google Developers.
- 4. Google Blog (2024). "AI Overviews in Google Search — how they work and what's next."
- 5. OpenAI. "GPTBot and OAI-SearchBot web crawler documentation." OpenAI Developer Platform.