Definition
Generative Engine Optimization (GEO) is the practice of increasing how often and how prominently a brand, page or fact appears in answers written by generative engines such as ChatGPT, Google AI Overviews and AI Mode, Gemini, Perplexity, Claude and Microsoft Copilot. Where SEO competes for positions in a list of links, GEO competes to be named in the answer or cited as one of its sources.
The term comes from the paper GEO: Generative Engine Optimization, posted to arXiv on November 16, 2023 by Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande. The first version lists Princeton University, Georgia Tech, the Allen Institute for AI and IIT Delhi as affiliations, and the paper was published at KDD 2024 in Barcelona. The authors defined a generative engine as a system that retrieves documents and has a large language model write a response grounded in them, naming Bing Chat, Google's SGE and Perplexity as examples.
To test what helps, the researchers built GEO-bench, a benchmark of 10,000 queries across 25 domains, and rewrote source pages in nine different ways. They measured visibility with two new metrics: Position-Adjusted Word Count (how much of the answer is attributed to a source, weighted toward the top) and Subjective Impression (an LLM judgment of relevance, influence and other qualities). The main findings:
- Adding quotations, statistics and citations to credible sources raised visibility by roughly 30 to 40% on Position-Adjusted Word Count. This is the origin of the widely quoted 'up to 40%' figure.
- Keyword stuffing, the classic SEO tactic, gave little to no improvement.
- A more persuasive or authoritative tone did not help significantly, while fluency and readability edits added 15 to 30%.
- Lower-ranked pages gained most. When every source was optimized, the Cite Sources method raised visibility for pages ranked fifth in search results by 115%, while top-ranked pages lost ground.
- On Perplexity, a live engine, the best methods improved visibility by up to 37%.
The experiments ran on GPT-3.5 Turbo in 2023, so the numbers are directional rather than current ranking factors.
In practice, generative engines decide visibility in two places. During retrieval, the engine rewrites a prompt into several searches (query fan-out), fetches pages and selects passages, so crawl access and coverage of those sub-queries matter. During synthesis, the model chooses which sources to cite and which brands to name, shaped by what it learned in training and by how clearly and credibly each page states its facts. Google's guide to optimizing for generative AI features, published May 15, 2026, confirms that AI Overviews and AI Mode use retrieval-augmented generation on the Search index plus query fan-out. It states that from Google Search's perspective GEO 'is still SEO' and that llms.txt files, content chunking and special schema are not needed. Other engines run their own crawlers and indexes (OAI-SearchBot for ChatGPT search, Claude-SearchBot for Claude), so access and citation behavior differ per platform.
The audience is large. OpenAI said on October 7, 2026 that more than 1.2 billion people use ChatGPT each week, and Google said at I/O in May 2026 that AI Overviews have over 2.5 billion monthly users and AI Mode more than 1 billion.
GEO is measured with a fixed set of prompts run repeatedly across engines, because answers vary between runs and models. Teams track whether the brand is mentioned, its AI visibility score and position, AI share of voice against competitors, the citation share of their own URLs, and sentiment. Mentions and citations move separately: mentions follow reputation and third-party coverage, citations follow crawlable pages that answer the fan-out queries. First-party data has started to appear too. Bing Webmaster Tools added an AI Performance report with citation counts in February 2026, and Google Search Console's generative AI report, which shows impressions only, reached all sites on August 31, 2026. Promptwatch covers the prompt-level side across ChatGPT, AI Overviews, AI Mode, Perplexity, Claude, Gemini and Copilot.
GEO is now the most common name for the discipline, partly because it has a citable origin. Answer engine optimization is older and grew out of featured snippets and voice search, LLMO puts the model at the center, and generative search optimization is a near-synonym focused on search products. Our explainer on what GEO is and the comparison of GEO vs SEO go further.
Examples of Generative Engine Optimization (GEO)
- A project management SaaS runs 150 buying prompts every week in ChatGPT, AI Mode and Perplexity. It is named in 22% of ChatGPT answers but only 6% of Perplexity answers, and the fan-out queries show Perplexity searching for per-seat pricing comparisons the company never published. A public pricing comparison page closes most of the gap within two months.
- A regional bank applies the GEO paper's findings to its mortgage guides: instead of adding keyword variants, it adds sourced rate statistics, short quotes from named loan officers and links to the central bank's data. The guides start appearing as cited sources in AI Overviews for rate questions.
- A B2B security vendor is mentioned often in ChatGPT answers but never cited. Its robots.txt, copied from an old template, disallowed OAI-SearchBot, so its own pages could not appear in ChatGPT search results. After unblocking it, the vendor's comparison pages begin showing up as sources.
- A skincare brand finds AI answers describing a product with an ingredient it removed in 2024. The team traces the claim to an outdated retailer listing and an old review article, gets both corrected, and checks accuracy and sentiment in its monthly prompt runs until the old claim disappears.
