A generative engine is an AI system whose output is newly generated text, in contrast to a retrieval system that finds and returns existing documents. The term is most commonly used in the phrase “Generative Engine Optimization” (GEO) to distinguish this new class of AI answers from traditional search.

What makes an engine “generative”

Traditional search engines are retrieval systems: they index a corpus of documents, score each document for relevance to a query, and return a ranked list. The documents themselves are unchanged.

Generative engines work differently. Given a query, they produce new text by predicting what words should come next given everything in their training data (and any retrieved context). The output is composed on the fly - it is not a copy of any single source.

This distinction matters because:

  • The same query can produce different outputs on different runs (the model is not deterministic)
  • The output may contain information drawn from many sources, remixed into a single answer
  • There is no direct URL to attribute the answer to - the model has synthesized it

Generative engines vs retrieval-augmented engines

Many modern “generative engines” are actually hybrid systems. Perplexity, for example, uses a generative LLM but augments it with real-time web retrieval. The LLM generates the prose, but the content is grounded in fetched documents.

This hybrid approach - called Retrieval-Augmented Generation (RAG) - is increasingly common because it reduces the problem of hallucination (the model inventing facts). When a generative engine is RAG-based, being cited by sources that the retrieval system fetches becomes a lever for AI visibility.

Implications for brand visibility

Because generative engines compose their own answers, they can mention a brand without any user explicitly searching for that brand. A user asks “what should I use to track my brand in ChatGPT?” and the engine generates an answer that includes product names from its training and retrieval context.

This is the core opportunity for GEO and AEO: getting your brand into the training data and cited sources that generative engines draw from when composing answers in your category.