Glossary · Beginner
What Is a Generative Engine? Definition and Example
A generative engine is an AI-powered search or assistant system that writes a new answer to a query by retrieving information and combining it with a large language model, rather than returning existing documents. The term was popularised by the 2023 paper that defined generative engine optimization (Aggarwal et al., arXiv). ChatGPT, Gemini, Perplexity and Google's AI features are common examples.
What are its parts?
| Part | Role |
|---|---|
| Retrieval | Finds relevant pages from an index or the live web |
| Language model | Writes the answer |
| Citation layer | Attaches sources to claims |
| Training data | Background knowledge the model already has |
Why does retrieval vs training matter?
If an answer is retrieved, updating your pages and the sources being cited can change it fairly quickly. If it comes from training data, it changes only when the model is updated. When you audit an answer, check whether it shows sources: that's a clue to which one you're dealing with.
What does it mean for brands?
You're no longer competing for a position on a page but for a place in a sentence. That rewards clear facts about who you are, specific claims, and a presence on the sources the engine trusts for your category.
Example
Ask "Which accounting software suits a UK freelancer?" A generative engine searches, reads several review and vendor pages, and writes "For UK freelancers, popular options include..." with links. The brands it names come from what it retrieved and how clearly those sources describe them.
How do you optimize for one?
The 2023 GEO paper tested changes such as adding citations, quotations and statistics to content, and measured how visible the content became in generated answers (Aggarwal et al.). In practice: make claims specific and sourced, keep facts about your brand consistent everywhere, and measure on a fixed prompt set to see what moves.
Related
Questions
Is a generative engine the same as an answer engine?
They overlap heavily. "Generative engine" stresses that the answer is newly written; "answer engine" stresses that it answers directly.
Where does the term come from?
It was popularised by the 2023 research paper that introduced generative engine optimization (GEO).
Do generative engines always search the web?
No. Some answers come from what the model learned in training, while others retrieve live sources. Many tools combine both.