Answer Engine

Ask a traditional search engine “what’s the best way to store fresh basil?” and you get ten links to read. Ask an answer engine and you get a paragraph that says what to do, with small citation markers pointing to the pages it drew on. Perplexity, Google’s AI Overviews and AI Mode, ChatGPT search and Microsoft Copilot search all work this way.

Under the hood an answer engine is retrieval-augmented generation over the web. It turns the question into one or more searches, fetches and reads the top results, and has a large language model write an answer grounded in those pages, citing them. OpenAI’s WebGPT research system showed the pattern in 2021 by letting a model browse and quote pages before answering.

The design changes who does the checking. A list of links leaves judgment to the reader. A confident paragraph invites the reader to accept it. Liu, Zhang and Liang audited four generative search engines and found answers that read well but were often unsupported: on average only about half of generated sentences were fully backed by their citations. In episode 54 at 1:05:26, Dan Klein frames the general problem: “If you have a technology that always sounds right, but often isn’t, it becomes very hard as a user to know what to trust.” That is his view of language technology broadly, not a measurement of any one product. The episode was sponsored by Galileo (sponsor mention).

Why it matters (editorial synthesis): answer engines are becoming the first stop for many questions, which is convenient and also shifts traffic away from the sites the answers come from. For readers, the practical habit is to open the cited source for anything that matters and check that it actually says what the answer claims, especially for health, money and legal questions. For publishers, it is the reason generative engine optimization has become a topic.