AI Search Engine.

A classic search engine delivers a list of pages that a person has to open and evaluate themselves. An AI search engine takes over this step: it searches for suitable pages, reads them and summarises the answer in its own words. The sources are given as links or footnotes. Follow-up questions are possible in a dialogue without searching again.

Several steps run in the background. The question is interpreted and often broken down into several sub-searches (query fan-out). For each sub-search, results are retrieved from an index, either its own or a third party's, such as Bing or Google. A language model evaluates the passages found and formulates an answer based on these sources. This basic principle is called retrieval-augmented generation and is meant to reduce hallucinations, but it does not prevent them completely.

The best-known offerings differ in origin and focus. Perplexity is built as a standalone answer engine and shows sources particularly clearly. ChatGPT has a search function that retrieves current pages when a question requires it. Microsoft Copilot is built on the Bing index. Google integrates AI answers directly into search, as AI Overviews above the results and as AI Mode for detailed dialogues.

For companies, the question shifts from 'What position am I in?' to 'Am I named or recommended as a source?'. Some users no longer click through after an AI answer (zero-click search). Those who are named, on the other hand, gain trust, because the recommendation comes from a seemingly neutral authority. This can be measured through mentions for defined questions, for example as share of model or citation rate.