AI Mode

When you search for something on Google, you normally get a list of links. You click yourself, read yourself, and piece the answer together yourself. AI Mode is a different operating mode of the same search. There, you type your question in full sentences and receive a finished answer text that a computer program has written for you. The links only remain alongside or beneath it as evidence. And you can ask follow-up questions directly, without starting the search over – the mode remembers what you asked before.

What’s at stake for websites and for Google

The internet as we know it runs on clicks. A news site, a travel blog, or a recipe portal earns money when visitors come by. For decades, a large share of these visitors came via Google. But if the answer already appears in the search itself, there’s less reason to click. Experts call this a zero-click search: a search that ends without a visit to another site.

For publishers, this is an economic problem. Their texts are used to formulate the answer, but the advertising revenue stays with Google. Several media companies have therefore filed lawsuits or negotiated payment agreements. Conversely, Google is under pressure from the other side: chat programs like ChatGPT have long been answering questions this way, and many users now go there instead of to a search engine.

On top of that is the question of how Google will make money in the future. Classic search sells advertising space alongside the links. In an answer text, there’s less room for ads. Google is therefore testing advertising that is built into the answer itself. How well users will accept that remains to be seen.

From the question to the formulated answer

Behind the scenes, a so-called language model is at work. This is a program that has learned from huge amounts of text which words typically follow one another. With this, it can formulate sentences that read as if written by a human. At Google, this is the Gemini model family.

On its own, based only on what it has learned, the model would state outdated or made-up things. That’s why your question is first broken down into several smaller search queries. Google calls this process query fan-out, meaning a fanning-out of the question. From the question about an affordable running shoe for wide feet, several individual searches are thus generated. The pages found then serve as material for the model, which builds the answer from them.

This looking-things-up before answering is called, in technical terms, retrieval augmented generation. It reduces the number of errors, but does not eliminate them. The model can summarize sources incorrectly or mix up two things. Such confidently presented false statements are called hallucinations. A common misconception is therefore to take AI Mode for a verified source of facts – for important information, it’s worth checking the linked sources.

Where you actually encounter AI Mode

On Google, AI Mode appears as its own tab next to Images, Videos, and News. So you have to actively select it. This is to be distinguished from AI Overviews: these are short AI summaries that appear unprompted above the normal search results. AI Mode is the more extensive variant with follow-up questions. It was first rolled out in the US, with further countries and languages added gradually.

Similar modes exist among the competition. Microsoft's Bing works with Copilot, the search engine Perplexity has been built this way from the start, and ChatGPT has its own web search. Once you understand AI Mode, you’ll recognize the pattern everywhere.

In business news, you mostly encounter the term in connection with numbers: how many users use it, how sharply traffic to news sites declines, what the computing power costs. In reports about search engine optimization, the question is how a website can still manage to appear in such answers at all. Both areas are currently in a lot of flux.