AI Overviews

Anyone searching on Google used to get mainly a list of blue links. Since 2024, many search queries now show a text box above the results with a ready-made answer. Google calls this box AI Overviews. The text inside it is not written by a human, but by a computer program that builds a summary from several websites. Next to it are small references to the sources the information comes from. To the user, it looks as if the search engine is answering the question itself, rather than just pointing to pages that might answer it.

What this means for websites and their clicks

The business model of many websites depends on visitors from search. A recipe portal, a travel blog, or a news site earns money through advertising or subscriptions. Both only work if someone actually opens the page. If the answer is already shown in the search result, many users no longer have a reason to click.

Experts call this a zero-click search: the search ends without a click on any result. Studies by SEO firms showed clear declines in visitor numbers for some sites after the launch of AI Overviews, in some cases in the double-digit percentage range. Google disputes such figures and points out that users are asking more and longer questions. Reliable, cross-site data is hard to come by, because only Google itself has all the numbers.

That is why there is dispute and even lawsuits. Publishers and comparison portals argue that Google uses their content without providing the return of visitors. Competition authorities in the EU and the US are examining this. For investors, this is relevant because many media companies get a large share of their reach through Google.

How Google builds the summary

In the background, a language model is at work — a program that has learned from huge amounts of text how language is structured. Such models can formulate sentences, but they don't know anything with certainty about current facts. That's why the model doesn't get its facts from memory, but delivered by the search engine.

The process happens in two steps. First, Google searches for suitable pages as usual. Then excerpts from them are given to the model together with the question, and the model formulates a short text from them. This procedure is called Retrieval Augmented Generation, roughly: generation using previously looked-up material. The advantage is that the answer is based on real sources, which can then be linked.

It is not error-free. Shortly after launch, AI Overviews recommended, among other things, using glue to hold cheese on pizza. The source was an old joke post from a forum. The model does not reliably recognize irony or poor sources. When a language model makes a false claim while sounding convincing, this is called a hallucination.

Where AI Overviews appear and what competitors do differently

The boxes appear most often for explanatory questions: how something works, what a term means, what differences there are. For simple navigational searches, such as a company's name, they usually do not appear. For sensitive topics like medicine, finance, or elections, Google is more cautious and shows a summary less often.

Google is not alone. Perplexity built a search engine from the start that provides only answers with source citations. ChatGPT and Microsoft's Bing also work with similar summaries. The main difference lies in how prominently the classic links still appear alongside them.

In business news, you mostly encounter this term in two contexts. First, with publishers complaining about declining reach. Second, with Google itself, where the question is whether advertising can be sold just as well in such boxes as next to a list of links. For search engine optimization, or SEO for short, the goal has shifted: it is no longer just about ranking first, but about being named as a source within the summary.