Google heralds an AI era of search as summaries edge out links

Google CEO Sundar Pichai speaks at Google I/O. At the developer conference, everything revolved around the topic of artificial intelligence (AI). Christoph Dernbach/dpa
Google CEO Sundar Pichai speaks at Google I/O. At the developer conference, everything revolved around the topic of artificial intelligence (AI). Christoph Dernbach/dpa

The world's leading search engine being reinvented with the help of AI. Anyone searching Google in the US - and soon in other countries - will soon stop seeing just a list of links and instead see a text summary of search results generated by artificial intelligence.

It remains to be seen what impact this will have on the business model of websites that largely depend on users finding their content via Google.

It is also unclear what this will mean for Google's revenue from advertising in searches, which is a key source of income for the company.

"Let Google do the searching for you" is the new motto for web searches, Liz Reid, the company's head of Search, said at the Google I/O developer conference on May 14.

You can now ask several questions in one search query: For example, not just where Pilates studios are located in Boston, but also how to get there and book a time slot.

Google's dominance as a search enginge was accompanied by the business idea that turned Google into a real money-printing machine: companies pay money for their links to appear next to related search results.

However, chatbots and other AI applications are making it easier to offer an answer to a question, rather than just linking to various websites where the answer might be found.

AI start-ups such as Perplexity AI want to step up to Google in its core area, but so far they have not been able to seriously challenge the market leader. However, Google doesn't want to wait around to see if they can become good enough to rival its own search engine.

Google also made it clear that the company does not want to leave this market to challengers such as ChatGPT inventor OpenAI when it comes to AI assistants in the future.

OpenAI had made headlines the day before with a live demonstration of a version of ChatGPT that now has a voice it can use to have conversations with users while also taking visual information from the smartphone camera into account.

The chatbot gave verbal instructions on how to solve a maths equation that an OpenAI employee had written down on a piece of paper. ChatGPT was also able to interpret moods based on facial expressions.

The competition never sleeps

Google demonstrated similar overarching capabilities of a new AI software called Project Astra - not live on stage, however, but in a previously filmed video.

Google is relying on its in-house AI model Gemini to compete with the popular chatbot ChatGPT and other AI software. Some of the features that could be seen in the Astra demonstration are soon to be added to Gemini, Google manager Koray Kavukcuoglu announced.

Google is also trying to allay the concerns of website owners, who fear that their links being replaced by AI summaries could dry up their traffic - and revenue.

Google executive Hema Budaraju said that previous test runs had shown that there were more diverse links in the AI Overviews than usual - and that users also clicked through to them.

Google had previously introduced a feature in which it is sufficient to circle an object on a photo or words on the smartphone display to search the internet.

At the Google I/O, it was also demonstrated that it's enough to make a short video of a broken record player to get instructions on how to fix the problem on this model. "Google Search is generative AI at the scale of human curiosity," said chief executive Sundar Pichai.

Google sees humanity only at the beginning of AI use

"We have been investing in AI for more than a decade," emphasised Pichai. And yet the technology is only just beginning. Among other things, Google is working on expanding the so-called "context window" - ie. the amount of information that an AI model can analyse simultaneously.

The subscription version of the AI software can currently analyse a PDF document up to 1,500 pages long or a one-hour video at once and answer questions about it. No other chatbot is able to do this, emphasised Google manager Sissie Hsiao. By the end of the year, the aim is to double this figure.

Google is also focusing on "AI agents" - assistants that can independently fulfil tasks involving several steps. They could, for example, take care of all returns or complete the necessary re-registrations when moving house and look up useful addresses in the new neighbourhood.

Pichai demonstrated how the software can now search for all images from the collection of personal photos on request, for example those relating to a child's swimming progress. Finally, the Google boss had Gemini count how many times the term AI was used during the two-hour presentation: more than 120 times.

Every day, hundreds of millions of people start their journey through the web with a Google search. AI start-ups are still trying in vain to break this dominance. Now Google itself is going on the offensive. Andrej Sokolow/dpa
Every day, hundreds of millions of people start their journey through the web with a Google search. AI start-ups are still trying in vain to break this dominance. Now Google itself is going on the offensive. Andrej Sokolow/dpa