AI is changing the web, and that includes search. Google is now responding with more personalization and improved source attribution. The latter might also be related to an investigation by the EU Commission.

AI slop, i.e., AI-generated garbage, threatens to make search results almost unusable. It is easier than ever before for anyone to put content online – AI-generated texts, images, graphics, but also entire websites can be created with a few clicks by AI. The operators of such sites hope to generate advertising revenue with little effort. However, this usually makes searchers rather unhappy due to the quality of the content.

Google is now countering this with a form of search personalization. The feature is called "preferred sources." It allows users to specify which sources and websites they particularly like. These will then be displayed preferentially. Google thus considers the sources in every search query. It is clear that one must be logged in for this.

Initial tests conducted by Google have been very successful. Around 90,000 different websites have been marked as preferred sources. According to Google, users who have set such a source have clicked on the corresponding link twice as often as usual.

(Image: Google)

In addition, Google will in the future mark links to sources for which one has a subscription and display them additionally in a carousel. This should allow people to get more value from their access. However, this will only appear in the Gemini app and subsequently in the AI Overviews and AI Mode. The "preferred sources" will also not be available until next year. In AI Mode, more sources will also be provided than before. These will then receive a brief explanation of why they are intended to be helpful.

Google’s helpful summaries and partnerships

The Web Guide, a way to collect and sort information on a topic, is available for more topics and is faster. This is still an experiment that requires consent. Those who try out the guide will be shown further links and can sort the information. It is intended to facilitate a kind of deeper research. All major internet and AI companies are interested in offering us more information than we actually search for – because they obviously know that we need all of it. This ranges from social media feeds to deep research functions, AI agents, and now search.

Google always talks about wanting to make its own products more helpful. The current blog post also mentions this. In addition to personalization on the searcher’s side, Google also wants to attack on the other side and save the web from AI collapse. The origin of content used for AI answers is already proving problematic. Without clicks on their websites, publishers lack revenue from advertising. Without content from publishers, there are no AI answers to users’ current questions. The EU Commission is already investigating this relationship.

Therefore, Google offers commercial partnerships, "under which we pay for extended advertising rights and methods for providing content such as APIs." Last year, more than 3,000 such partnerships were established with publications, platforms, and content creators. This is to be further expanded. Other AI providers, such as OpenAI, also enter into license agreements with publishers. However, this is limited to a few individuals and is expensive. It is questionable how, for example, the several tens of millions of euros that OpenAI pays Axel Springer can be recouped. In Germany, Google cooperates with Der Spiegel.

One of the experiments that Google wants to expand is AI summaries of articles. One might think that an article already contains the most important information, provided in understandable language and with necessary background information by the creator. The initial experiences with AI summaries of news under Google Discover have certainly been noticed for errors – Google itself admits this with a note on possible errors. Whether this is really helpful is up to each individual to decide.

(emw)

Don’t miss any news – follow us on Facebook, LinkedIn or Mastodon.

This article was originally published in German. It was translated with technical assistance and editorially reviewed before publication.