Fewer clicks on the web? EU Commission investigates Google's AI overviews
Google's AI overviews show more content directly in search, resulting in less traffic and advertising revenue for publishers. The EU Commission is investigating
(Image: YueStock/Shutterstock.com)
The EU Commission has its sights set on the new "Overview with AI" function in Google Search. The Brussels-based government institution is currently examining how "Google AI Overviews" works in practice and how the continuous texts generated by artificial intelligence in response to more complex search queries interact with EU copyright regulations, Commission spokesperson Thomas Regnier told Euractiv magazine. The stricter competition requirements under the Digital Markets Act (DMA) for "gatekeepers", the additional platform requirements under the Digital Services Act (DSA) and the older rules against unfair competition could also be relevant.
But that's not all: according to the report, the Commission is also investigating whether the AI overviews are compatible with the requirements to protect media diversity under the European Media Freedom Act (EMFA). According to this, citizens must, among other things, "have access to a wide range of editorially independent media content". Platform operators are required to guarantee this.
Google introduced AI Overviews in Germany and other European countries – with some delay compared to the USA – at the end of March. Publishers and independent smaller content producers in particular see it as problematic that users are often likely to be satisfied with the AI overviews and no longer go to the source's website for the information. With the clicks, site operators also lose the opportunity to earn money with advertising on their own site. This traditional online business model is also being challenged by AI searches from ChatGPT, Perplexity and Microsoft's Copilot.
Drastic drop in traffic for publishing sites
Many publishers have seen a drastic drop in traffic following the introduction of the AI function. According to the financial service Bloomberg, Google acknowledged this effect in October at a meeting with 20 website creators behind closed doors. In a submission to the British government, the BBC complains that AI intermediaries pose a threat to its brand. They would not represent the value of the public broadcaster's reporting or would summarize articles inaccurately. Such hallucinations are a known problem with systems using generative AI. Google's AI Overview has repeatedly attracted attention for its highly questionable summaries.
According to Euractiv, a high-ranking Google manager signaled at the beginning of April that the number of search queries had increased following the introduction of AI Overviews. This could also benefit publishers. However, the insider did not want to disclose click rates, as the figures were constantly changing.
According to the EU Copyright Directive, protected works may be used for research purposes under an exception for text and data mining (TDM). "AI-generated summaries that reproduce protected content can fall under the TDM exceptions", explained Regnier. However, the material must be legally accessible and not subject to a reservation of rights by the rights holder.
No opt-out from AI overviews possible
Authors and exploiters who want to prevent TDM for their works available online can, in principle, reserve the right of use themselves. Such an announcement is only effective if it is made "in machine-readable form". This usually means in the robots.txt file. However, according to Google, rights holders cannot currently exclude inclusion in AI summaries – the clause only applies to AI training. According to the US company, the separation between AI search and normal search will generally become less important over time.
Google states that the rejection of AI training has no influence on the visibility of the corresponding website in the search results. However, news producers still fear that they will slip down the hit lists if they are excluded, reported Renate Schroeder, Director of the European Federation of Journalists, Euractiv. She advocates a high level of granularity in the disclosure of "sufficiently detailed" summaries of training data prescribed in the AI Act so that rights holders could more easily claim compensation for the use of their content. Google argues the opposite. According to a study commissioned by the Copyright Initiative, the reproduction of works using models for generative AI generally constitutes copyrightable reproduction and is therefore illegal.
Videos by heise
In a preliminary decision, the Commission has found that Google has violated the DMA by giving preferential treatment to its own products in search engine results. However, according to Regnier, these preliminary findings do not relate specifically to AI overviews. These were only introduced after the decision was published. However, the same principles also apply here. Just one week before the introduction of AI overviews in Europe, Google published an experiment according to which the deletion of content from European news publishers from search queries had "no measurable impact" on the company's revenue. This obvious demonstration of power angered publishers. According to them, Google benefits from the credibility and reliability of their content, even if the direct revenue in the Group's news business was relatively low.
(nie)