This summer, many internet communities have been mocking a new feature on Google: automatically generated AI responses. The feature is supposed to help internet users find the information they need more quickly. However, in far too many cases, the responses are incorrect and block users from instantly seeing the actual search results. Not that many of us care: according to one study, significantly fewer people click on Google search results when an AI summary is presented.
Beyond causing a few new memes on the internet, AI summaries are part of a larger symptom, according to tech critic Ketan Joshi. In the wake of the AI hype, more and more companies are equipping their services with AI features based on unreliable language models. As we have explored in recent months, these systems are also extremely energy-intensive, consume vast amounts of water and have serious consequences for people in countries in the Global South.
And yet, finding ways to avoid AI features is becoming increasingly difficult.
Digital ‘walled gardens’ are restricting internet users
We hate to admit it, but it can be quite practical to choose one company for all your digital services. Google, for example, offers an impressive number of services that are compatible with one another. Users can upload documents and share them directly via email. Or they can embed presentations into a collaborative document and edit them together with colleagues. For several months now, Google’s Gemini language model has also been helping with this, summarising emails or attempting to create presentations or even podcasts directly from documents.
Apple allows users to copy text from their smartphones and paste it onto their supposedly carbon-neutral desktop computers. While this has many advantages, it also comes with a major disadvantage: in recent decades, more and more closed systems have developed in the digital world.
Can you disable Google’s AI overviews?
Google itself does not offer the option to disable AI summaries in search results (as of September 2025).
However, there are two ways to stop seeing automatic responses:
On the one hand, you can use an alternative search engine such as Ecosia or DuckDuckGo.
On the other hand, there are several browser plugins that hide the response window. To install them, simply search for ‘Hide Google AI’ in your browser’s plugin directory.
However, it’s not confirmed whether a plugin stops the Gemini search query taking place or simply stops you seeing it. So if you want to be on the safe side, you should use a green search engine.
The internet has already coined a term for this: a ‘walled garden’. Imagine a flowering garden enclosed by solid walls. We can enjoy these walled gardens as long as we don’t need any services or devices operated by different providers. But if we want to connect an Android mobile phone to a MacBook, for example, we come up against rock-hard—albeit artificial—walls.
These closed ecosystems restrict users. And they also become a problem when the providers of our gardens suddenly make changes.
The current AI boom illustrates this very well. Users of Google’s online services have little choice but to use Gemini. Google’s AI summaries can only be hidden using third-party browser plugins. This presents a major dilemma for those who want to navigate the web in the most CO2-efficient way possible.
Why is it problematic that AI is being forced upon us?
In August 2025, Google published a study on the energy consumption of its current Gemini language model. In this study, the company examines the energy consumption of a single search query in its system. The reported power consumption is extremely low. So low, in fact, that it is negligible compared to other everyday activities and online activities.
However, Ketan Joshi argues that this analysis was deliberately one-sided. First of all, Google’s analysis is very one-dimensional. This is because it deliberately focuses on the most economical use of language models. At the same time, Google does not consider the mean value, but rather the statistical median.
“[The study] provides an estimate of the impact of a single text query. But not only that—it focuses on the impact of the “averaged” text task. That is, from the total number of queries sent to Gemini, the one in the middle was selected.” – Ketan Joshi
Google justifies the use of the median instead of the mean by arguing that it protects the evaluation from distortions caused by extreme values. Statistically speaking, this is a legitimate approach. And Google also states that the data basis is distorted. However, Ketan Joshi criticises Google for not revealing in which direction the data is distorted and to what extent.
Without this information, it is extremely difficult to correctly interpret Google’s results. It’s also questionable that Google focuses on the most energy-efficient task within its system because simple search queries only account for a small part of the work done by language models. But this is how Google, Meta and others present their technologies.
AI automisation increases the resource consumption of language models
Google’s AI capabilities are heavily focused on generating video content. This is a use case that requires much more computing power than for individual search queries (the scenario Google presents in its energy consumption study). In addition, many queries to language models such as Gemini aren’t active but passive, as Google’s AI-overview feature automatically pops up when users search.
AI data centres consume so much energy that they are sometimes powered by illegal gas turbines. This has a detrimental impact on the environment and the local population. This brings us back to our initial observation. More and more ecosystems in the digital world are designed in such a way that we can’t simply leave. We also can’t change or customise these services because they’re offered and developed by individual companies. It’s precisely these companies that are now adding unreliable and energy-hungry AI features.
And if you want to avoid AI features, your hands will be increasingly tied.
AI hype shows that we must protect open-source and decentralised services
Digital sovereignty means, among other things, being able to move freely and independently on the internet. It also means being able to decide for or against a programme or service when new system functions are introduced. However, this is no longer possible in the walled gardens of the internet, where more and more Big Tech companies are capitalising on their lucrative business model.
This development, which has reached a new level with the AI boom, clearly shows the importance of open-source licences, decentralised networks such as Mastodon and democratic communities on the internet.
Let’s imagine that a potentially harmful new technology is integrated into software that is designed to be open-source. Here, users would have the option of deleting the critical functions in the software’s source code and creating their own customised version. Since the software is made available collaboratively, other interested parties could also switch to the new version.
The same applies to decentralised networks. In an interview with RESET in April 2025, Andy Piper from the decentralised short message service Mastodon assumed that decentralised networks would not “implement energy-intensive AI integration” in the future. His assumption, until now, has proven correct. And even if there were AI functions in Mastodon, Pixelfed or PeerTube, users could redesign and customise the services accordingly.
Collaborative software is generally protected by open-source licences and copyright law. However, it is precisely this that makes it vulnerable when taken over by large companies.
How can we ensure a green digital future?
Growing e-waste, carbon emissions from AI, data centre water usage—is rampant digitalisation compatible with a healthy planet? Our latest project explores how digital tools and services can be developed with sustainability in mind.
The Android operating system, for example, started out as an open-source operating system before it was bought by Google. However, Google is now making it increasingly difficult for developers to offer their own versions of the operating system. This makes it especially hard to continue operating older smartphones, as the Topio association revealed to us in an interview.
We can extend the life of technology with open-source software
Thanks to free licences, however, software can be designed in a much more resource-efficient way. This is one of the keys to extending the life of technology. Together with better customisation options, this results in many advantages that make open-source programmes exciting.
In our guide to switching to decentralised social media networks, you can find out how to join an alternative to Instagram and co. AI- and Big Tech-free messengers also exist, and the ZOOOM initiative in Europe shows how intellectual property can be combined with the open-source concept.
