Between March and April Google began rolling out their AI Overview search results in the UK. These have largely already been rolled out elsewhere, including across the US and Europe.
Speaking to my UK based SEO clients about this, most of them are (quite surprisingly) happy with the feature - at least as a consumer - they find it saves them time, and makes it easier to digest information.
On the other hand - in April I saw a 30% drop in traffic, in average, across all my clients, when compared to March.
As a consumer I would probably agree that the AI Overviews are quite handy - but - knowing a little around the added environmental costs of this feature, I don't feel the benefits outweigh those costs. And that's before jumping into the ethical considerations of how Google (and AI in general) have literally stolen the content that humans have written, and have basically monetised it in the process.
Google might not be selling access to it's AI Overview feature, but of course it's main product is its search engine - so improving searcher satisfaction is a key goal for them - which likely helps to grow their market share and keep their shareholders happy. Most of Google's (their parent company is Alphabet) revenue comes from advertising and other channels anyway.
It remains to be seen how they will monetise AI (will they allow advertisers to feature here in the Overviews at some point?)
What are the environmental costs of running a Google AI Overview search?
[Disclaimer here that I have not seen any officially confirmed figures on this matter. Perhaps they will only come if someone at Google ever releases this data publicly (and I'm not convinced that will happen any time soon). I will of course update this article if I see any such data come available! Please do reach out to me if you find any data on this.]
Speaking to people within the environmental and sustainability tech sector, estimates put AI Overview searches as being 10x more costly than a traditional search result.
For context - an average (traditional) Google search query is responsible for 0.192g EqCO2 (gram equivalent of CO2) according to a Greenspector study in 2020 (via the brilliant Gerry McGovern).
Gerry continues that for traditional Google queries "this is the equivalent carbon emissions of a 1 meter journey in a light vehicle. That alone might not sound much - but when Google gets over 80,000 requests per second, that's the carbon equivalent of a 6.7 million km journey (1, 325 tons of CO2 every day!)"
So, if we scale up accordingly for Google's AI Oversearch feature, each search query would have an estimated carbon of 1.92g CO2. As you can imagine... this is a huge increase for every single query made using AI.
However - it is very likely that Google is including some form of query-caching here, so that instead of recalculating the answer to queries as they've made by a user, they can instead refer to the cached version, thus saving on costs. Hat tip to Chris Adams, Author of the Green Web Foundation who shared that Google's own LLM models that are available for public use do include query caching so it's very likely Google will be making use of this feature within their AI results - although we can't say this for sure.
For fun* I asked Google this exact question: What is the estimated environmental cost of google's ai overview searches, in terms of carbon and water? and below is the AI Overview response I got:
Google's AI overview searches, like those using generative AI, have a significant environmental impact. They contribute to increased carbon emissions due to the energy-intensive nature of AI model training and operation. Additionally, AI's power requirements necessitate substantial water consumption for cooling data centers. While specific numbers for AI overview searches are difficult to isolate, Google's overall emissions have risen significantly due to AI.
Elaboration:
Carbon Emissions:
AI models require immense computing power to run, leading to higher electricity consumption and increased greenhouse gas emissions. Researchers have estimated that one ChatGPT query consumes roughly as much electricity as a light bulb for 20 minutes, and this adds up exponentially with millions of users.
Water Consumption:
Data centers need large amounts of water to cool down servers, and the increasing demand for AI processing is driving up water use. Google's data centers in the US alone consumed billions of liters of fresh water in 2021 for cooling.
Google's Overall Emissions:
Google's own reports indicate that AI is a major contributor to the company's increased greenhouse gas emissions, with a 48% rise since 2019. This highlights the broader impact of AI's energy needs on the environment.
*I will plant some trees later to compensate for carrying out this AI search!
And when considering water usage, which is now a massive environmental consideration for all forms of AI usage, we can likely extrapolate based on the estimated costs of other AI LLMs including ChatGPT. It is commonly believed on the web that one query on ChatGPT uses over 500 milliliters of water (over half a litre!) but this figure has since been disputed. It is likely to be far less than this for individual queries made on ChatGPT, although that shouldn't make it's usage environmentally sound.
So, it's hard to determine the exact water cost here for Google's AI Overview feature - it's unlikely to be as high as 0.5 liters of water per search query - and if I can find a reliable figure to cite here I will.
Water usage is required for AI usage because the data centers used to power these calculations use up a lot of energy, and this produces a lot of heat, and so water is used as a cooling mechanism.
Sometimes this water is being taken from local communities, being diverted from agricultural usage or from drinking water for humans - and all for what - helping users in the Global North (eg people like me) to get the answer to their question faster, saving them clicking-through to a website to read the answer for themselves? Is it really all worth it?
How do we disable Google's AI Overview search feature?
If you add -ai to the end of your query it should prevent AI results from being returned.
So - your Google query should read "how to boil an egg -ai".
Hat tip to sustainability specialist (and former SEO consultant!) Adam Bastock of Small99 for sharing this tip recently on LinkedIn 💪
What's frustrating for Google users is that there's not a more UX friendly way of disabling or opting out of AI powered search results. Maybe there should be a public campaign aimed at getting Google to allow us to disable it permanently if we want to?
Perhaps if enough people start adding -ai to the end of their search results Google will start getting the idea.
Previously people had suggested that as a work-around, if you swear within your query, this can prevent the AI search results from appearing (they will have a profanity filter built in). I'm not sure how healthy this will be on our language usage after longterm use...
Another suggestion was go to within the Google search options and select a specific date range for your results, ending in December 2024, but of course this will mean your search results won't include pages that have been published so far in 2025, which again isn't ideal.
From my own usage AI seems to kick-in when I am running a super specific, long tail query (when perhaps there wouldn't be great results to serve me), but even then sometimes I will also see AI results for generic queries, so who knows?
As someone who is quite strongly against most usage of AI - the most frustrating element here is that the use of AI is being widely forced upon us - there is no kind of easy opt out here - and when you are aware of the massive environmental costs of AI (compared to the minimal gains received), you can probably understand the frustration many users have.
Perhaps this is a great time to find a better, more privacy-focused and ethical alternative to Google?
A few more ethical Google alternatives here would be:
Brave - a popular privacy-focused search engine
Ecosia (they plant trees based on the profits - this appears to be powered by Yahoo)
DuckDuckGo - again, very privacy focused and no search personalisation
Qwant - another privacy focused search engine.
Final thoughts - What is the real (hidden) cost of our AI adoption?
I don't like how the environmental costs of AI usage are being purposely obscured from the end user. I think it is incredibly irresponsible for tech companies, especially those that are used by billions of people around the world on a daily basis, to suddenly implement such a feature without the ability to disable it.
I'm hoping that when people become aware of these energy and water usage costs associated with AI - and this applies to ChatGPT and other AI tools - that they either curb their usage or find alternative ways of doing what they want to do.
For many that might mean finally making the switch to a more ethical search engine and browser - one which doesn't force it's users into adopting AI search results.
