What is the True Energy Cost of Google Search’s AI Overview?

Over the past two years, Google’s AI Overviews have become part of the furniture of everyday web surfing. 

Where links to websites used to be the first thing an internet user saw upon entering a query on Google, more often than not the world’s most popular search engine now returns a summary, generated by its in-house AI system.

Since launching the function in 2024, these AI-generated summaries have become far more common. In fact, a recent study by software firm Ahrefs suggested that 55% of all Google searches now result in one.

The change has been welcomed by many people, who appreciate receiving concise answers without needing to scavenge through multiple websites.

Nevertheless, behind that convenience lies a growing debate over the environmental cost of turning the world's most popular search engine into an AI-powered answer machine.

The data

By now it is well understood that using an AI platform requires far more energy and water than a traditional query on a search engine. 

Last year, the International Energy Agency confirmed that a single query on ChatGPT consumed 10 times as much electricity as an old Google search.

But with AI now baked into more than half of all Google searches, the energy the average internet user is responsible for is skyrocketing.

In recent years, one of the biggest challenges in deciphering the exact impact of modern internet usage has been the lack of transparency from major tech companies.

However, that changed in August 2025 when Google published what it described as the first comprehensive estimate of the energy required for a prompt submitted through its Gemini applications.

According to the company, a median Gemini text prompt consumes 0.24 watt-hours of electricity, produces 0.03 grams of carbon dioxide emissions and uses 0.26 millilitres of water for data centre cooling.

According to the AI Overview function itself, generating a summary through Google’s search bar uses the same amount of energy as a standard Gemini prompt, which suggests that the internet’s demand for power is growing far faster than many people realise.

The problem of scale

While Google's figures suggest an individual AI interaction consumes relatively little energy, the environmental challenge emerges when those queries are multiplied across billions of searches.

Google itself says it processes five trillion searches each year, meaning even modest increases in energy demand per query can translate into enormous consumption at full scale.

The impacts of AI are already visible in Google's own sustainability reporting.

The company's greenhouse gas emissions were 48% higher in 2023 than in 2019, with Google attributing much of the increase to expanding data centre infrastructure and growing AI-related computing demands. That has grown even more in the years since.

Google has acknowledged in its environmental reporting that reducing emissions may become increasingly difficult as AI is integrated more deeply across its products because of the greater intensity of AI computing workloads.

The wider tech sector faces a similar challenge, with the likes of Microsoft admitting that AI has changed its climate targets.

Researchers have warned that rapidly growing demand for AI services is driving a new wave of data centre construction, increasing pressure on electricity grids and, in some regions, extending reliance on fossil fuel generation.

A question of necessity

The debate surrounding AI Overviews extends beyond energy and water consumption as well.

Publishers, website owners and SEO professionals have reported significant declines in traffic since AI-generated answers began appearing at the top of search results.

To put a figure on it, Ahrefs has estimated that AI Overviews can reduce clicks to websites by more than a third.

At the same time, users are often unable to disable the feature permanently. Instead, they must go out of their way to type "-ai" at the end of each query if they do not wish to receive an overview.

That raises questions about whether consumers should have greater control over AI-powered search experiences, particularly when those experiences carry additional environmental costs.

Transparency is only the beginning

Google's decision to publish energy data for Gemini marked an important step towards understanding AI's environmental footprint, but actually addressing that consumption is still a huge problem for the firm.

The environmental impact of a single query may appear negligible. The impact of trillions of them is another matter entirely.

Still, as the old sustainability adage suggests: "You cannot manage what you cannot measure." If Google can get a handle on the appetite of its AI model, the mitigation can begin.