If you asked me six months ago where the climate impact of my digital life sits, I would have said data centres. That is the story most of us know. According to researchers at Oxford’s Environmental Change Institute, it is also the least interesting part.
Speaking with Felippa “Pippa” Amanta, a DPhil researcher at the Institute, what began as a conversation about energy consumption raised a different question entirely: not how much energy your apps use, but how much demand your apps are designed to produce.
Before Oxford, Pippa worked in Jakarta on food security and trade policy. She is careful not to frame this as only a wealthy-country conversation. In her living lab work, some households were “good reminders” of digital exclusion: rapid digital transformation risks leaving the most vulnerable further behind.
“Digitalisation is a double-edged sword for climate change,” Pippa told me.
The answer everyone gives
The ICT sector accounts for at least 1.7 percent of global emissions, and data centre energy use is projected to more than double by 2026. Those numbers are real, but Pippa was surprisingly unbothered by them.
An evidence synthesis by Pippa’s team found that some digital habits are already a net positive. Streaming can displace DVDs, video calls can replace flights. So if some substitutions are genuinely cleaner, where is the problem?
The efficiency trap
Consider the robot vacuum. Pippa pointed me to research by Dr Emilie Vrain.
Automation may save effort, but many users resist handing over control. People run other appliances while it cleans. They do extra cleaning in places it cannot reach. The time saved does not vanish; it gets filled.The net effect, as Pippa’s team found, is that efficiency gains are offset by increases in overall consumption.
Researchers have a name for this: rebound. More efficient cars lead to more driving. Cheaper lighting leads to more lit spaces. Build a wider motorway and you get more traffic, not less.
Digital technology follows the same logic, but with a twist. A more efficient boiler does not actively encourage you to use more heat. A digital platform can. Every reduction in friction (faster checkout, one-click ordering, autoplay) also removes the pause where a different choice might have been made. That efficiency gain is captured, by design, as increased consumption.
In most energy systems, rebound is a side effect. In digital systems, it can be the business model.
Recommendation algorithms sit at the centre of how platforms make money. A widely cited McKinsey analysis found that 35 percent of Amazon purchases and 75 percent of Netflix viewing come from product recommendations. A New America case study put Netflix’s figure at roughly 80 percent of hours streamed. These algorithms are not cosmetic features, but rather infrastructure for steering behaviour at enormous scale.
Algorithmic steering, she said, can “ratchet our consumption to an unsustainable level.” Platforms are built to keep users “engaged” and “locked-in” through “a smooth, convenient, frictionless, and personalised” experience.
We measure the electricity an app consumes. We do not measure the emissions that follow because the app made it easier to buy new, order extra, or upgrade sooner. A 2025 commentary in Big Data & Society calls these “algorithmically facilitated emissions”: downstream emissions that intensify because people respond to systems built to promote high-carbon consumption.
If platforms are built to increase demand, a large share of their climate impact is real, significant, and entirely absent from any reporting framework.
The gap
“These indirect effects have not really come up in policy discussions,” Pippa told me. “The focus has been on the direct energy consumption of data centres.”
There are early signs of movement. Germany’s Digital Policy Agenda for the Environment argues that platforms could promote sustainable consumption by prioritising environmentally friendly products in search results. In a separate piece, Pippa and Professor Charlie Wilson argue that AI’s climate risks go well beyond data centre energy.
But the core gap is stark. It is far easier to regulate what a meter can read (energy, water, hardware) than to regulate behavioural steering and induced demand. The climate impact of digitalisation will be shaped by both. Right now, we are only watching one.
What we do with this
Pippa’s research draws on four years of tracking 80 “living lab” households in the UK and surveying 6,000 respondents across three countries. The project is called iDODDLE, funded by the European Research Council, and it is wrapping up this year.
Her first message to students was, “don’t feel guilty about being imperfect and don’t worry about every single small decision.” Some digital substitutions are already lower-carbon. The bigger question is where digital tools push wasteful consumption, and whether we can change the systems doing the pushing.
She suggests a few starting points: a “Digital Cleanup” of duplicated files, choosing second-hand marketplaces over ultra-fast-fashion apps, checking sustainability claims before buying. The Digital Humanities Climate Coalition toolkit offers a practical guide.
But the bigger lever is institutional. Many organisations already report their carbon footprints. There is now an opportunity to include “the carbon footprint of digital tools and services.” Oxford’s own Gardens, Libraries and Museums division has recently started doing exactly this, measuring the carbon cost of its digital operations including generative AI.
For students, that translates into questions worth raising with colleges and departments. What digital services are procured? What emissions disclosures exist? Are sustainability criteria applied to software and cloud choices as seriously as they are to estates, travel, or catering?
“The outcome will depend on the policies and choices we make right now,” Pippa told me. I started this conversation thinking the climate cost of digital life was about electricity. It is not. It is about what “frictionless” was designed to make us do. And right now, nobody is counting it.