A Google AI trial rerouted hundreds of American Airlines flights from the US to Europe, reducing contrail formation that contributes to global warming. The 17-week study showed a 62 per cent drop in visible contrails for optimised routes. Overall contrail reduction across the trial was 11.6 per cent.
Google researchers, led by Dinesh Sanekommu, conducted a randomised control trial with over 2400 American Airlines flights flying eastward from the US to Europe. The trial ran for 17 weeks from January to May 2025, focusing on night flights when contrails have a stronger warming effect than during the day, when they can reflect sunlight and cool the planet slightly. Contrails, formed by aircraft engine soot in ice-rich upper atmosphere regions, are estimated to cause more warming than the planes' CO2 emissions. An AI contrail-forecasting tool predicted these regions using detailed weather data and suggested alternative low-contrail routes in flight-planning software for one group of flights, while the control group received no such options. Of 1232 flights offered the AI-optimised routes, dispatchers selected them for only 112 due to concerns like cost or safety. Satellite imagery analysis showed a 62 per cent lower amount of visible contrails for those optimised flights, with an overall 11.6 per cent reduction when including all flights in the test group. The warming effect dropped by 13.7 per cent for the suggested-route group and 69.3 per cent for actual optimised flights, with no difference in fuel use. “It validated the thesis of, if we could figure out how to safely and correctly integrate into the flight planning process, then this is a scalable route to consider contrail avoidance across many flights,” said Sanekommu. Edward Gryspeerdt at Imperial College London noted, “This is probably the best you can do, at least with the tools we have at the moment,” adding that even a 10 per cent reduction would be significant, though scaling up faces flight-planning challenges. The study is detailed in arXiv DOI: 10.48550/arXiv.2603.06909.
