American Airlines and Google say an AI-powered flight planning system reduced heat-trapping contrails during a large-scale trial on transatlantic routes, with a reported 62% reduction on flights that followed the suggested routing exactly. The test covered 2,400 flights from the United States to Europe between January and May 2025, integrating contrail-avoidance guidance into normal dispatch and flight-planning workflows as an advisory tool. The results, shared publicly and supported by a mix of peer-reviewed and preprint research, suggest that small route adjustments guided by machine learning can reduce contrail formation without major fuel-cost impacts.
What the Trial Actually Tested
Most public attention on aviation and climate focuses on carbon dioxide, but contrails may account for roughly half of flying’s total warming impact. These thin ice-crystal clouds form when hot engine exhaust meets cold, humid air at altitude. Some persist for hours, trapping outgoing heat much like a greenhouse gas. The catch is that contrails are highly localized and short-lived, which means avoiding them requires knowing exactly where and when atmospheric conditions will produce them.
That prediction challenge is what Google’s AI system was designed to solve. The trial used a control-versus-route design, according to The Associated Press: some flights received AI-generated alternate routes expected to dodge contrail-forming zones, while others flew standard paths. Pilots and dispatchers could accept or reject the suggestions, keeping the system within normal airline safety protocols.
The scale was notable. Prior feasibility work had tested the concept on a handful of flights. This time, 2,400 U.S.-to-Europe flights participated over roughly five months of operations, making it one of the largest randomized controlled trials of contrail avoidance reported to date, per the researchers.
Competing Numbers Tell a Consistent Story
The headline figure of 62% fewer contrails comes with an important asterisk. According to the scaled-trial preprint, that reduction applied specifically to the 112 per-protocol flights that executed contrail-avoidance maneuvers exactly as planned. Flights where pilots partially followed or declined the route suggestions showed smaller gains, which is expected in any real-world operational test.
An earlier peer-reviewed study published in Nature Communications Engineering reported a somewhat different metric: a 54.4% reduction in detectable contrails per flight kilometer for avoidance flights in a smaller randomized test with American Airlines. That study established the basic feasibility of the approach before the larger trial scaled it up.
The gap between 54.4% and 62% does not necessarily signal a contradiction. The earlier study measured contrails per flight kilometer across all avoidance flights, while the newer trial isolated flights that followed the AI routing precisely. Both figures point in the same direction: when pilots fly the suggested routes, contrail formation drops substantially. The Associated Press reported that an estimated warming metric tied to the remaining contrails fell by roughly 69%, suggesting that the avoided contrails were disproportionately the ones with the strongest heat-trapping potential.
Satellite Eyes and Open Data
None of these results would be credible without a reliable way to verify whether contrails actually formed. The measurement system behind the trial uses an automated pipeline that compares individual flight segments to contrails detected by computer vision applied to geostationary satellite imagery from the GOES-16 Advanced Baseline Imager. A technical paper describing this per-flight detection system explains how the approach matches specific aircraft tracks to visible contrail signatures in near-real time, creating a feedback loop that can validate whether avoidance maneuvers worked.
The detection algorithms themselves were trained and benchmarked using OpenContrails, a publicly available labeled dataset hosted on Google Cloud Storage. That benchmark dataset for GOES-16 ABI contrail detection gives independent researchers access to the same labeled imagery, meaning the measurement claims can be tested and challenged by outside scientists. Open access to the underlying data is a meaningful check on corporate self-reporting, though it does not eliminate all questions about model accuracy in varying atmospheric conditions.
Why Rerouting Beats Waiting for New Engines
The aviation industry’s primary decarbonization strategy relies on sustainable aviation fuels and more efficient engines, both of which require years of development and massive capital investment. Contrail avoidance, by contrast, is a software problem. If atmospheric predictions are accurate enough, airlines can reduce warming simply by nudging flight paths a few thousand feet up or down, or a few miles laterally, at little to no extra fuel burn.
That distinction matters because contrails produce warming now, and their effects dissipate within hours or days once they stop forming. Carbon dioxide, on the other hand, stays in the atmosphere for centuries. Cutting contrails offers a way to reduce aviation’s near-term warming contribution while longer-term CO2 solutions mature. The research archive cited by Cornell Tech connects this work to a broader body of climate science exploring short-lived climate forcers and how targeting them can buy time against the worst warming scenarios.
Still, the approach has limits that the trial data alone cannot resolve. Contrail prediction depends on weather forecasts, which lose accuracy beyond a few hours. The 112 per-protocol flights that achieved the 62% reduction represent a fraction of the 2,400-flight trial, raising questions about how often pilots will follow AI-generated route changes in routine operations. And scaling from one carrier’s transatlantic routes to the global fleet introduces regulatory, airspace coordination, and cost-sharing challenges that no single trial can address.
What Stands Between a Trial and Standard Practice
Turning a promising experiment into everyday procedure will require changes at several levels of the aviation system. First is operational trust. Dispatchers and pilots must be confident that contrail-avoidance suggestions will not compromise safety, significantly lengthen flight times, or create conflicts with air traffic control. The trial integrated the AI tool as an advisory layer, and crews retained full authority to reject route changes, which helps explain why only a subset of flights followed the recommended paths exactly.
Embedding contrail avoidance more deeply into airline workflows likely means moving from optional suggestions to default plans, with clear criteria for when deviations are warranted. That, in turn, depends on regulators accepting contrail mitigation as a legitimate planning objective alongside fuel efficiency and on-time performance. Airspace managers would need procedures that allow modest altitude or lateral changes without overloading already busy corridors on popular transatlantic tracks.
Economics will also shape adoption. Even if most contrail-avoiding routes add little or no fuel burn, some will carry modest penalties in distance or time. Airlines will want assurance that competitors are playing by the same rules, rather than unilaterally accepting extra costs for climate benefits that are shared globally. Industry groups and governments may need to define standard metrics for contrail-related warming and incorporate them into climate reporting frameworks, so that carriers can credibly claim and compare reductions.
On the technical side, the AI models and forecast systems must prove robust across seasons, regions, and different types of aircraft. The initial trials focused on specific transatlantic routes, where atmospheric data coverage is strong and traffic patterns are well understood. Extending the approach to other long-haul corridors, and eventually to dense short-haul networks, will test how well contrail predictions hold up in more varied meteorological conditions and more constrained airspace.
Finally, the credibility of contrail avoidance as a climate tool will hinge on continued transparency. The use of open datasets and published methods has already allowed outside researchers to scrutinize the detection pipeline and trial design. Maintaining that openness as more airlines and technology providers enter the space can help prevent exaggerated claims and ensure that reported warming reductions reflect real, independently verifiable changes in the sky.
If those operational, regulatory, economic, and technical hurdles can be cleared, the American Airlines–Google experiment suggests that aviation could start cutting a major slice of its warming footprint long before new fuels and engines arrive at scale. In a sector often criticized for slow climate progress, the ability to trim heat-trapping clouds with software and satellite data alone is likely to remain an attractive, and increasingly hard to ignore, option.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
