The tide is turning on distracted driving in America. Almost nobody has noticed.
Traffic deaths have fallen for 15 consecutive quarters. In 2024, 39,345 people died on US roads, the first year below 40,000 since 2020. NHTSA’s early estimate for 2025 is 36,640, down another 6.7% and low enough to erase most of the pandemic-era surge. That is thousands of people alive today who, on the 2022 trendline, would not be.
These gains did not come from AI. They came from stronger laws, better feedback, and the ability to measure changes in driving behavior. AI’s role is in what comes next: helping safety professionals connect risks conventional analysis cannot, intervene earlier, and accelerate a decline still far too slow. First, what changed.
What turned the tide
Phone use behind the wheel fell before the deaths did. Cambridge Mobile Telematics first documented the scale of the problem in 2019, when our analysis of 54 million trips showed drivers distracted by their phones on 37% of trips, and claims data tied roughly one in five crashes to phone distraction, far more than official statistics captured. Advocates and safety experts pressed the issue as distraction climbed to record levels during the pandemic. In 2023 the trend reversed. In 2024, our analysis found US phone distraction dropped 8.6%, a change we estimate prevented some 105,000 crashes and 480 deaths in one year. The mechanism is no mystery: research CMT published with the Governors Highway Safety Association found a driver using a phone is 240% more likely to crash, and Nationwide’s independent analysis of collision claims puts a distracted driver at 3.5 times the crash risk. Reduce the behavior and the crashes follow it down. Safer vehicles and calmer post-pandemic roads deserve a share of the credit; no national trend has a single cause. But distraction is the one major risk behavior that has verifiably fallen, year after year, wherever it is measured.
State laws are doing the heavy lifting, and each new hands-free law is a natural experiment whose result keeps replicating. The four states that went hands-free in 2023 (Ohio, Michigan, Missouri, and Alabama) have averaged an 11.8% reduction in distraction, preventing an estimated 31,000 crashes and 140 deaths, with Michigan leading at 18.7%. In Iowa’s first year under its 2025 law, phone motion fell 6.9%, preventing an estimated 470 crashes, 210 injuries, and three deaths. Ohio Governor Mike DeWine credits his state’s law with 18,000 fewer crashes and 280 fewer traffic deaths. “Distracted-driving crashes are completely avoidable,” DeWine said.
Insurance programs reinforce the laws. Voluntary programs such as Nationwide’s SmartRide and Focused Driving Rewards use trip feedback and discounts to reward focused driving. The evidence that feedback works is randomized and peer-reviewed. In a PNAS trial, University of Pennsylvania researchers gave drivers weekly goals and friendly competition built on trip feedback: handheld phone use fell 20.5% versus control, and 27.6% with modest financial incentives added, reductions that persisted after the program ended, the signature of a habit, not a performance. An earlier JAMA Network Open trial with more than 2,000 drivers cut handheld use 15% to 21%. In 2025, the AAA Foundation for Traffic Safety published the first experimental demonstration that this feedback-and-rewards model improves safety across behaviors, with gains that held after the feedback stopped, consistent across age, sex, and race.
Why the turn is hard to see
If the tide is turning, why has almost no one noticed? Partly because the roads are still dangerous. More than 36,000 deaths a year does not feel like progress, and it should not; less bad is not good. And partly because the official instruments cannot show the change.
The nation’s official measure of driver phone use, the National Occupant Protection Use Survey, works like this: trained observers stand beside the road in daylight and count what they can see. They are stationed at intersections from 7 a.m. to 6 p.m. over several weeks each summer and record, as the Insurance Institute for Highway Safety has noted, only drivers stopped in traffic where an observer can stand. A phone held low never appears in the count. Nothing after dark is counted. Its latest edition reports that handheld calling has halved in a decade, yet visible device manipulation jumped from 3.0% to 4.5% in one year. Winning or losing? The curb cannot say.
The daylight blind spot matters most after sunset. In 2023, The New York Times asked “Why are so many American pedestrians dying at night?”, documenting a uniquely American pattern: pedestrian deaths climbing since 2009, three quarters after dark, with smartphones among the prime suspects. That curve has bent. The Governors Highway Safety Association reports pedestrian deaths fell for a second straight year in 2024, then dropped 11% in the first half of 2025, the largest decline in 15 years of tracking. Even in that half-year, pedestrian deaths rose in 24 states; the tide is turning, not turned. An instrument that observes only daylight could not have seen the rise and cannot see the recovery.
Crash statistics are no sharper. NHTSA attributes 3,275 deaths to distraction in 2023, about 8% of the toll, a figure nearly everyone in road safety regards as a floor, since distraction must be proven from a police report and drivers rarely volunteer it. When NHTSA researchers re-analyzed crash data with a validated imputation model that estimates the distraction police reports miss, they concluded distraction contributed to about 12,400 deaths a year, roughly 29% of the toll and $98 billion in costs, on the same scale as drunk driving. Officially, distraction looks like a minor character in American road deaths. Corrected, it is a co-lead. IIHS President David Harkey put it plainly: the field lacks “good information about where, when and how drivers are using their phones.”
The measurement moved inside the vehicle
The turn is visible because measurement no longer depends on the roadside. For drivers who choose to participate in safe-driving programs, a smartphone’s motion sensors generate rich signals about each trip. Privacy is a design requirement here: for public-safety analyses such as these, telematics measurements are anonymized and aggregated to show how a population’s driving changes, never to follow an individual. It sees the whole trip, not a glimpse; the night as well as the day; the county as well as the country. The measurement itself is not AI, any more than a thermometer is a physician. Telematics is the instrument: it records what happened. AI can manufacture confidence faster than reality can validate it. Telematics supplies the constraint: a record of what actually happened on real roads.
Software counts simple physical events in those signals: a phone handled while the vehicle is moving, a screen tapped, a call held to the ear. In this public-safety use, it is counting rather than judging, a pedometer for phone handling, aggregated across millions of drivers into a picture of what a state’s drivers did, together, after a law changed. The aggregate analysis does not score or make decisions about individual drivers. The method holds up to independent scrutiny. When IIHS scientist Ian Reagan tested telematics against the federal roadside survey, the trends aligned, but telematics saw far more: drivers on their phones over 3% of total driving time. “Telematics data offer a lot more nuanced information than we have now,” Reagan observed, “because the information is collected all day long, from a large number of drivers over the entire duration of their drives.”
That is how Iowa got its answer in months, not years. Nationwide’s analysis of anonymized data from Iowa drivers who opted into its safe-driving programs tracked every phase of the rollout: screen tapping down 5.9% during the education-first warning period, down 13.6% once the law took full effect, handheld calling down 9.1%, and roughly 140,700 fewer phone-motion events than pre-law behavior predicted. Awareness produced real change before a single ticket was written, and the gains grew from there.
AI is about the future, not the past
To be clear about credit: AI did not turn this tide. Lawmakers did, advocates did, insurers did, and above all drivers did. But the playbook that produced this turn will not, on its own, deliver the next meaningful reduction. The gains are uneven, a turning tide can turn back, and this pace of progress still leaves tens of thousands dying every year. Road safety has never lacked hypotheses; it has lacked ground truth rich enough, and feedback loops fast enough, to test them. That is where AI comes in: understanding risk at a level no clipboard or crash report ever allowed, the conditions in which crashes occur, how behaviors combine, where they cluster, which interventions work, and for whom.
The foundation for that future is already taking shape. Even without AI, continuous telematics analysis is revealing relationships that roadside surveys and crash reports miss. This April, IIHS published an analysis of nearly 600,000 trips that overturned a long-standing assumption: drivers do not use their phones mostly at low speeds. In free-flowing traffic the opposite is true, with phone handling rising 12% for every 5 mph over the limit on highways, strongest on the fastest roads. Two deadly behaviors compound where the consequences are worst. Context shapes risk in other ways: phone use spikes during rush hour and at school drop-off, and drivers are 50% more likely to handle phones stopped at intersections than moving on an open road. Findings like these become the next generation of risk factors, and education aimed at the moment’s risk spikes rather than at drivers in general.
For drivers, this future should feel like help, not homework. AI can make the feedback more personal and timely: a heads-up that the road ahead is where distraction and speeding tend to mix, or encouragement after a week of focused trips. Separately, voluntary programs can recognize and reward measured improvement. The driver stays in charge, opts in, and collects the benefit twice, in safety and in savings.
AI can also build on that measured foundation: connecting driving behavior with roadway conditions, infrastructure, and other context; spotting relationships people would miss; prioritizing attention; and helping the people responsible act earlier. Picture an intersection where a tree has grown over a stop sign. Traditionally it becomes a “problem intersection” only through its crash history. AI can help a safety agency connect what was previously unconnectable: imagery flags a potentially obscured sign, aggregated driving data shows a pattern of late braking, and AI surfaces the combination for human review, giving staff evidence to inspect the intersection while its crash record is still zero. The problem can be solved with a chainsaw, not a citation. That is the idea behind road-intelligence platforms such as CMT’s StreetVision, and it is where AI moves from the background of road safety to the forefront. Vehicles benefit too: IIHS researchers have pointed out that the same smartphone sensors could deliver low-cost safety warnings to millions of older vehicles, while population-scale behavior data gives designers a real-world picture of the driving their systems must handle. In Washington, the bipartisan Roadway Safety Modernization Act would let state highway safety programs use telematics and predictive analytics to find and fix dangerous roads, and NHTSA’s Pathways to Safer Streets plan targets distraction as a leading cause of traffic deaths.
Keep going
This is what a turning tide looks like: laws grounded in measurement, education aimed where it matters, feedback that builds habits, a death toll falling quarter after quarter. It is a turn, not a victory. More than 36,000 people will still die on American roads this year, and nearly a third of states have no hands-free law. But for the first time, we are working with information worthy of the problem. We no longer have to wonder whether we are saving lives. We can count them, and what we do next decides how many more.
