I. The Regulatory Framework

The Federal Trade Commission Act was enacted in 1914, creating an independent agency with broad authority to prevent “unfair methods of competition” and, as subsequently amended, “unfair or deceptive acts or practices in or affecting commerce.”¹ The statute does not define “deceptive.” Congress left that task to the Commission and the courts.

On October 14, 1983, the Commission issued its Policy Statement on Deception in a letter to the Honorable John D. Dingell, Chairman of the House Committee on Energy and Commerce. The statement synthesized decades of case law into a three-part framework. A practice is deceptive if: (1) there is a representation, omission, or practice that is likely to mislead the consumer; (2) the consumer is acting reasonably under the circumstances; and (3) the representation, omission, or practice is material, meaning it is likely to affect the consumer’s conduct or decision with regard to a product or service.²

Separately, the Commission’s Advertising Substantiation doctrine, articulated in the 1984 Policy Statement appended to Thompson Medical Co., 104 F.T.C. 648, requires that advertisers possess “a reasonable basis for advertising claims before they are disseminated.” An advertiser who makes an objective claim about a product without possessing competent and reliable evidence to support that claim has engaged in an unfair and deceptive act or practice within the meaning of Section 5.³

The Commission has enforced these principles with considerable vigor. It has pursued companies that overstated the accuracy, reliability, or efficacy of their products across virtually every industry in American commerce. It has obtained consent orders against manufacturers of dietary supplements, fitness equipment, educational software, automobile additives, and air purifiers. It has, in at least one memorable instance, required a shoe company to pay forty million dollars for claiming that its sneakers would tone the wearer’s leg muscles, on the ground that the company lacked competent and reliable scientific evidence to support the claim.⁴

It has never, in the 112 years of its existence, examined whether the claim “Saturday: 74°F, Partly Cloudy, 20% chance of rain” is supported by competent and reliable scientific evidence when the Saturday in question is eighty-seven days away.

II. The Mathematical Proof

In 1961, Edward Norton Lorenz, a mathematician and meteorologist at the Massachusetts Institute of Technology, was running a numerical weather simulation on a Royal McBee LGP-30 computer. To save time, he restarted a run from the middle by manually entering the intermediate values. He entered 0.506 instead of the full value 0.506127. The result should have been negligible. Instead, the simulation produced an entirely different weather pattern.⁵

Lorenz published his findings in 1963 in a paper titled “Deterministic Nonperiodic Flow,” which appeared in the Journal of the Atmospheric Sciences, Volume 20, pages 130 through 141. The paper demonstrated that certain nonlinear dynamical systems are exquisitely sensitive to initial conditions: infinitesimally small differences in starting states grow exponentially over time, producing fundamentally different outcomes. The atmosphere, Lorenz proved, is such a system. A difference in the third decimal place of a single variable was sufficient to produce a completely different weather pattern within a matter of simulated days.⁶

The implications for weather prediction were immediate and devastating. If the atmosphere is a chaotic system in which tiny perturbations amplify exponentially, then any forecast based on measured initial conditions contains errors that will grow until the forecast is indistinguishable from a random guess. The question was not whether this would happen but how quickly.

In 1966, Jule Charney, chair of the Global Atmospheric Research Program, convened a committee to determine the feasibility of extended-range weather prediction. The committee ran perturbation experiments using the Mintz-Arakawa general circulation model and concluded that small errors in temperature and wind patterns doubled approximately every five days. Given realistic estimates of observational error, the committee calculated that deterministic weather forecasts would lose all useful skill after approximately two weeks. The committee’s report stated the conclusion directly: “the limit of deterministic predictability for the atmosphere is about two weeks in the winter and somewhat longer in the summer.”⁷

This result has been confirmed repeatedly in the sixty years since it was first proposed. In 2019, Fuqing Zhang and colleagues at Penn State University published a study in the Journal of the Atmospheric Sciences using high-resolution global models to quantify the ultimate predictability limit of midlatitude weather. Their finding: approximately two weeks, with current technology achieving useful deterministic skill out to nine or ten days and potential improvements over coming decades adding perhaps four to five additional days, but never exceeding the theoretical ceiling.⁸

Kerry Emanuel, professor of atmospheric science at MIT and a co-author of the study, summarized the result: “Edward Lorenz proved that one cannot predict the weather beyond some time horizon, even in principle. Our research shows that this weather predictability horizon is around two weeks, remarkably close to Lorenz’s estimate.”⁹

The proof is not controversial. It is not disputed. It is not the subject of active scientific debate. It is settled atmospheric science, confirmed by six decades of theoretical work, numerical experimentation, and operational forecast verification. The atmosphere is a chaotic system. Chaotic systems have finite predictability horizons. The atmosphere’s predictability horizon is approximately two weeks. Beyond that horizon, a forecast has no more skill than a table of historical averages compiled from decades of past observations.

This is what the scientific community calls “climatology.” It is what the rest of us call “guessing.”

III. What the Government’s Own Data Shows

The National Weather Service, an agency of the National Oceanic and Atmospheric Administration within the Department of Commerce, operates the Weather Prediction Center in College Park, Maryland. The WPC issues forecasts and, unusually among government agencies, publishes detailed verification statistics documenting how accurate those forecasts actually were.¹⁰

The verification data tell a clear story. For maximum temperature forecasts, the mean absolute error at a one-day lead time is approximately 2.5 to 3 degrees Fahrenheit. At three to four days, the error grows to approximately 3.5 to 4.5 degrees. At seven days, it reaches 5 to 6 degrees. The degradation is steady, monotonic, and well-documented across decades of operational forecasting.¹¹

ForecastWatch, an independent forecast verification service founded by Eric Floehr, has tracked the accuracy of consumer weather forecasts from ten major providers since 2005. Its twelve-year study covering 2005 through 2016, based on approximately 200 million forecast verifications, provides the most comprehensive publicly available assessment of how weather forecasts actually perform. The key findings:

One-day-out forecasts average less than three degrees Fahrenheit in error. This is genuinely impressive. Five-day-out forecasts in 2016 were approximately as accurate as one-day-out forecasts were in 2005, reflecting substantial improvement. But nine-day-out forecasts had, by 2016, only recently become marginally better than the long-term climatological average. In other words, a nine-day forecast had only just begun to outperform a system that simply predicts tomorrow’s high temperature will be the historical average high for that date at that location. A system that requires no satellite data, no supercomputer, no atmospheric model, and no meteorologist. A system that could be operated by a calendar and a filing cabinet.¹²

Beyond nine days, the data are unambiguous. Forecasts perform at or below the level of climatological averages. The forecast contains no information that the consumer could not obtain by looking up the thirty-year average temperature for the date in question. The sun icon, the temperature number, the precipitation percentage, and the confident visual presentation are wrapping paper around a historical average.

IV. What the Industry Sells

AccuWeather, Inc., is the largest private weather forecasting company in the United States, founded in 1962 by Joel Myers, then a graduate student in meteorology at Penn State. The company provides forecasts to media outlets, businesses, and directly to consumers through its website and mobile application. It employs approximately 450 people and claims to reach 1.5 billion users worldwide.¹³

In 2013, AccuWeather introduced forty-five-day forecasts. The forecasts presented specific temperatures, precipitation probabilities, and sky conditions for individual days up to six and a half weeks in advance. The meteorological community responded with what can charitably be described as alarm.

Jason Samenow, meteorologist and editor of the Washington Post’s Capital Weather Gang, conducted an independent verification of AccuWeather’s forty-five-day forecasts. His conclusion: the forecasts offered “no more accurate information than historical average conditions.” He wrote that the product was “not rooted in any science currently available to meteorologists and [has] not demonstrated value.” He concluded with two words: “Caveat emptor.”¹⁴

In 2016, AccuWeather extended its forecasts to ninety days. The response was not warmer. Gizmodo called the product “misleading as hell.” Dan Satterfield, writing for the American Geophysical Union, said the forecasts were “actually even worse than the Farmer’s Almanac, since they give rain chances and temperatures for exact points months into the future,” and advised consumers to “stand up for science and replace” the AccuWeather app on their phones.¹⁵

AccuWeather’s response to the criticism was revealing. Jon Porter, vice president of Innovation and Development, acknowledged in an interview with Gizmodo that the ninety-day forecast was “more about long-term trends than individual days.”¹⁶ The company itself, in a public statement, conceded that consumers “should not use long-range forecasts as a strict guide.” Joel Myers, the company’s founder, told NPR that consumers should not plan a baseball game around an eighty-eight-day forecast, despite the fact that AccuWeather’s own marketing materials had suggested precisely that use case. NPR host Robert Siegel noted the contradiction on the air.¹⁷

The product nevertheless remains available. It generates revenue through advertising impressions and premium subscriptions. It presents, on the same screen, in the same font, with the same precision, a forecast for tomorrow that is accurate to within three degrees and a forecast for three months from now that is accurate to within whatever the historical average happens to be. The user interface does not distinguish between them. No disclaimer accompanies the ninety-day forecast. No confidence interval is displayed. No notation indicates that the forecast for Day 87 contains no more predictive information than the corresponding entry in the 1991-2020 U.S. Climate Normals published by the National Centers for Environmental Information.

V. The Systematic Misrepresentation

The weather forecasting industry does not merely fail to inform consumers about the declining accuracy of extended forecasts. It actively misrepresents accuracy in the opposite direction, making forecasts appear worse than they are in the short term in order to avoid consumer dissatisfaction.

This practice is known as “wet bias,” and its existence has been documented by independent researchers and acknowledged by the industry itself.

In 2002, Eric Floehr began systematically collecting and verifying weather forecasts from the National Weather Service, The Weather Channel, and AccuWeather across the United States. His data, collected on ForecastWatch.com, revealed a striking pattern: commercial forecasters consistently predicted higher probabilities of precipitation than actually occurred. The NWS forecasts were statistically unbiased. The commercial forecasts were not.¹⁸

The magnitude of the bias was considerable. When The Weather Channel predicted a twenty percent probability of precipitation, it actually rained approximately five percent of the time. A stated probability of seventy percent corresponded to an observed frequency that roughly matched. Probabilities at the upper end were rounded up to one hundred percent regardless of actual likelihood. Local television stations exhibited even greater bias, with some reporting one hundred percent probability of precipitation in situations where it rained only seventy percent of the time.¹⁹

Nate Silver, in his 2012 book The Signal and the Noise: Why So Many Predictions Fail — but Some Don’t, documented the phenomenon and the industry’s rationale for it. Commercial forecasters bias their precipitation predictions upward because consumers are more upset by unexpected rain than by unexpected sunshine. A false alarm (predicting rain that does not occur) costs the consumer the inconvenience of carrying an unnecessary umbrella. A miss (failing to predict rain that does occur) costs the consumer wet clothing, a ruined event, and a loss of trust in the forecast. The industry has calculated that the asymmetric cost of these errors makes it commercially rational to overpredict rain.²⁰

The practice is, by the industry’s own account, intentional. It is not an artifact of model imperfection. It is not a consequence of insufficient data. It is a deliberate decision to present consumers with probability estimates that the forecaster knows to be higher than the best available scientific assessment. The forecaster possesses an unbiased estimate. The forecaster chooses to present a biased one. The consumer receives a number that is, in the most literal sense, a knowing misrepresentation of the forecaster’s own best estimate of the truth.

The FTC’s Deception Policy Statement does not contain an exception for misrepresentations motivated by the desire to help the consumer pack an umbrella.

VI. The FTC’s Own Precedent

The Commission has not been shy about enforcing the substantiation doctrine against companies that make claims unsupported by scientific evidence. The enforcement record is extensive and the penalties are substantial.

In 2012, the FTC reached a $40 million settlement with Skechers USA, Inc., for advertising that its Shape-ups, Resistance Runner, and Tone-ups shoes would help wearers lose weight, tone muscles, and improve cardiovascular health. The Commission found that Skechers had failed to possess competent and reliable scientific evidence to support these claims at the time they were made. The company had cited a study whose results, the FTC alleged, were cherry-picked to support claims the study did not actually produce. The Commission was not persuaded.⁴

In 2013, the FTC obtained a final order against POM Wonderful LLC, the pomegranate juice company, for advertising that its products could treat, prevent, or reduce the risk of heart disease, prostate cancer, and erectile dysfunction. The D.C. Circuit upheld the order in POM Wonderful LLC v. FTC, 777 F.3d 478 (D.C. Cir. 2015), holding that the company’s health claims required substantiation by “competent and reliable scientific evidence,” which the court defined as “tests, analyses, research, studies, or other evidence based on the expertise of professionals in the relevant area, that has been conducted and evaluated in an objective manner by persons qualified to do so, using procedures generally accepted in the profession to yield accurate and reliable results.”²²

In 2016, the FTC issued an Enforcement Policy Statement on Marketing Claims for Over-the-Counter Homeopathic Drugs, holding that efficacy claims for homeopathic products must be supported by competent and reliable scientific evidence or must be accompanied by prominent disclosures that there is no scientific evidence the product works and that the product’s claims are based only on theories from the 1700s that are not accepted by modern medical experts.²³

The standard is clear. If a company makes an objective claim about the efficacy or accuracy of a product, it must possess competent and reliable scientific evidence to support that claim. If it does not possess such evidence, the claim is deceptive. The meteorological community has provided competent and reliable scientific evidence, published in peer-reviewed journals and confirmed by six decades of operational verification, that weather forecasts beyond approximately ten days have no demonstrated skill above climatological averages. AccuWeather presents forecasts out to ninety days with no disclaimer, no confidence interval, and no notation that the scientific evidence supporting the Day 87 forecast is precisely as robust as the scientific evidence supporting the claim that Shape-ups shoes tone your glutes.

VII. The Scale of the Enterprise

The American weather enterprise is not a cottage industry. It is a multi-billion-dollar sector of the economy that shapes daily decisions for virtually every person in the United States.

The Pew Research Center has reported that weather is the single most popular category of smartphone application usage, with approximately seventy percent of smartphone owners checking weather forecasts on their devices.²⁴ Given that there are approximately 310 million smartphone users in the United States as of 2026, this implies approximately 217 million people consulting weather forecasts on their phones alone, not counting television, radio, newspaper, and direct NWS access. A conservative estimate of total daily weather forecast consumption in the United States is approximately 330 million individual forecast consultations per day.

These are not idle consultations. NOAA has estimated that approximately one-third of the United States gross domestic product, or roughly $7.5 trillion, is sensitive to weather conditions. Agriculture, energy, construction, transportation, retail, tourism, and insurance all make decisions based on weather forecasts. Airlines adjust flight schedules. Power companies purchase natural gas futures. Farmers decide when to plant and when to harvest. Construction crews decide whether to pour concrete. Event planners decide whether to rent a tent.²⁵

When these decisions are based on a one-day forecast with a mean absolute error of three degrees Fahrenheit, they are based on genuine predictive information. When they are based on a fourteen-day forecast that has not been demonstrated to outperform climatological averages, they are based on a representation that looks like a forecast, sounds like a forecast, is presented in the format of a forecast, and contains no more predictive information than the National Centers for Environmental Information’s thirty-year averages, which are available for free in a downloadable CSV file.

The FTC’s materiality standard asks whether the representation is “likely to affect the consumer’s conduct or decision with regard to a product or service.” A farmer who plants based on a fourteen-day forecast has been affected. A bride who books an outdoor venue based on a thirty-day forecast has been affected. A family that cancels a vacation based on a forty-five-day forecast has been affected. These are precisely the decisions that AccuWeather markets its extended forecasts to support. The company told NPR that its ninety-day forecast would help consumers plan vacations. It simultaneously told Gizmodo that the forecast was “more about long-term trends than individual days.” These statements are difficult to reconcile. They are easy to enforce against.

VIII. The Government-Funded Paradox

Perhaps the most remarkable aspect of the weather forecasting deception is that the government itself provides the data that proves the deception is occurring, and then continues to issue the forecasts anyway.

The National Weather Service is funded by the United States taxpayer at an annual budget of approximately $1.3 billion.²⁶ The NWS produces forecasts. The NWS also produces verification data documenting the accuracy of those forecasts. The verification data show that NWS forecasts degrade steadily with lead time and that forecasts beyond seven to ten days have limited skill. The NWS publishes this data on its own website. The NWS then continues to issue seven-day forecasts, which is responsible, and makes its data available to commercial providers who extend those forecasts to ninety days, which is not.

The commercial weather industry relies overwhelmingly on data produced by the National Weather Service and the European Centre for Medium-Range Weather Forecasts. The Global Forecast System (GFS), operated by NOAA’s National Centers for Environmental Prediction, produces forecast guidance out to sixteen days. The ECMWF’s Integrated Forecasting System produces guidance out to fifteen days. Neither model produces daily deterministic guidance beyond sixteen days because the atmospheric science community has determined, based on six decades of research initiated by Lorenz, that such guidance would be useless.²⁷

AccuWeather takes this sixteen-day guidance, supplements it with proprietary statistical methods that it has never subjected to independent peer review, and extends it to ninety days. The company has described its methodology as “proprietary.” It has not published the methodology in a peer-reviewed journal. It has not submitted its extended forecasts to independent verification by ForecastWatch or any other third-party evaluator. When asked to provide evidence that its ninety-day forecasts outperform climatological averages, its founder responded by noting that the forecasts “will give you information that will be better than you can have figured out in any other way.” Robert Siegel of NPR did not press the point. The FTC has not pressed the point either.

IX. The Arithmetic of Non-Enforcement

AccuWeather is not the only provider issuing forecasts beyond the demonstrated limits of atmospheric predictability. Apple Weather, which is preinstalled on approximately 1.46 billion active iPhones worldwide, displays a ten-day forecast. The Weather Channel app presents a fifteen-day forecast. Weather Underground presents a ten-day forecast. Dark Sky, before its acquisition by Apple, presented an eight-day forecast. Each of these products presents its extended forecasts in the same typographic style, with the same numerical precision, and with the same apparent confidence as its short-range forecasts. None displays a confidence interval. None displays a disclaimer. None distinguishes between a forecast with demonstrated skill and a forecast that performs no better than a desk calendar.

The aggregate annual revenue of the commercial weather forecasting industry in the United States exceeds $9 billion.²⁸ A substantial portion of this revenue is generated by advertising served alongside extended forecasts on mobile applications and websites. The longer the forecast, the more screens the user views. The more screens the user views, the more advertisements are served. The more advertisements are served, the more revenue the provider earns. The commercial incentive to extend forecasts beyond the limits of scientific validity is direct, measurable, and worth billions of dollars per year.

The FTC employs approximately 1,100 people.²⁹ The commercial weather forecasting industry serves approximately 330 million consumers per day. The ratio of enforcers to forecast consumers is approximately 1 to 300,000. The ratio of enforcement actions against weather forecast providers to enforcement actions against shoe companies is zero to one.

X. Conclusion

The Federal Trade Commission Act declares deceptive practices unlawful. The Commission’s own Deception Policy Statement defines deception as a representation likely to mislead consumers acting reasonably under the circumstances. The Commission’s own Advertising Substantiation doctrine requires that objective claims be supported by competent and reliable scientific evidence before they are disseminated.

Edward Lorenz proved in 1963 that the atmosphere is a chaotic system with finite deterministic predictability. Charney, Fleagle, Lally, Riehl, and Wark estimated in 1966 that the predictability limit is approximately two weeks. Zhang, Emanuel, and colleagues confirmed in 2019 that the limit is approximately two weeks. The NWS Weather Prediction Center’s own verification statistics show that forecast skill degrades steadily from a mean absolute error of three degrees Fahrenheit at one day to five or six degrees at seven days and no demonstrated skill above climatology at nine or ten days. ForecastWatch’s twelve-year, 200-million-verification study confirms the same degradation profile.

AccuWeather issues ninety-day forecasts. It does not publish its methodology. It does not submit its extended forecasts to independent verification. Its own executives have acknowledged that the ninety-day forecast is “more about long-term trends than individual days,” while its own marketing materials present the forecast as a day-by-day guide to planning outdoor activities three months in advance. The meteorological community has called these forecasts “not rooted in any science currently available to meteorologists.” The company continues to issue them.

The FTC obtained a forty-million-dollar settlement from a company that claimed unstable shoes would tone leg muscles. It obtained a final order against a company that claimed pomegranate juice could prevent heart disease. It issued an enforcement policy requiring that homeopathic products either be supported by scientific evidence or be accompanied by disclosures that no such evidence exists. In each case, the standard was the same: the company made an objective claim, the claim was not supported by competent and reliable scientific evidence, and the claim was material to consumers.

A weather forecast for Day 45 is an objective claim. It is not supported by competent and reliable scientific evidence. It is material to every consumer who has ever planned a wedding, booked a vacation, scheduled a construction project, or decided whether to plant winter wheat based on a forecast that, by the admission of the scientific community that created the underlying models, contains no predictive information whatsoever.

Approximately 330 million Americans consult weather forecasts every day. They receive, embedded within genuinely useful short-range predictions, extended forecasts that the atmospheric science community has determined cannot outperform a table of historical averages. The forecasts are presented identically. The consumer has no way to distinguish between them. The information asymmetry is total, the affected population is universal, and the commercial incentive to maintain the deception is measured in billions of dollars per year.

The Commission that extracted forty million dollars from a shoe company for an unsupported claim about calf muscles has not extracted a single dollar from a weather company for an unsupported claim about next month’s temperature. The Commission that required homeopathic products to disclose the absence of scientific evidence has not required a single weather app to disclose the absence of forecast skill. The Commission that has dedicated a century of enforcement to ensuring that consumers receive truthful representations about the products they use has not yet noticed that the most frequently consulted information product in the United States presents, on the same screen, in the same font, with the same implied authority, predictions that are supported by cutting-edge atmospheric science and predictions that are supported by nothing at all.

Lorenz proved that the atmosphere is unpredictable beyond approximately two weeks. The NWS proved that operational forecasts confirm Lorenz’s theory. ForecastWatch proved that commercial forecasts confirm the NWS data. The commercial weather industry proved that none of this matters if the revenue model depends on showing the user one more screen.

Ergo.