COVID-19 and the Cost of False Certainty

Crisis of Knowledge—and Communication

The COVID-19 pandemic confronted governments with an unavoidable problem: officials had to act before the evidence was complete. The virus was new, diagnostic capacity was initially limited, hospitals faced uncertain demand, and scientists were trying to understand transmission, immunity, treatment, and disease severity in real time. Some inaccurate forecasts, inconsistent statistics, and changing recommendations were inevitable.

The central failure was not simply that public-health authorities sometimes got things wrong. Mistakes were unavoidable during an emergency involving an unfamiliar pathogen. Nor was every discrepancy in case and death counts evidence of misconduct. Countries used different definitions, testing systems, and reporting practices, making some international comparisons closer to apples and oranges than standardized measurements. Once surveillance systems had been established, they were also difficult to change without disrupting comparisons over time.

The more consequential failure concerned the parts of the narrative that could have been restrained. Provisional judgments were often communicated as settled facts. Distinctions between evidence, inference, aspiration, and precaution disappeared from public messaging. Questions that should have remained open became tests of social responsibility, while disagreement was too readily characterized as ignorance, selfishness, or misinformation.

This problem was especially visible in the United States, where public discussion repeatedly flattened different questions into one. Who was most at risk? Could the virus be eliminated? Did vaccination prevent infection, severe disease, or transmission? How much protection followed prior infection? Did every kind of mask work under every condition? Did persistent symptoms after infection constitute one disease?

The answers were seldom binary. Yet binary answers were often what the public received.

A Virus With More Than One Risk Curve

One of the most important misconceptions concerned the distribution of risk. COVID-19 could infect people of every age, but the probability of hospitalization or death was never evenly distributed. Age was one of the strongest risk factors from the beginning, with chronic illness, obesity, immune status, and other medical conditions adding further variation.

Nevertheless, many Americans developed a mental picture of the virus as approximately equally dangerous to everyone. A July 2020 Franklin Templeton–Gallup survey found that respondents believed people 55 and older accounted for only slightly more than half of COVID-19 deaths. At the time, the actual proportion was approximately 92%. Respondents estimated that people 44 and younger accounted for about 30% of deaths, compared with an actual figure of approximately 2.7%.

This did not mean younger people faced no danger. Some became seriously ill, some died, and others experienced prolonged complications. But “not zero” is not equivalent to “evenly distributed.” Public-health communication frequently failed to preserve that distinction. Messages intended to persuade younger people to take precautions sometimes blurred relative and absolute risk rather than explaining both.

The mortality data confirmed the steep gradient. Approximately 81% of American COVID-19 deaths in 2020 occurred among people 65 or older. Archived CDC comparisons later estimated that, relative to people ages 18–29, those 85 and older faced roughly fifteen times the hospitalization risk and hundreds of times the risk of death.

A more candid message would have said that COVID-19 was a population-wide problem with radically unequal individual consequences. Younger people could transmit the virus to vulnerable people and could themselves experience uncommon but serious outcomes. At the same time, an elderly person with several medical conditions did not face the same risk as a healthy college student. Acknowledging that difference would not have trivialized COVID-19; it would have accurately described it.

Hospitalization Risk and the Loss of Proportion

Public estimates of hospitalization risk provide one of the clearest examples of distorted risk perception. In December 2020, a Gallup/Falcon Economics of Recovery survey asked Americans what percentage of people infected with COVID-19 had required hospitalization. Only 18% selected Gallup’s designated range of 1% to 5%. By contrast, 35% of all respondents believed that at least half of infected people had been hospitalized. That included 41% of Democrats and 28% of Republicans. Nearly two-thirds of Democrats—about 63%—selected an estimate of 31% or higher.

These were not cautious estimates that happened to be somewhat high. They were incompatible with the scale of hospital use the country was experiencing. A CDC-associated model estimated approximately 52.9 million infections and 2.4 million hospitalizations through September 2020. A 30% hospitalization rate would have produced almost 16 million hospitalizations in seven months; a 50% rate would have produced more than 26 million. Had either estimate been accurate, hospitals would not merely have experienced serious regional strain. Beds, emergency departments, intensive-care units, staffing, and ordinary medical services would have broken down across much of the country.

The partisan findings should not be reduced to a story in which one side understood COVID-19 and the other did not. Republicans were more likely to select Gallup’s hospitalization range, but they were also more likely to underestimate other dangers, including COVID’s lethality relative to influenza and its capacity for asymptomatic transmission. Democrats were more likely to recognize some of those dangers but greatly overestimated hospitalization and mortality risks, particularly for younger people. The broader problem was that political identity increasingly influenced which risks people magnified and which they minimized.

Gallup’s 1%–5% answer also requires qualification. It was a reasonable description of the general scale of risk based on the evidence available in late 2020, not a precise or permanent probability. Gallup’s appendix cited estimates of approximately 1.7% and 4.5%, both of which depended on modeled infections that surveillance systems had missed. The probability also changed with age, health, immunity, treatment, and the circulating variant. The range is most useful for showing why estimates of 30% or 50% were implausible; it should not be read as proof that everyone faced the same risk throughout the pandemic.

Four sources of noise are especially important:

• The denominator was incomplete. Official case counts missed asymptomatic infections, mild illnesses, people who never sought testing, unreported home tests, and some reinfections. These omissions were especially consequential among younger and healthier people. Because hospitalizations were more likely to be detected than mild infections, dividing hospitalizations by confirmed cases could overstate the probability that an infection led to admission.

• “With COVID” did not always mean “because of COVID.” COVID-NET included patients who tested positive within fourteen days before admission or during hospitalization. Most counted admissions during the Delta period were primarily for COVID-19, but the share admitted for other reasons rose during early Omicron. A trauma or surgical patient who happened to test positive did not represent the same event as a patient admitted with viral pneumonia.

• Hospitalization was partly a behavioral and institutional outcome. Patients decided when to seek emergency care, and clinicians decided whom to admit. Fear, clinical uncertainty, local protocols, underlying conditions, and available capacity could all influence that decision. A person who believed a fever meant there was a one-in-two chance of hospitalization might respond differently from someone who understood that risk in context.

• The underlying risk changed over time. A percentage calculated during the original strain could not simply be carried forward to Delta or Omicron. Vaccination, prior infection, improved treatment, the ages of those infected, and the characteristics of each variant all changed the probability of severe illness.

A national analysis using antibody surveillance to estimate infections during Delta and the initial Omicron wave illustrates that variation:

These are modeled population estimates, not directly observed individual probabilities.

The table reinforces two central points. Age produced enormous differences in risk, and the average changed substantially between variants. During initial Omicron, the estimated percentage among adults 70 and older was approximately twelve times that among adults aged 18–49. Across all ages, the estimate fell from approximately 6.7% during Delta to 1.6% during initial Omicron.

Even these estimates may be somewhat high if interpreted as the probability that COVID itself would cause hospitalization. The study could not reliably count reinfections, making its infection denominator too small, and it acknowledged that some included hospitalizations were principally caused by unrelated conditions. Separate CDC medical-record reviews found that COVID was the primary reason for approximately 95.5% of counted adult admissions during Delta but only 87.5% during early Omicron.

It is therefore reasonable to suspect that the average risk was often closer to the bottom of Gallup’s range than to the top, and that it fell below 1% for many younger and healthier people during parts of the pandemic. The evidence does not support declaring that the national risk was always 1% or less: Delta, advanced age and serious medical conditions produced substantially higher rates. But a 1% hospitalization probability would still be consequential for a virus infecting tens of millions of people. At 100 million infections, it would represent approximately one million hospitalizations.

That is the proper scale of the issue. COVID-19 did not need a 30% or 50% hospitalization rate to constitute a serious public-health emergency. A highly transmissible virus could place enormous pressure on hospitals even when severe outcomes remained uncommon for the average infected person. The communication failure was allowing population burden and individual risk to blur together. Many Americans came to believe that an infection commonly led to hospitalization across the population, when the available evidence showed a much lower average risk that varied sharply by age, health, immunity and time.

Zero COVID, Measles, and the Conditions for Eradication

The United States never adopted a formal national Zero COVID program comparable to those pursued in China, Australia, or New Zealand. Nevertheless, American rhetoric frequently encouraged the impression that sufficiently widespread vaccination would produce a decisive endpoint after which infections would largely cease.

It is too sweeping to say that SARS-CoV-2 could never have been eliminated merely because it was a coronavirus or because it mutated. SARS-CoV-1 was also a mutating coronavirus, yet sustained human transmission was stopped. The crucial difference was epidemiological. SARS-CoV-1 transmission was more closely associated with recognizable illness, enabling detection, isolation, and contact tracing to break transmission chains.

SARS-CoV-2 was much more difficult. People could transmit it before symptoms appeared or without recognizing that they were ill. By the time intensive controls were adopted, it had spread widely. Protection against infection was not permanent, and susceptible animal populations created opportunities for the virus to persist outside humans.

Part of the public misunderstanding appears to have arisen from an implicit comparison between COVID-19 vaccination and vaccination against diseases such as measles. No polling appears to have measured how many Americans believed the two viruses behaved similarly, so this should not be presented as a quantified national belief. Still, the comparison appeared regularly in public conversation and obscured important biological differences.

Measles has multiple genotypes, but they belong to one antigenically stable serotype. The existing vaccine protects against all circulating genotypes. Measles also has no natural reservoir outside humans, and infection or vaccination ordinarily produces highly durable immunity. These characteristics are why global measles eradication is considered biologically feasible.

Even measles has not been eradicated globally. It was declared eliminated in the United States in 2000, meaning continuous domestic transmission had been interrupted for at least twelve months. Imported cases and outbreaks still occur, particularly in communities with low vaccination coverage. Elimination and eradication are not the same.

SARS-CoV-2 presented a fundamentally different vaccine target. Mutations—particularly in the spike protein—produced variants capable of partially evading neutralizing antibodies generated by prior vaccination or infection. Vaccination could continue to protect against severe disease through broader immune memory even as protection against infection weakened. The virus therefore became a moving target for transmission-blocking immunity, though not a completely new target with every mutation.

Animal infections further complicated the picture. By late 2020, transmission among farmed mink had already generated international concern. Research later documented widespread infection among free-ranging white-tailed deer. A campaign directed exclusively at humans could not eradicate a virus capable of circulating in large animal populations. Put less clinically, Bambi wasn’t waiting in line with us.

These facts did not make vaccination futile. They meant that a successful vaccine against severe COVID-19 was not necessarily an eradication vaccine comparable to the measles vaccine. That distinction was rarely explained with sufficient clarity.

International experts were already discussing a long-term endemic future in 2020. At a WHO briefing on December 28, Professor David Heymann said that SARS-CoV-2 appeared destined to become endemic and continue mutating. That possibility did not receive equivalent prominence in American public discourse. Endemic management became central only after breakthrough infections and immune-evasive variants forced it into view.

What the Vaccines Demonstrated—and Why Scrutiny Was Rational

The initial Pfizer-BioNTech and Moderna trials produced genuinely impressive results. During the trial period, both vaccines greatly reduced symptomatic COVID-19, and later evidence demonstrated strong protection against hospitalization and death. For older adults and people with serious medical conditions—the groups most likely to suffer severe consequences—the case for vaccination was especially strong.

A clinical trial, however, establishes what it was designed and large enough to measure. Pfizer’s primary endpoint was prevention of symptomatic COVID-19, not complete interruption of infection or onward transmission. FDA reviewers noted in December 2020 that the evidence was insufficient to determine the vaccine’s effect on asymptomatic infection and viral shedding. The trial established that vaccinated participants were much less likely to develop symptomatic illness during the study period. It did not establish that vaccinated people could not become infected or transmit the virus.

That uncertainty was acknowledged in prime time. On December 3, 2020, NBC aired a Dateline special in which Lester Holt interviewed leaders from Pfizer, Moderna, and Johnson & Johnson. Asked whether a vaccinated person could still carry and transmit the virus, Pfizer CEO Albert Bourla said the question needed to be examined and that Pfizer was not certain based on the available evidence. Moderna chief medical officer Tal Zaks likewise explained that Moderna had demonstrated protection against illness but did not yet have enough evidence to prove that the vaccine reduced transmission.

Many scientists reasonably expected vaccination to reduce spread, because preventing infection or shortening illness usually reduces opportunities for transmission. Early real-world evidence supported that expectation. A CDC study of healthcare personnel, first responders, and other essential workers reported in March 2021 that the mRNA vaccines were about 90% effective against detected infection after full immunization during the thirteen-week study. The finding was encouraging, but it came from a limited pre-Delta period and did not establish complete or permanent transmission blocking.

Public rhetoric soon outran those limits. On March 29, 2021, Rachel Maddow told viewers that “the virus stops with every vaccinated person.” At a CNN town hall on July 21, President Biden said, “You’re not going to get COVID if you have these vaccinations.” Neither claim accurately expressed what the evidence could establish and was akin to announcing “Mission Accomplished”. The vaccines reduced infection and transmission; they did not end them altogether. Days after Biden’s statement, the CDC publicly acknowledged that vaccinated people infected with Delta could transmit the virus and revised its masking guidance.

Waning protection added another layer of complexity. Protection against infection generally reached its highest level soon after the second dose and then declined. A large Qatar study estimated Pfizer’s effectiveness against infection at approximately 77.5% during the first month after the second dose, followed by a progressive decline that became more pronounced after the fourth month. A Kaiser Permanente study found that effectiveness against Delta infection declined from approximately 93% during the first month to 53% after four to five months.

Protection against severe disease proved considerably more durable. A CDC study estimated effectiveness against COVID-19 hospitalization at 86% during weeks 2 through 12 after the second mRNA dose and 84% during weeks 13 through 24, with no statistically significant decline between those periods. The vaccines therefore continued performing their most important medical function even as their usefulness as a lasting barrier against infection and transmission weakened.

Three propositions were repeatedly compressed into the phrase “the vaccines work.” Vaccination reduced severe illness and death. It reduced the recipient’s chance of infection, particularly during the first months after vaccination. It also reduced onward transmission under certain conditions. All three received evidentiary support, but they were not equally strong or durable. Protection against hospitalization and death was the most reliable; protection against infection and transmission depended much more on timing, prior immunity, and the circulating variant.

Those distinctions mattered because benefit and risk were not identical across the population. For an elderly person facing a substantial risk of hospitalization or death, even a modest additional reduction in severe disease could be extremely valuable. As age and underlying risk declined, the absolute benefit also became smaller. That did not make vaccination useless for younger healthy people, but it made age, sex, medical history, prior infection, and earlier doses increasingly relevant to an individual decision.

The emergence of myocarditis after mRNA vaccination made that calculation particularly important for adolescent boys and young men. The condition was rare and generally resolved clinically, and it had to be weighed against cardiovascular complications associated with COVID-19 itself. Nevertheless, it was a genuine vaccine-associated risk concentrated in a group with a comparatively low probability of severe COVID-19. CDC surveillance in 2021 estimated approximately 62.8 reported cases per million second doses among males ages 12–17 and 50.5 per million among males ages 18–24, with substantially lower reporting rates among older men and women.

The booster debate exposed the same need for demographic precision. On August 18, 2021, President Biden announced plans to make boosters broadly available before the FDA’s independent vaccine advisory committee had completed its review. Anthony Fauci supported the administration’s broader approach. Paul Offit, a member of the committee, argued that the evidence most clearly supported boosters for older and medically vulnerable people rather than every healthy younger adult.

On September 17, the committee voted 16–2 against authorizing Pfizer boosters for everyone sixteen and older, then unanimously supported them for people 65 and older and those at elevated risk. The FDA initially adopted the narrower authorization, although federal recommendations were later expanded. The disagreement was not between people who believed vaccines worked and people who believed they did not. It concerned how much additional benefit another dose provided, which populations clearly needed it, and whether the evidence justified a universal policy.

Private government discussions likewise showed that responsible officials still had unresolved questions. In January 2021, Surgeon General Vivek Murthy asked Fauci and CDC Director Rochelle Walensky whether there was any reason to prefer vaccination earlier or later in pregnancy. Walensky initially called the question “data free,” and Fauci agreed that there was no known reason to prefer one period. After consulting NIAID Vaccine Research Center director John Mascola, Fauci raised a theoretical concern about fever and a strong inflammatory response after the second dose during the first trimester; Walensky replied that it was “definitely a good point.” The exchange showed senior officials doing what sound science requires: identifying a gap, consulting expertise, and reconsidering an initial answer. Members of the public were entitled to ask similarly careful questions, particularly because pregnant women had not been deliberately enrolled in the original trials.

Post-authorization surveillance further demonstrated why scrutiny had to continue. The Johnson & Johnson vaccine was temporarily paused in April 2021 after reports of a rare but serious clotting disorder. The pause was lifted with updated warnings, but the CDC later recommended Pfizer or Moderna over Johnson & Johnson, and in May 2022 the FDA restricted its authorized use. AstraZeneca’s related viral-vector vaccine was never authorized in the United States and was restricted by age or deemphasized in several other countries after reports of the same rare syndrome. Its European authorization was ultimately withdrawn at the manufacturer’s request amid declining demand and the availability of newer vaccines. These histories did not establish that COVID-19 vaccination was generally unsafe; they showed that products were not interchangeable and that benefit-risk judgments could change as evidence accumulated.

That distinction became especially important when vaccination moved from a voluntary medical recommendation to a legal or institutional requirement. Federal rules reached healthcare workers, federal employees and contractors, and military personnel. The Biden administration also attempted to require vaccination or regular testing for employees of large private businesses through OSHA. States, universities, hospitals, and private employers imposed additional requirements, placing jobs, education, and professional opportunities at stake. The Supreme Court allowed the federal healthcare-worker rule to proceed while blocking OSHA’s broader private-workplace rule.

A mandate raises a different question from a recommendation. A recommendation asks whether vaccination is likely to benefit an individual. A mandate must also show that compelling that person serves a sufficiently strong public purpose to override individual choice. That argument was strongest when vaccination was expected to provide a durable barrier against transmission. As protection against infection waned and prior infection became more widespread, the justification for treating every person as presenting the same risk became less persuasive.

The evidence strongly supported vaccination as protection against severe disease, especially for older and medically vulnerable Americans. That conclusion can coexist with recognition that the vaccines did not permanently stop infection or transmission, that uncommon adverse effects differed by product and demographic group, and that the value of additional doses was not identical for everyone. Scrutinizing those questions was not inherently anti-vaccine. It was part of the process by which medical recommendations should be refined.

Natural Immunity and a False Binary

The debate over natural immunity became needlessly polarized. Infection with a virus ordinarily stimulates immune memory, including antibodies and B- and T-cell responses. There was never a sound biological reason to assume that SARS-CoV-2 infection would produce no subsequent protection. The legitimate questions concerned its strength, consistency, duration, and ability to withstand new variants.

Infection-acquired immunity also came at the cost of acquiring the disease. Vaccination offered a way to develop protection without accepting the full risks of an initial infection. That difference was especially consequential for older adults and people with serious medical conditions. Recognizing natural immunity did not require recommending infection as a public-health strategy.

The original vaccine trials included some participants with evidence of prior infection, but there were too few later cases in that subgroup to estimate reliably how much additional benefit vaccination provided. It is more accurate to say previously infected people were not adequately studied for that question than to say they were entirely excluded.

Paul Offit later described a private 2021 consultation attended by Fauci, NIH Director Francis Collins, Walensky, and Murthy. Offit and three other outside experts were asked whether prior infection should count as vaccination under vaccine requirements. According to Offit, the four outside experts split evenly, and he favored recognizing prior infection. His account does not identify the other experts or place Fauci within that two-to-two division. It does establish that serious disagreement existed within the government’s own consultation process even as policy generally treated vaccination as necessary regardless of previous infection.

The CDC’s Kentucky study became an example of how a defensible finding could be communicated too broadly. Published in August 2021, it examined Kentucky residents infected during 2020 and compared those with and without documented reinfection during May and June 2021. Among this previously infected population, unvaccinated individuals had 2.34 times the odds of documented reinfection compared with fully vaccinated individuals. The result supported an additional benefit from vaccination after infection.

The CDC’s accompanying headline—“Vaccination Offers Higher Protection than Previous COVID-19 Infection”—implied a comparison the study had not made. Everyone in the study had previously been infected. It compared prior infection alone with prior infection followed by vaccination: natural immunity with hybrid immunity.

The study was retrospective and researchers used laboratory and immunization records. Its limitations were more specific: it covered one state and two months, could miss vaccinations or infections absent from the relevant databases, could not fully account for differences in testing behavior, and could not establish causation. In the final paragraph the report itself acknowledged that its retrospective design and restricted geographic and temporal scope limited the conclusions.

Later evidence made the broader picture clearer. During Delta, CDC data from California and New York showed that previous infection provided substantial protection against later cases and hospitalization. People with neither vaccination nor prior infection consistently faced the greatest risk, while those with both forms of immunity generally possessed broad protection. Other studies found that infection-derived protection waned and varied by variant, while hybrid immunity often produced stronger or broader protection than either vaccination or infection alone.

The evidence never fit comfortably into the slogan “vaccines work, natural immunity is irrelevant,” any more than it supported the opposite claim that infection was preferable to vaccination. Both infection and vaccination shape immune memory. Their effects change with time and variants, and their value depends heavily on age, medical vulnerability, prior exposure, and the outcome being measured.

Masks: Filtration, Fit, and Population Effectiveness

Masking underwent a similar transformation from a conditional intervention into an identity-laden absolute. “Masks work” concealed more than it explained. Work for whom? Against which virus? In what setting? With what type of mask, fit, and consistency? Was the outcome infection in the wearer, source control from an infected person, or community transmission?

Before COVID-19, the WHO commissioned a systematic review of non-pharmaceutical interventions against pandemic influenza. Its literature search covered publications from 1946 through 2018; that does not mean equivalent mask trials were conducted continuously across seven decades. The review identified ten randomized community trials involving more than 6,500 participants. In pooled analysis, mask interventions did not produce a statistically significant reduction in laboratory-confirmed influenza. None of the household trials individually found a significant reduction in secondary laboratory-confirmed infections, although many were underpowered and adherence was often poor. (WHO-commissioned review)

That evidence cannot simply be transferred unchanged to COVID-19. Influenza and SARS-CoV-2 are different viruses, and the trials primarily examined surgical or similar masks rather than professionally fitted respirators. Nevertheless, the pre-pandemic evidence should have discouraged absolute claims about community masking.

The review excluded respirator use in the community and noted that N95 and similar respirators should provide better protection because of their filtration efficiency when properly fitted. The phrase “when properly fitted” is essential. A filter’s laboratory performance is not the same as the protection produced when air leaks around its edges.

Healthcare studies show how quickly real-world performance can depart from an initial fit test. In a multicenter study across six American emergency departments, 412 healthcare workers used the same N95 for more than half a clinical shift and reused it for as many as five shifts. After one shift, 38.7% of the respirators failed a qualitative fit test; after five shifts, 92.8% had failed. Failure varied substantially by respirator design. Those results should not be generalized into a claim that the same percentage of all N95s fails during ordinary use—the study examined extended use and reuse under emergency-department conditions—but they demonstrate why fit and behavior cannot be treated as footnotes.

Performance depends on respirator design, facial structure, training, moisture, facial movement, speaking, repeated adjustment, donning and doffing, and reuse. Touching a mask matters because it can disturb the seal and create opportunities for hand or surface contamination. though it should not be presented as a principal mechanism of SARS-CoV-2 transmission, which was predominantly airborne.

This is the difference between individual efficacy and population effectiveness. A properly selected and fitted respirator can substantially reduce an individual wearer’s exposure. That does not guarantee an equivalent community-wide effect. Population policies involve different devices, inconsistent fitting, removal for speaking or eating, reuse, varied exposure, and imperfect adherence.

COVID-era randomized evidence was mixed but not empty. Denmark’s DANMASK trial did not find a statistically significant reduction in infection among people advised to wear surgical masks, although it primarily measured protection of the wearer rather than community source control and remained compatible with a modest benefit. A much larger cluster-randomized trial in Bangladesh found a modest reduction in symptomatic seroprevalence from a program promoting surgical masks, while the evidence for cloth masks was less clear.

The appropriate conclusion is neither that masks were useless nor that any face covering provided reliable protection under all circumstances. A well-fitted N95 worn consistently in a crowded, poorly ventilated indoor environment is not equivalent to a loose cloth mask repeatedly adjusted by its wearer. The scientific question “Can this device reduce exposure when correctly used?” is distinct from the policy question “Will requiring an entire population to wear a wide range of face coverings produce a large, measurable reduction in transmission?” During the pandemic, those questions were too often treated as interchangeable.

Long COVID and the Sherlock Holmes Problem

In Sir Arthur Conan Doyle’s A Scandal in Bohemia, Sherlock Holmes offers an appropriate warning: “It is a capital mistake to theorize before one has data. Insensibly one begins to twist facts to suit theories, instead of theories to suit facts.”

Persistent illness after viral infection is medically plausible and historically familiar. COVID-19 has produced identifiable long-term injuries, including persistent loss of smell, lung damage following severe pneumonia, cardiovascular complications, and autonomic and neurological problems. It would be wrong to treat all patients reporting prolonged illness as mistaken or psychologically motivated.

The classification problem is different. Early descriptions of long COVID incorporated dozens of symptoms, including fatigue, headache, insomnia, anxiety, shortness of breath, gastrointestinal complaints, pain, dizziness, and problems with memory or concentration. These symptoms can be genuine and disabling. They are also common in the general population and can arise from many medical, psychological, and social causes.

Fatigue deserves particular attention because it became one of the most frequently reported symptoms while remaining one of medicine’s least specific complaints. It can follow a viral infection, but it can also arise from depression, chronic stress, disturbed sleep, inactivity, medication, anemia, endocrine disorders, and numerous other conditions. The same attribution problem applies to headache, poor concentration, anxiety, insomnia, and gastrointestinal discomfort.

The pandemic created an unusual background against which symptoms were experienced and interpreted. People faced fear of illness, social isolation, disrupted sleep, reduced physical activity, interrupted medical care, employment uncertainty, and heightened attention to bodily sensations. CDC Household Pulse Survey data found that measures of anxiety and depression increased nationally between August 2020 and February 2021 before declining during the following months.

Fear was neither an all-purpose explanation nor an irrelevant variable. It does not show that an individual patient’s symptoms were psychological, and it cannot explain measurable organ damage or every post-viral syndrome. It was, however, part of the environment in which symptoms were experienced, remembered, reported, and attributed. Any credible estimate of virus-specific long-COVID prevalence must attempt to account for that environment.

A French analysis of 26,823 adults from the CONSTANCES cohort illustrates the attribution problem. The belief that one had experienced COVID-19 was associated with most of the persistent symptoms studied, including fatigue, breathlessness, and impaired attention. Positive SARS-CoV-2 serology, by contrast, was associated only with persistent loss of smell. The study did not prove that belief or fear caused the other symptoms. It was cross-sectional, its antibody test was imperfect, and symptoms may have led people to believe they had experienced COVID rather than the reverse. Its narrower lesson is that self-report and self-attribution alone cannot establish cause.

Broad definitions maximize sensitivity but sacrifice specificity. Defining long COVID as almost any symptom continuing for a set period after infection captures people with genuine post-viral disease. It may also capture incomplete recovery from severe illness, preexisting conditions, effects of pandemic disruption, unrelated diseases, and background symptoms newly interpreted through the long-COVID framework.

CDC prevalence estimates have often relied on the experimental Household Pulse Survey, which asks respondents whether they had COVID-19 and whether they currently have symptoms lasting three months or longer that were not present before infection. These weighted self-reports are useful for identifying a potential population burden, but they are not diagnoses confirmed through biomarkers or prospective clinical examinations. They cannot by themselves determine how many symptoms were caused by the virus.

The age distribution raises a related question. Self-reported long COVID has generally been most prevalent among middle-aged adults rather than among the oldest groups that experienced the highest acute hospitalization and mortality rates. This mismatch does not necessarily disprove long COVID. Acute fatal disease and chronic post-infectious illness need not have identical mechanisms or risk curves. Survivorship, immune differences, competing chronic illnesses, reporting behavior, employment demands, and underrecognition among the elderly could all contribute. But the divergent curves should have encouraged caution about presenting long COVID as a simple continuation of severe acute COVID.

More rigorous studies using infected and uninfected comparison groups have found symptom clusters occurring more often after infection. At the same time, the NIH RECOVER initiative and major scientific reviews have acknowledged the lack of one definitive diagnostic biomarker and the tradeoff between capturing genuine cases and falsely classifying unrelated symptoms.

The most plausible interpretation is that long COVID is an umbrella rather than one disease with one mechanism. It may include organ damage after severe infection, persistent loss of smell or taste, post-viral fatigue syndromes, autonomic dysfunction, immune or inflammatory abnormalities, exacerbations of prior illness, and symptoms whose relationship to infection remains uncertain.

Scrutinizing the category’s boundaries is not the same as denying patient suffering. Scientific progress requires narrower case definitions, appropriate uninfected control groups, verified infection histories, pre-infection baselines, prospective measurement, and objective clinical findings where possible. The theory must be refined to fit the data rather than every persistent symptom being fitted retrospectively into an already successful theory.

The Emergence of a Monopoly of Information

Pandemic communication did not occur in an open scientific vacuum. Government agencies, technology companies, news organizations, professional associations, and academic institutions were under enormous pressure to stop false claims from spreading. Some intervention was understandable. Fabricated cures, fraudulent products, and genuinely dangerous medical advice circulated widely.

The category of “misinformation,” however, frequently expanded beyond demonstrably false claims. It came to include disputed interpretations, inconvenient evidence, premature conclusions, and occasionally positions that later became compatible with mainstream understanding. Government officials communicated extensively with social-media companies, flagged content, requested information about enforcement, and sometimes pushed platforms to take stronger action.

Mark Zuckerberg later acknowledged that pressure directly. In an August 2024 letter to Congress, he wrote that senior Biden administration officials had “repeatedly pressured our teams for months to censor certain COVID-19 content, including humor and satire.” Zuckerberg emphasized that Meta made—and remained responsible for—its own decisions, but concluded: “I believe the government pressure was wrong.”

Former Twitter CEO Jack Dorsey offered a broader acknowledgment. Discussing disclosures from the Twitter Files after leaving the company, he said, “I was surprised by the level of engagement with government agencies,” while also saying that Twitter employees had generally tried to do the right thing. His statement was not limited to COVID and did not establish that every government contact was coercive. It did confirm that the degree of engagement was greater than even the company’s former chief executive had understood.

The constitutional meaning of “the press” should not be confused with the modern journalism profession. The First Amendment creates no credentialing system that determines who qualifies as a journalist. In Lovell v. Griffin, the Supreme Court explained that freedom of the press was not confined to newspapers and periodicals but extended to “every sort of publication which affords a vehicle of information and opinion.” The protection therefore reaches ordinary citizens engaged in publishing and distributing ideas, not only established news organizations.

That principle became especially consequential during COVID-19. In 2020, Pew found that 53% of American adults obtained news from social media at least sometimes. These platforms were important channels of public information, not merely private spaces for casual conversation. Government efforts to influence what could be published or encountered there therefore raised First Amendment concerns even when the affected speakers were not professional journalists. Those concerns do not erase the platforms’ editorial rights or make every government communication coercive. They do make it inadequate to treat pressure on social-media speech as unrelated to freedom of the press.

The constitutional dispute reached the Supreme Court in Murthy v. Missouri. The record described extensive communications between federal officials and social-media platforms, including requests concerning COVID-19 content. In 2024, however, the Court did not decide whether those communications amounted to unconstitutional coercion or significant encouragement. It held that the plaintiffs lacked standing to seek a forward-looking injunction because they had not sufficiently connected specific restrictions—or a likely future injury—to the actions of the named federal officials. The case was resolved on standing, not on the underlying First Amendment question.

The record supports the conclusion that government pressure, content referrals, and requests for stronger enforcement occurred. It does not establish that every removal resulted from coercion or that one authority controlled all pandemic speech. Technology companies had their own policies, employees, political pressures, and commercial incentives. Many dissenting claims were genuinely wrong, and some were reckless.

The deeper problem was the feedback loop. Officials simplified claims to encourage compliance. Platforms increasingly treated official claims as proxies for truth. Journalists looked to government statements and platform enforcement when deciding which arguments were credible. Scientists who challenged prevailing conclusions could then be portrayed as outside the consensus—even when the apparent consensus had been strengthened by excluding competing interpretations.

This was a network of powerful institutions repeatedly validating one another. Government officials cited scientific consensus, platforms enforced the government’s interpretation of that consensus, and news organizations treated those enforcement decisions as evidence that dissenting claims were unreliable. Once the loop formed, legitimate uncertainty could be mistaken for misinformation simply because the institutions defining misinformation were relying on one another.

Scientific consensus is essential, but it is a conclusion reached through evidence and argument—not an administrative category that determines which evidence may be discussed. Restricting demonstrably fraudulent or dangerous claims may prevent immediate harm. Suppressing legitimate uncertainty, particularly under government pressure, creates a different danger: the delayed collapse of public trust when provisional claims are later revised.

The Cost of Treating Confidence as Evidence

COVID-19 vaccines prevented severe illness and saved lives. Reductions in close contact lowered opportunities for transmission. Ventilation reduced the concentration of infectious aerosols. Properly fitted respirators could reduce individual exposure. Prior infection generated meaningful immune protection. Some patients developed genuine and sometimes disabling post-viral complications. None of these conclusions requires accepting every claim made in their name.

Nor does criticism of pandemic communication require believing that officials acted from uniformly malicious motives. They faced a real emergency, incomplete evidence, political pressure, and the challenge of persuading millions of people to alter their behavior. Simplification was sometimes unavoidable.

What was not unavoidable was the repeated conversion of uncertainty into certainty.

Officials could have explained that risk was steeply age-dependent while still asking younger people to protect vulnerable neighbors. They could have celebrated strong vaccine protection against severe disease while acknowledging that transmission remained under study and would change with time and variants. They could have recommended vaccination after infection while recognizing that infection itself produced substantial immunity. They could have recommended high-quality respirators in high-risk environments without pretending that every mask and every mandate had equivalent evidence behind it. They could have investigated persistent post-COVID illness without prematurely presenting a broad inventory of nonspecific symptoms as one settled disease.

The lesson of COVID-19 is not that expertise has no value. It is that expertise becomes less trustworthy when it conceals its own limits. During an emergency, authorities may fear that acknowledging uncertainty will weaken compliance. Over time, the opposite can occur. People who discover that yesterday’s certainty was provisional may begin to distrust today’s well-established evidence as well.

Scientific authority is preserved not by claiming a monopoly on truth, but by showing the public how truth is approached: through evidence, disagreement, replication, revision, and an honest account of what remains unknown.