The algorithm learned that rural users were harder to engage, so it stopped trying.
A World Bank study conducted across India has revealed that the power of social media to prevent disease is real — but unevenly distributed. Facebook and Instagram ads meaningfully reduced malaria in cities, yet the same campaign left rural, high-risk communities largely untouched, not by intention, but by the quiet logic of algorithmic efficiency. The findings ask a question older than any platform: when a tool works well for those already advantaged, does it serve the public good?
- Urban residents exposed to the campaign used bed nets 11% more often and saw malaria cases fall by nearly a third — proof that the intervention itself was sound.
- Rural populations living in non-solid dwellings showed no measurable change, revealing that 130 million impressions can mask a profound failure of reach.
- A controlled experiment confirmed the algorithm, not the message, was the obstacle — when rural users actually saw the ads, their behavior shifted meaningfully.
- Reaching a vulnerable rural resident cost ten times more than reaching an urban one, forcing a reckoning between cost-efficiency and equitable public health.
- Researchers are now calling for first-party data integration and deliberate targeting tools to override algorithmic bias toward easy audiences.
When World Bank researchers asked whether Facebook ads could save lives, they discovered the answer depended entirely on where someone lived. Working with the NGO Malaria No More and Facebook's Campaigns for a Healthier World initiative, the team tracked a malaria prevention campaign across 80 districts in three Indian states, encouraging bed net use and prompt treatment-seeking across more than 130 million people.
In cities, the results were striking. Urban residents in solid dwellings increased bed net use by 11 percent, sought treatment 13 percent more readily, and health records showed a 30 percent monthly drop in malaria cases. The campaign appeared to be a public health success — until researchers looked at the rural data, where the ads had produced no measurable change at all.
A second experiment revealed the culprit. When rural residents in non-solid dwellings were directly shown the same ads, they responded — bed net use rose 8 percent and treatment-seeking improved. The content worked. The algorithm did not. Facebook's system, built to maximize engagement, was quietly routing ads toward easier-to-reach urban users and bypassing the most vulnerable populations entirely.
The cost gap was stark: reaching someone in a non-solid dwelling cost nearly ten times more than reaching someone in a solid one. The $202,000 national campaign averted urban malaria cases at $3 to $6 each — competitive with traditional interventions — but those savings came at the expense of those most at risk.
The researchers argue that social media campaigns can meaningfully complement traditional health programs, but only if platforms and agencies are willing to use smarter targeting tools — and accept higher costs as the price of genuine equity. The deeper question is whether optimizing for reach and optimizing for justice can ever be made to point in the same direction.
A team of researchers from the World Bank set out to answer a deceptively simple question: can Facebook ads save lives? What they found was more complicated than a yes or no. The answer, it turned out, depended entirely on where you lived.
The study, conducted in collaboration with Facebook's Campaigns for a Healthier World initiative, tracked a malaria prevention campaign across 80 districts in three Indian states. Researchers Dante Donati, Nandan Rao, Victor Orozco-Olvera, and Ana Maria Muñoz-Boudet worked with the NGO Malaria No More to distribute ads on Facebook and Instagram encouraging two simple behaviors: using mosquito nets consistently and seeking treatment quickly if symptoms appeared. The campaign reached more than 130 million people across 22 states. On paper, it looked like a public health success waiting to happen.
But the results told two starkly different stories. In cities, the campaign worked. People living in solid dwellings—concrete, brick, stone—who saw the ads increased their bed net use by 11 percent. They were 13 percent more likely to seek treatment promptly if they developed symptoms. Health facility records showed a 30 percent drop in monthly malaria cases in urban districts. The campaign even reported a 44 percent reduction in self-reported cases. These were the kinds of numbers that make public health officials sit up and take notice.
In rural areas, the ads changed nothing. Among people living in non-solid dwellings—structures made of mud, straw, or tin—the campaign had no measurable impact on behavior or disease incidence. This was the puzzle. The ads worked in cities. The ads existed. So why weren't they reaching the people who needed them most?
The researchers designed a second experiment to find out. They took the same ads and made sure they appeared directly in the Facebook feeds of people in both solid and non-solid dwellings. When rural residents actually saw the ads, they responded. Bed net use rose 8 percent. Treatment-seeking improved by 2.5 percent. The content was fine. The problem was the algorithm. Facebook's targeting system, optimized to maximize engagement and reach the easiest-to-reach users, was systematically bypassing rural, lower-income populations. It wasn't malice. It was mathematics. The algorithm simply found it more efficient to show ads to people already primed to engage.
The cost disparity was stunning. Reaching someone in a non-solid dwelling cost nearly ten times more than reaching someone in a solid dwelling. For a public health campaign operating on a fixed budget, that math was brutal. The researchers calculated that the national campaign, which cost $202,000 total, averted malaria cases at roughly $3.35 to $6.54 each in urban areas—competitive with traditional interventions like bed net distribution, which costs around $5 per case averted. But those savings came at the expense of the people most vulnerable to malaria.
The researchers concluded that social media campaigns could complement traditional health programs, especially in cities where they're cost-effective and scalable. But to reach their potential, they need smarter targeting. The researchers propose using first-party data collected directly from users, combined with advanced techniques like Facebook's Lookalike Audience tool, to deliberately reach high-risk populations even when the algorithm wouldn't naturally send them there. It means accepting higher costs per person reached in rural areas, but treating that as a feature of equitable public health, not a bug to be engineered away. The question now is whether platforms and public health agencies are willing to prioritize reaching the hardest-to-reach people over the easiest metrics.
Notable Quotes
The core issue was not the ad content but the targeting mechanisms used, which tended to prioritize reaching users more easily engaged or more accessible on social media, often bypassing rural, lower-income, and high-risk populations.— World Bank researchers
The Hearth Conversation Another angle on the story
So the ads worked. They reduced malaria cases in cities by 30 percent. Why is that a problem?
Because the people who got sick most often never saw the ads. The algorithm was solving for engagement, not for health equity. It was like having a medicine that works but only giving it to people who don't need it as much.
But you said the ads worked when rural people actually saw them. So the content was good?
Yes. That's what makes it worse in a way. It wasn't that the message was wrong. It was that the system was designed to avoid sending the message to the people at highest risk. The algorithm learned that rural users were harder to engage, so it stopped trying.
How much harder are we talking about?
Ten times more expensive to reach someone in a mud or straw dwelling than someone in a concrete house. For a campaign with a fixed budget, that's a choice between reaching many people cheaply or reaching the most vulnerable people at all.
Did the researchers suggest a fix?
They want campaigns to use first-party data—information people give directly to Facebook—combined with tools designed to find people similar to high-risk groups, even if the algorithm wouldn't naturally show them ads. Basically, deliberately paying more to reach the people who need it most.
That sounds expensive.
It is. But so is malaria. The study showed that when you factor in unreported cases, the cost per prevented case drops to 20 cents. That's cheaper than almost any other intervention. The question is whether we're willing to spend more on the people the algorithm ignores.