The Election Machine: How Campaigns Turn Data Into Your Political Feed

A political campaign can decide which voters it wants to reach. It can select people who live in a particular district, belong to a certain age group, appear persuadable on an issue, or have a history of voting in similar elections.

That is only the first decision.

The platform carrying the advertisement may then decide who actually sees it, how often it appears, what it costs, and which audience receives more or less of the campaign’s money. Those decisions can be made by automated systems that voters cannot inspect and campaigns may not fully control.

This is the central fact behind modern political advertising: targeting is no longer only a choice made by a campaign. It is a negotiation between voter data, campaign strategy, advertising markets, and platform algorithms.

The result is not a machine that can read every voter’s mind or determine an election by itself. The evidence does not support that claim. But it does show a political system in which campaigns, data brokers, and technology platforms can assemble detailed voter profiles, divide audiences into narrow groups, send different messages to different people, and sometimes reach a population different from the one the campaign selected.

That changes what voters know about politics, what campaigns know about voters, and who gets to see which version of a candidate’s argument.

The voter file is where the process begins

Political data collection did not begin with Facebook.

Campaigns have long used voter registration records containing information such as a person’s name, address, party registration, and participation in past elections. In many jurisdictions, records can show whether someone voted, although they do not reveal whom that person supported. Campaigns and political organizations can use those records to classify people broadly as likely supporters, likely opponents, or possible persuadable voters. (Electronic Frontier Foundation)

Those categories are imperfect. A voter registered with one party may disagree with its candidate. A person who voted in several elections may stay home in the next one. A person who signed a petition may support placing an issue on the ballot without supporting the issue itself.

Campaigns nevertheless use these records because they provide a starting map of the electorate. They indicate where voters live, whether they have participated before, and sometimes which political organization may have a relationship with them.

The data also moves between organizations. Campaign privacy policies commonly permit information to be shared with political groups, candidates, causes, or organizations with similar objectives. The EFF cites campaign privacy policies that allow this kind of exchange and notes that candidate email lists can have substantial financial value. In 2017, for example, the Hillary Clinton campaign gave its email list to the Democratic National Committee in a transaction valued at $3.5 million. (Electronic Frontier Foundation)

A voter’s first interaction with one campaign can therefore become useful to another campaign later. Signing up for an email list, attending an event, donating money, downloading an app, or requesting information may create a record that remains valuable beyond the immediate election.

How campaigns add consumer and online data

The voter file rarely contains enough information to determine what message will work. Campaigns and political organizations may try to add other data, often through commercial data brokers.

The combined record can include consumer information, demographic characteristics, location signals, browser activity, mobile-device data, social-media information, and details gathered directly by campaigns. Some of these records are supplied by individuals. Others are inferred. (Electronic Frontier Foundation; Reuters)

An inference is not a fact supplied by the voter. It is a prediction made from other information. A data company might estimate that someone is interested in a particular issue, likely to own a home, receptive to a type of appeal, or unlikely to vote. The person may never have stated any of those things.

The EFF describes a simplified process in which a campaign begins with voter records, gives them to a data broker for enrichment, selects categories of people it believes will respond to particular messages, and then works with an advertising company to place the message on people’s devices. Location data or an internet address may help advertisers reach people who are not already present in the campaign’s own files. (Electronic Frontier Foundation)

Reuters reported in 2020 that the two major U.S. parties were working with national databases covering more than 200 million voting-age Americans. Those systems combined voter files with information from multiple sources and used models containing thousands of data points to estimate voters’ political positions and likely behavior. (Reuters)

The scale of these systems does not mean every data point is accurate. It means that campaigns can work with large populations and make decisions based on probabilities. A voter may be treated as part of a category because the data suggests a pattern, even when the classification is wrong.

That distinction matters. The election machine does not need to know exactly who a person is. It only needs to decide whether that person belongs in an audience worth reaching.

Where this story sits

5 related pieces

Hover a story · click to read

1. No. 36 · Sep 5, 2026The Army Is Turning Military Land Into an AI Compute Platform
2. No. 17 · Aug 27, 2026The Kidnappers Outsmart Nigeria’s Surveillance State
3. No. 55 · Sep 24, 202610 Gigawatts Overhead: The Orbital Data Centers the FCC Reviews One Filing at a Time
4. No. 62 · Oct 7, 2026The Authorized Leak: When Lawful Access to National ID Systems Becomes the Vulnerability
5. No. 57 · Sep 28, 2026The ShinyHunters FBI Breach: When Stolen Personnel Files Become Leverage Against the State

From a profile to a message

Once a campaign has divided voters into groups, it can vary more than the audience.

It can test different images, arguments, videos, emotional appeals, and calls to action. One audience may see a message about taxes. Another may see a message about abortion, public safety, immigration, climate policy, or threats to democracy. A campaign may emphasize turnout among supporters, persuasion among undecided voters, or fundraising among people likely to donate. Research on political microtargeting describes this broader practice of matching messages or campaign objectives to selected groups. (Tappin et al., PNAS; Fowler et al., American Political Science Review)

Traditional television advertising made it difficult to deliver every message to a narrowly defined individual. A campaign could choose a television program, station, market, or time slot, but many different kinds of viewers would see the same commercial.

Digital advertising makes narrower targeting possible. Research comparing Facebook and television advertising during the 2018 election found that Facebook attracted a broader set of candidates, particularly in lower-level races. Among candidates who used both forms of advertising, Facebook advertisements appeared earlier in campaigns, were less negative, less focused on issues, and more partisan than television advertisements. (Fowler et al., American Political Science Review)

The shift is not simply from one screen to another. It changes what campaigns can attempt. An online ad can be directed at a smaller group, tested quickly, and changed when the results appear weak.

That does not guarantee effective persuasion. It creates more opportunities for experimentation.

The campaign chooses the audience, then the algorithm intervenes

The most important part of the system is often the least visible.

A campaign may tell a platform to reach a particular group. The platform then uses its own delivery system to determine which eligible users receive the ad and how much the advertiser pays to reach them. Research describes these systems as responding to factors involving the advertiser, the ad, the selected audience, and the platform’s automated predictions about delivery and value. (Votta et al., “The Cost of Reach,” Political Communication)

The campaign’s selected audience and the audience that actually receives the ad may therefore differ.

A study published in 2024 examined more than 80,000 political advertisements shown during Germany’s 2021 federal election and more than 1.1 billion impressions on Meta platforms. The researchers found substantial discrepancies between targeted audiences and actual audiences. They also found systematic differences in the reach of ads among parties, including evidence that the platform’s delivery system favored advertisements from populist parties in the data they analyzed. (Hagen et al., “Systematic Discrepancies in the Delivery of Political Ads on Facebook and Instagram,” PNAS Nexus)

The finding does not establish that every election platform favors every populist campaign, or that the result would be identical in every country or election. It does show why campaign targeting cannot be understood by looking only at the advertiser’s settings. The platform’s optimization system is part of the political communication process.

A 2024 preregistered study of political advertising on Facebook and Instagram reached a related conclusion. Researchers placed 135 identical ads for three Dutch political parties, targeting nine audiences during the 2022 municipal elections. The ads used identical budgets, text, images, and timing. Even under those controlled conditions, the price of reaching audiences differed. The study found that lower-educated citizens, women, and younger adults aged 18 to 24 were more expensive to reach in the tested campaigns. (Votta et al., “The Cost of Reach,” Political Communication)

This creates a democratic problem that is different from ordinary campaign messaging. If a platform makes certain groups more expensive to reach, a campaign with a fixed budget may reach fewer of those voters. If its delivery system directs an ad toward people predicted to respond, the campaign’s stated audience may become only a rough instruction.

The campaign chooses. The system filters.

Who gets left out?

Digital targeting is often described as a way to reach the right people. It can also determine who is not reached.

Campaigns may exclude people who are unlikely to vote, unlikely to donate, or unlikely to respond. Data-driven systems can also make some voters harder to reach when their records are incomplete, inaccurate, or absent from a campaign’s files. The available research does not establish that every such exclusion is deliberate discrimination. It does show that advertiser-selected audiences and actual audiences can diverge. (Electronic Frontier Foundation; PNAS Nexus)

Some voters may see repeated appeals from several campaigns. Others may see little political advertising at all. Some groups may receive policy information, while others receive fundraising requests or messages designed to encourage a particular response. The campaign’s objective and the platform’s delivery system can shape which of those messages reaches a given person. (Tappin et al., PNAS; Votta et al., “The Cost of Reach,” Political Communication)

The practical effect can still matter. Voters do not experience an election as one shared stream of arguments. They experience different selections of political information.

The study of German political ads found that the actual audiences receiving ads differed from the audiences advertisers intended to target. The Dutch experiment likewise showed that the cost of reaching groups could vary even when the advertising content and budgets were held constant. Those findings make it difficult to treat digital political advertising as a neutral pipe between campaigns and voters. (PNAS Nexus; Political Communication)

Does microtargeting actually persuade voters?

The strongest claims about political data often concern persuasion. If a campaign knows enough about a voter, the argument goes, it can deliver exactly the message needed to change that person’s mind.

The evidence is more limited.

A 2023 study in PNAS tested whether machine-learning-based microtargeting produced a persuasive advantage over other messaging strategies. The researchers used survey experiments involving U.S. issue advocacy. In one context, their microtargeting strategy produced an estimated persuasive effect at least 70 percent larger than the comparison strategies. But the result depended heavily on the issue and the conditions. In a second study, where the targeting system chose which policy attitudes to address, the advantage was much smaller. The researchers also found no evidence that targeting people using more than one characteristic produced additional persuasive gains. (Tappin et al., “Quantifying the Potential Persuasive Returns to Political Microtargeting,” PNAS)

That is evidence for a conditional advantage, not proof that detailed personal data reliably changes elections.

The researchers emphasized that the effects depended on the context. Political messages often affect different groups in broadly similar ways, which limits the value of matching a different message to every individual. A campaign can spend heavily to identify fine-grained differences without finding enough meaningful variation to justify the cost. (Tappin et al., “Quantifying the Potential Persuasive Returns to Political Microtargeting,” PNAS)

This is one of the strongest pieces of evidence against the most dramatic version of the election-machine theory. Microtargeting can sometimes improve persuasion, but it is not a universal technology for controlling voters.

The system may still be useful even when persuasion is weak. Campaigns may use data to identify likely supporters, find people who need reminders, prioritize door-knocking, solicit donations, or decide where to spend staff time. A tool does not need to change many minds to affect turnout operations or campaign spending. (Electronic Frontier Foundation; Tappin et al., “Quantifying the Potential Persuasive Returns to Political Microtargeting,” PNAS)

Data can matter without changing minds

Political campaigns use information for more than persuasion.

A campaign may want to estimate who is likely to vote, who needs a reminder, who might volunteer, who might donate, or who could persuade friends and family. A voter profile can influence which people receive a text message, phone call, mailer, canvassing visit, donation request, or invitation to an event. (Electronic Frontier Foundation)

These uses are difficult to measure from public advertising data alone. A platform’s ad library may show an advertisement, its sponsor, spending range, dates, and estimated reach, but it may not reveal the full chain of data used to select the audience. It may also provide little information about the people who were considered and excluded.

The public can inspect some campaign messages through tools such as Meta’s Ad Library and Google’s Ads Transparency Center. These repositories make some political advertising more visible, but they do not expose every element of a campaign’s internal data model or the platform’s delivery logic.

The Wesleyan Media Project’s work illustrates the value of public tracking. The project has monitored political advertising since 2010 and has used Meta and Google platform libraries to track digital and social-media advertising since 2018. It describes its purpose as making information about political advertising available for public and scholarly analysis. (Wesleyan Media Project)

That visibility remains partial. A voter may be able to identify who paid for an ad without knowing why the platform delivered it, what data category placed the voter in the audience, or how many different versions of the message were shown to other people.

The privacy problem is larger than the ad

Political targeting depends on information that may reveal much more than a person’s likely vote.

Campaign records can connect names, addresses, political affiliations, voting participation, donations, online behavior, location signals, consumer characteristics, and inferred interests. Some data comes from public records. Some comes from commercial sources. Some is collected directly by campaigns. Some is generated by automated predictions. (Electronic Frontier Foundation; Reuters)

The danger is not limited to a campaign making an inaccurate assumption. Political data can expose sensitive associations, reveal participation in political causes, or make people easier to target or harass. The EFF notes that voter records can be available for purchase in most states and that online access can create risks such as doxing. (Electronic Frontier Foundation)

The EFF also describes political data brokers that offer campaign-oriented categories and predictive analytics. In 2020, OpenSecrets found that political groups paid at least $23 million to 37 data brokers for access to services or data. (Electronic Frontier Foundation)

The rights available to voters vary by jurisdiction and by the organization handling the information. That variation makes it difficult for a person to determine which data was collected, which company supplied it, which model produced an inference, or how to correct a classification.

The practical lesson is that a campaign may know more about a voter than the voter can discover about the campaign’s record of them.

Can voters tell why they saw an ad?

Usually, voters can learn less than they want to know.

A political ad may carry a sponsor disclaimer identifying the candidate, party, committee, or organization that paid for it. Platform libraries can provide additional information about an ad’s dates, estimated spending, or reach.

That answers one question: who paid?

It does not necessarily answer the more important questions:

• Which data category put me in this audience?
• Did the campaign select me, or did the platform’s algorithm find me?
• What other messages did different voters receive?
• Was the campaign trying to persuade me, mobilize me, raise money, or collect more data?
• How many people were excluded from the same message?
• Was the ad shown because of my location, browsing behavior, consumer profile, political record, or an inferred interest?

The transparency gap is partly technical. Advertising systems are proprietary. It is also institutional. Campaign-finance disclosure rules were built largely around money, sponsors, and communications, not the full data pipeline behind individualized delivery.

The gap affects accountability. A television commercial can be recorded, archived, and discussed by a broad audience. A narrowly targeted digital ad may be seen by a small group and disappear before journalists, opponents, or researchers can examine it.

Political-ad libraries reduce that problem, but they do not solve it. They are records of visible advertisements, not complete records of political influence.

The law follows money more easily than data

Campaign-finance law generally asks who paid, how much was spent, and whether spending was properly disclosed. Those questions remain important, but they do not fully describe digital persuasion.

A campaign can disclose the cost of an advertisement without revealing the data categories used to target it. A platform can publish an ad in a searchable library without explaining the factors that determined delivery. A data broker can supply an inferred characteristic without appearing in the advertisement’s sponsor disclaimer.

This creates a chain of responsibility with no single actor controlling the whole process:

• The campaign decides what it wants to accomplish.
• A consultant or data firm may help define the audience.
• A broker may supply or infer information.
• The platform decides how to deliver the ad.
• The voter sees the result, often without knowing the decisions that produced it.

Regulators and platforms have responded with disclosure rules, ad libraries, political-ad policies, privacy rights, and restrictions on certain kinds of targeting. Their approaches vary by country and platform, and they change over time.

The available evidence suggests that transparency has improved since the early days of digital campaigning, but public visibility remains incomplete. Researchers can now study large collections of political ads and actual delivery outcomes, yet important decisions still occur inside private databases and proprietary systems. (PNAS Nexus; Wesleyan Media Project)

What the evidence supports

The evidence supports a narrower and more defensible account of the election machine.

Campaigns do use voter files, consumer data, direct interactions, online signals, and inferred characteristics to classify voters and guide communications. Political organizations can combine information from multiple sources, share data within political networks, and buy campaign-oriented services from data brokers. (Electronic Frontier Foundation; Reuters)

Digital platforms allow campaigns to target narrower groups than television usually permits. They also use automated delivery systems that can alter the audience, price, and reach of an advertisement. Research from Germany and the Netherlands shows that the people a campaign intends to reach may not be the same people who receive the message, and that some groups can cost more to reach than others. (PNAS Nexus; Political Communication)

Campaign data can improve some forms of political communication, including identifying supporters, organizing turnout, and selecting audiences. Research also finds that microtargeting can produce a persuasive advantage under particular conditions. (Tappin et al., PNAS)

But the evidence does not show that personal data gives campaigns reliable control over voters, that every targeted message is persuasive, or that microtargeting alone determines election outcomes. The persuasive benefits can be small, inconsistent, and dependent on the issue, message, audience, and delivery system.

The strongest concern is therefore not that campaigns possess a perfect instrument of political control. It is that political communication is being organized by a system in which data collection, inference, audience selection, and automated delivery are difficult for voters to see.

The machine’s real power is deciding what becomes visible

The political consequences of targeting may begin before a voter changes their mind.

They begin when one voter sees a message and another does not. They begin when a platform makes one group cheaper to reach than another. They begin when a campaign chooses to speak about an issue only to a narrow audience. They begin when an inferred profile determines whether someone receives a turnout reminder, a donation request, a fear-based appeal, or nothing at all.

The system does not need to persuade everyone to matter. It only needs to shape the information environment in which voters make decisions.

That environment is now split among campaigns, data firms, platforms, regulators, researchers, and voters. Campaigns know what they want to test. Data firms know how they classify people. Platforms know how their delivery systems behave. Voters usually see only the final message.

That imbalance is the lasting problem. Political data may help campaigns organize and communicate more efficiently, but democratic accountability requires more than efficient delivery. Voters need to know who is speaking, what information is being used to reach them, how messages differ across groups, and which parts of the process are controlled by a campaign versus an opaque platform.

The election machine is real, but its power is less magical than often claimed. It does not reliably predict or command human behavior. It does something more ordinary and more consequential: it sorts people, assigns probabilities, allocates attention, and determines which political messages become visible to whom.

That is enough to change the campaign. It is also enough to require public scrutiny.

Sources / References

• Electronic Frontier Foundation. “How Political Campaigns Use Your Data to Target You.” April 16, 2024. https://www.eff.org/deeplinks/2024/04/how-political-campaigns-use-your-data-target-you
• Culliford, Elizabeth, and Minami Funakoshi. “How Political Campaigns Use Your Data in Elections.” Reuters, October 12, 2020. https://www.reuters.com/graphics/USA-ELECTION/DATA-VISUAL/yxmvjjgojvr/
• Votta, Fabio, Tom Dobber, Benjamin Guinaudeau, Natali Helberger, and Claes de Vreese. “The Cost of Reach: Testing the Role of Ad Delivery Algorithms in Online Political Campaigns.” Political Communication, published online December 23, 2024. https://www.tandfonline.com/doi/full/10.1080/10584609.2024.2439317
• De Francisci Morales, Gianmarco, et al. “Systematic Discrepancies in the Delivery of Political Ads on Facebook and Instagram.” PNAS Nexus, vol. 3, no. 7, 2024. https://pmc.ncbi.nlm.nih.gov/articles/PMC11229815/
• Tappin, Ben M., Chloe Wittenberg, Luke B. Hewitt, Adam J. Berinsky, and David G. Rand. “Quantifying the Potential Persuasive Returns to Political Microtargeting.” Proceedings of the National Academy of Sciences, 2023. https://www.pnas.org/doi/10.1073/pnas.2216261120
• Fowler, Erika Franklin, Michael M. Franz, Gregory J. Martin, Zachary Peskowitz, and Travis N. Ridout. “Political Advertising Online and Offline.” American Political Science Review, published online August 19, 2020. https://www.cambridge.org/core/journals/american-political-science-review/article/abs/political-advertising-online-and-offline/9E24E81AC74E4644494FF451D5373B71
• Wesleyan Media Project. “Project Background.” https://mediaproject.wesleyan.edu/about/project-background/
• Meta. “Ad Library.” https://www.facebook.com/ads/library/
• Google. “Ads Transparency Center.” https://adstransparency.google.com/