Fundamentals

Before you pick a channel, do the math: how many people will vote, how many votes you need, with a margin, and where they will come from. Your theory of victory is the sentence that ties that math to the plan: 'we win if our base turns out, if we bring in X supporters who often skip elections and if we win over Y undecided voters who do vote'. Anything that doesn't serve that sentence is secondary.

• Vote goal = registered voters × expected turnout × share needed to win, plus a safety margin of 5–10%.
• The share depends on the system: a majority in a runoff, first place on a proportional list, the seat in a single-member district, 50% + 1 in a referendum.
• Split the goal across three sources: a base that already votes, supporters who may not show up (mobilization) and undecided people who vote (persuasion).
• Use precinct-level results from past elections to find where each source lives.
• Test your theory early, with data: if the new electorate doesn't show up in the first measurements, change the plan before the end.
• In the cases: in Brazil 2026 the math was overturning a 2.2 million vote deficit in three weeks; in Barcelona 2015, coming first with a quarter of the vote and having partners to govern with.

For each group of voters, ask two questions: do they support you? Will they vote? The answer gives four audiences, each with its own list, script and goal. Mixing them up is a campaign's most expensive mistake: spending persuasion on people already convinced, or reminding people to vote who will vote for your opponent.

• Supports and votes (the base): doesn't need convincing; ask them for work, money and conversations with people they know.
• Supports but may not vote: mobilization (reminders, a voting plan, transport, timing).
• Undecided but votes: persuasion (a trusted messenger, a concrete issue, listening).
• Firm opponent: don't spend time on them; when you respond, speak to the undecided people listening.
• Where there is a voter file with turnout history (as in the US), your lists come from it; where there isn't, use precinct results, polls and what you hear at the doors, knowing it's an estimate.
• In general elections, contact barely persuades (see 'What the research says'); in primaries, referendums and with infrequent voters, it matters more.
• Undecided doesn't mean centrist: often it's someone with little interest in politics. Talk about everyday life, not ideological position.

A message is a choice: what you say, to whom, and what you leave out. Discipline is repeating that choice more often than seems necessary, with different examples and different voices. If you talk about everything, you're remembered for nothing.

• One contrast line: why you and not them, in the words your audience uses.
• Three proof points at most: concrete, checkable promises tied to the lives of the people you want to reach.
• The messenger matters as much as the message: doctors, neighbors, coworkers, people with the real story.
• Research before you write (focus groups, conversations at the doors, comments) and test before you scale.
• Repeat: when your team is sick of the message, the public is just starting to hear it.
• Every response to an attack ends on your issue, not your opponent's.
• In the cases: Mamdani built everything around three cost-of-living promises; in Senegal, 'Diomaye is Sonko' transferred trust from the leader to his substitute in a single sentence.

Signing people up is easy; turning them into people who do things is the work. The ladder is a sequence of ever bigger asks, and the campaign measures the step from one rung to the next, not the total number of sign-ups.

• Sign-up → contact within 24 hours → small first task → second task → named role → team leader.
• Fast contact: people who aren't called within a few days go cold. A personal message beats an automated email.
• The first task has a clear start and end and a visible result.
• Roles with a name, hours and training: neighborhood captain, shift coordinator, data lead.
• Leaders recruit and train others: that's how the network grows without the central team becoming a bottleneck.
• Look after the people you have: thank them, show the effect of their work, offer rest. Burnout is a limit in almost every case.
• In the cases: in 2016 Sanders' volunteers had no titles or management roles; in 2020 the campaign created captains with set hours, training and a coach.

There are four ways to reach people, and nearly every winning campaign uses all four. Owned media is yours (list, channel, website); earned media is given by others (press, creators, word of mouth); paid media is bought (TV, radio, billboards, ads); shared media is what supporters post and forward.

• Owned first: an email list and a messaging channel you control survive algorithm changes and takedowns.
• Earned: the press still sets the agenda for many people. Have spokespeople, checkable facts and actions designed to make news.
• Paid: TV hasn't gone away. The Obama campaign spent about $280 million on TV ads between January 2007 and November 2008 (TNS/CMAG, cited by FactCheck.org), and Mamdani's campaign ran its first broadcast TV ad on April 24, 2025. Organic came before paid, not instead of it.
• In the EU, since October 2025, Meta no longer delivers political, electoral or social-issue ads and Google no longer serves political ads; newspapers, radio, billboards and other media can still sell them, with a label and a transparency notice.
• Shared: ask each person to say what they think in their own words. Identical texts posted at the same time by many accounts look like coordinated inauthentic behavior.
• Creators paid to carry a political message are doing advertising: disclose the partnership.

Views and reactions tell you something was seen, not that it changed anything. Measure at three levels and test before you scale: a campaign that compares learns; one that only counts repeats its mistakes.

• Attention: views, reach, reactions. Useful to know whether a piece is spreading; not a goal.
• Action: what happens off the screen (sign-ups, shifts done, voting plans, donations, people brought to the polls). This is the goal of missions.
• Outcome: turnout and votes by precinct, compared with the previous election.
• A/B test: two versions of a message to similar, randomly assigned groups; compare action, not reaction.
• Control group: leave some similar precincts or lists uncontacted and compare afterwards. It's the only way to know whether contact changed anything.
• UTM-tagged links and forms tell you where each sign-up came from.
• Canvass data shows trends, it doesn't predict results: the doors that open aren't a random sample.
• Keep campaign numbers, official numbers and platform numbers apart, and always say where each one comes from.

There's no single rule. Silence works for attacks that haven't reached your audience yet; correction is for when the rumor is already circulating among the people who decide; prebunking is for attacks you know are coming. Research says correcting works and backfire is rare; what doesn't work is spending the campaign reacting.

• Listen first: how far it has spread, who is repeating it, and whether it has reached undecided people or only one side's true believers.
• Small rumor, outside your audience: don't amplify it. Quoting it gives it reach.
• Rumor that has reached your audience: respond quickly: fact first, the rumor only once, why it's false, the fact again. Then go back to your issue.
• Respond in the format the rumor travels in: image, short audio, vertical video, on the same network.
• Trusted messengers correct better than press releases: a doctor, a neighbor, an independent fact-checker.
• Predictable attack: inoculate in advance. Explain the technique that will be used ('they'll say that…, and here's how to spot it'), not just the fact.
• Never respond with rumors, fake profiles or insinuation.
• In the cases: in Taiwan, each ministry answered trending rumors within two hours, with two images, very short text and humor.

The last weeks aren't for convincing: they're for getting your people to vote and making sure every vote is counted. These are two operations, with different teams: mobilization (who, when, how they get there) and protection (poll watchers, tally sheets and your own count).

• Ask for a voting plan: what time, leaving from where, going with whom. In a US experiment, asking for a plan raised turnout by 4.1 points among those contacted.
• Say how many people will vote, not how many will stay home: 'lots of people vote' motivates more than 'lots of people don't'.
• Remove obstacles: transport, polling place, ID, changing your registered address, deadlines for early or overseas voting.
• Accredited poll watchers at the polling stations that matter, trained on the rules and with a legal contact.
• At close, each watcher photographs the signed tally sheet for their polling station (never anyone's ballot) and sends it to a center that adds them up and compares with the official numbers.
• Independent observers compare and publish on their own: don't coordinate with them, because their credibility depends on not being on anyone's side.
• Prepare for election night and the day after: who speaks, when, and what to do if the numbers don't match.

Organizing exposes people. In repressive settings the risk is arrest, surveillance and violence; in democracies it's harassment, doxxing and threats, especially against women candidates and volunteers and people from minorities. Safety is part of the plan from day one, not a reaction to the first incident.

• Map the risks by person and by role: who is public-facing, who holds data, who goes out on the street.
• Accounts: two-factor authentication on all of them, a password manager, more than one admin and a backup channel.
• Numbers: in a regular WhatsApp group, everyone sees everyone's number. Use channels for announcements; for sensitive coordination, Signal with a username and hidden number.
• On the street and at the doors: pairs, a contact person, an incident protocol and a record of every case.
• Online harassment: nobody responds alone. Document it with dated screenshots, report it on the platform and have someone on the team filter messages for the person targeted.
• Repressive settings: a protest phone (no face or fingerprint unlock, location off, disappearing messages), legal and medical support, and no free VPNs of unknown origin.
• Data: keep only what you need, for as long as you need it. A leaked list puts everyone on it at risk.
• The kit's share link carries your written plan inside it: in restricted settings, don't send it through apps the government controls. In the library, Security in a Box has step-by-step guides.

Almost everything these playbooks report comes from the people who ran the campaigns. The findings below come from experiments and studies, and help separate what works from what only looks like it works. An honest caveat: there are almost no experiments on WhatsApp channels, Discord servers or relational organizing at scale, and most studies are from the US.

1. In general elections, campaign contact barely changes votes

   An analysis of 49 field experiments in the US put the average persuasive effect of campaign contact and advertising on candidate choice in general elections at zero. There were effects in two rare situations: when a candidate takes very unpopular positions and the other side invests in identifying persuadable voters, and when contact comes very early (and then the effect fades). For organizers: in general elections, use contact mainly to mobilize; in primaries, referendums and low-information races, the math may differ. Caveat: US data.

   Kalla and Broockman, American Political Science Review, 2018 (SSRN preprint, 2017)
2. Asking for a voting plan raises turnout

   In an experiment with 287,228 people during the 2008 US presidential primaries, calls that asked for a plan (what time, from where, how) raised turnout by 4.1 points among those contacted, and by 9.1 points in single-voter households; a standard encouragement call had no effect. For organizers: replace 'don't forget to vote' with questions that get the person to build a plan. Caveat: one experiment, by phone, in a voluntary-voting country.

   Nickerson and Rogers, Psychological Science, 2010
3. 'Lots of people vote' motivates more than 'lots of people don't'

   In two get-out-the-vote experiments, in California and New Jersey, messages announcing high turnout motivated people more than messages lamenting low turnout, especially among infrequent voters. For organizers: say how many will vote, not how many will stay home. Caveat: the studies measured stated motivation to vote, not validated turnout.

   Gerber and Rogers, Journal of Politics, 2009
4. Correcting rumors works; backfire is rare

   This synthesis by misinformation researchers concludes that the backfire effect (a correction strengthening the rumor) happens only occasionally, and advises not to hold back corrections for fear of it; repeating the myth while refuting it is generally safe. For organizers: silence is a tactic for small rumors, not a rule. Caveat: it summarizes studies; it doesn't measure effects on votes.

   The Debunking Handbook 2020 (Skeptical Science)
5. Inoculating in advance helps people spot manipulation

   90-second videos teaching people to recognize techniques such as emotional language and false dichotomies were shown as ads to about 5.4 million YouTube users in the US; among those who answered the follow-up question, the ability to recognize the techniques rose by about 5% on average. In the lab, people who saw the videos were about twice as good at spotting false dichotomies. For organizers: anticipate the technique your opponent will use, not just the fact. Caveat: the field effect is small, and it measures recognition, not votes.

   Roozenbeek et al., Science Advances, 2022 (University of Cambridge)
6. Field offices mattered in 2008

   A study of 11 battleground states found a disproportionate increase in the Democratic vote in counties with Obama field offices, large enough to flip three states. For organizers: organized local presence counts, even in an election with lots of TV and digital. Caveat: it's an observational study, not an experiment; the campaign chose where to open offices.

   Masket, Public Opinion Quarterly, Winter 2009
7. Swapping stories changes attitudes; arguments alone don't

   In three pre-registered experiments, 230 canvassers talked with 6,869 voters in seven US locations about undocumented immigrants and transgender people. Conversations with a non-judgmental exchange of personal stories reduced exclusionary attitudes (in the first experiment, for at least four months); identical conversations using arguments alone had no effect. For organizers: on identity issues, ask for and tell stories, not just facts. Caveat: the average effect is small, and it concerns attitudes, not votes for a candidate.

   Kalla and Broockman, American Political Science Review, 2020

• Nobody trusts the messenger

  These methods amplify a voice that already has credibility: Janones had his own audience, and Mamdani and Obama drew people in themselves. If the face of the campaign is unknown or discredited among the audience that decides, missions and videos just spread distrust faster. Start with borrowed messengers (doctors, neighbors, local leaders) and smaller goals.

  Brazil 2026 New York 2025 United States 2008 Ireland 2018

• There's no structure off the screen

  The cases that turned into votes had years of organizing behind them: in New York, an organization with nearly a decade of electoral experience; in Ireland, six years of local groups; in Sri Lanka, decades of activist networks. A new channel with no one on the doors, in the neighborhoods and at the polling stations produces noise, not votes. If you only have weeks, pick a few precincts and do them well.

  New York 2025 Ireland 2018 Sri Lanka 2024 United States 2016–2020

• Heavy repression without a safety plan

  Where there are arrests, surveillance and violence, public lists, open groups and known faces expose people. Without legal, medical and digital support, and without rules for data, the method puts lives at risk before it wins anything. In Serbia, Kenya, Bangladesh and Georgia, the human cost was real.

  Serbia 2026 Kenya 2024 Bangladesh 2024 Georgia 2024–2026

• The deciding audience isn't online

  If the people who decide are older, rural or live in closed groups you can't reach, digital talks to the already convinced. In Barcelona, digital didn't reach everyone; in Taiwan, encrypted groups stayed dark to fact-checkers. Combine radio, local press, door-knocking and messengers from each community.

  Barcelona 2015 Taiwan 2018–2021 Poland 2023

• Compulsory or voluntary voting changes the math

  Where voting is voluntary, winning often means bringing in people who don't vote: in Chile in 2021, about 1.25 million more people voted in the runoff. With compulsory voting, which Chile brought back at the end of 2022, that reservoir almost disappears and the work becomes persuasion and removing obstacles. In Brazil, where voting is compulsory for most adults, the final stretch is mostly logistics.

  Chile 2021 Brazil 2026 Poland 2023

• The government can switch off the internet

  A VPN gets around a platform block, not a total shutdown. In Bangladesh, the shutdown stalled coordination and hid the violence for days. Set up offline channels in advance: phone, SMS, radio, agreed meeting points and people who carry the message by hand.

  Senegal 2024 Bangladesh 2024 Nepal 2025

• The political context is against you

  Mobilization can't make up for everything: with about twice as many people on the doors, Labour lost seats in 2019; Sanders lost two primaries with millions of calls and texts; Fridays for Future filled the streets without winning the laws it demanded. If the economy, the candidate or the issue is against you, the method narrows the defeat or builds for later. Promise that, not victory.

  United Kingdom 2017 United States 2016–2020 Global (began in Sweden) 2018–2019

• You need to persuade, but you only reach the converted

  Campaign channels and spectacular actions are followed by people who already agree. If the election is decided among undecided people who vote, contact yields little in general elections and shock actions can give the target a stage. In Ireland in 2015, door-knocking raised turnout but didn't change votes; persuasion came from messengers and stories.

  United Kingdom 2019–2026 Ireland 2015 Ireland 2018 Brazil 2026

Studying only the winners misleads: the same methods were used by people who lost, and results also depend on the candidate, money, the economy and the opponent. That's why this site includes defeats and partial wins and reads them with the same care. The Sanders campaign (Big Organizing) lost the 2016 and 2020 primaries. Momentum helped Labour strip the Conservatives of their majority in 2017, but not form a government, and in 2019, with about twice as many people on the doors, the party fell to 202 seats. In Georgia, more than 500 days of daily protest have not brought new elections. In Chile, Boric won in 2021 and the same opponent, Kast, won in 2025. Fridays for Future filled the streets without getting the laws it demanded, and in Sri Lanka the people who made the Aragalaya were not the ones who came to power. Barcelona en Comú won without a majority of its own, and in Poland record turnout didn't stop PiS from coming first. The playbooks from the other side (AfD, Bolsonaro, Georgescu, Milei, Trump) are here to study methods that worked for opponents (and, in Georgescu's case, an annulled election), never to copy what is inauthentic in them. Read each case like this: what happened, what part is down to the method and what part to context, and what the limits say. If a method only ever shows up in victories, be suspicious.

Since October 2025, paid campaigning in the EU has changed: the law requires labels and transparency for all political advertising, and the two biggest ad platforms stopped selling it. This sheet sums up what applies across the Union; each country adds its own rules (election silence, spending limits), which the country selector shows. This isn't legal advice: check with a lawyer.

• Regulation (EU) 2024/900 on political advertising

  Fully applicable since October 10, 2025, it covers online and offline political advertising in European, national and local elections and referendums. Political advertising is a paid message, or one made on behalf of a political actor, that is liable to influence an election, a referendum, voting behavior or a legislative process: issue campaigns by NGOs, unions and movements are covered too, even without naming a party. ↗

• Label and transparency notice

  Every paid political ad must say that it is political advertising, who sponsors it and which election or process it relates to, and lead to a transparency notice that includes, among other things, how much it cost. Publishers keep notices for seven years. The label format is set by a 2025 implementing regulation, and on October 12, 2026 the Commission launched a pilot of the European repository of online political ads. ↗

• Targeting only with explicit consent

  Online, personal data may be used to target or deliver political ads only if it was collected from the person themselves, with explicit consent given separately for political advertising. Profiling based on sensitive data, such as political opinion, religion or ethnic origin, is banned for this purpose. ↗

• Non-EU sponsors barred in the three months before a vote

  In the three months before an election or referendum, political advertising services may only be provided to sponsors who declare that they are EU citizens, foreign residents entitled to vote in that election, or EU entities not controlled from outside the Union. Campaigning from outside the EU doesn't exempt you: ads that reach voters in the EU follow the same rules. ↗

• Meta stopped political ads in the EU

  Since October 6, 2025, Meta has not delivered political, electoral or social-issue ads in the EU. Politicians and candidates can still post political content; they just can't boost it with ads. Outside the EU, Meta's political ad tools continue. ↗

• So did Google, including YouTube

  Google announced in November 2024 that it would stop serving political ads in the EU, including on YouTube, because of the regulation, and applied the decision in October 2025. Its current ads policy restricts political advertising on its platforms as the regulation defines it; official messages about how to vote can apply for an exemption. ↗

• What is still allowed

  The law doesn't ban ads: newspapers, radio, TV, billboards and local media can still sell them, with a label and a notice. And it leaves organic activity out: your own unpaid posts, messaging channels, websites, opinions expressed in a personal capacity and volunteer work. Creators paid to carry a political message, on the other hand, are doing advertising and need a label. Campaign emails and texts need prior consent and an easy way to unsubscribe. ↗

• DSA: big platforms must reduce election risks

  The Digital Services Act requires very large online platforms and search engines to assess and reduce risks to electoral processes; the Commission's guidelines of April 26, 2024 explain how. This shapes labeling, reduced reach and removal of manipulated content, and the Commission opened formal proceedings against TikTok over Romania's 2024 election. For campaigns: report manipulation through the platform's channels and keep evidence, and remember that your own inauthentic coordination is a target too. ↗

• GDPR: political opinion is sensitive data

  Data revealing political opinions may only be processed with explicit consent for specified purposes or, for parties, associations and other political not-for-profit bodies, in relation to their members and regular contacts, as part of their activities and without sharing it outside. Legitimate interest alone isn't enough for this data, and what a person has made public is read narrowly: it doesn't allow you to infer opinions from who someone liked or followed. ↗

• The campaign is accountable for the data it uses

  Whoever decides what the data is for (party, candidate or movement) is the data controller and must be able to show it follows the rules. Tell each person why they're getting a message, who is sending it and how to exercise their rights, including when the data came from third parties. Profiling to target campaign messages requires, in principle, explicit consent. Platforms, data brokers and agencies are supervised too. ↗

Vote goal

How many votes you need to win: registered voters × expected turnout × share needed, plus a margin.

Theory of victory

The sentence that says where the missing votes come from and what work brings them in.

Universe

The list of people an action is aimed at: for example, supporters in one area who often skip elections.

Base

People who already support you and usually vote; your source of volunteers, money and conversations.

Persuasion

Work to change the choice of people who will vote but haven't decided.

Mobilization (GOTV)

Get out the vote: work to make sure supporters who might not show up actually vote.

Turnout propensity

The estimated likelihood that a person will vote, based on their history or on polls.

Voter ID

Recording, at each contact, the person's support on a scale (for example 1 to 5) to build the next lists.

Canvassing

Face-to-face conversations at voters' doors, with a script and a record of every answer.

Phone bank

Calling sessions run by volunteers, in person or remotely, with a script and a list.

Peer-to-peer texting

Texts sent one by one by volunteers, who can reply; not a mass blast, and it needs consent where the law requires it.

Relational organizing

Asking each supporter to talk to people they already know, and recording those conversations.

Distributed organizing

Volunteers doing staff-level work with simple tools and a public central plan.

Barnstorm

A mass meeting that ends with teams formed and tasks scheduled, not just speeches.

Leadership ladder

The sequence of ever bigger asks that turns a sign-up into a team leader.

Message box

A four-quadrant grid: what we say about us and them, what they say about us and themselves.

Poll watcher

A person accredited to observe voting and counting at a polling station on behalf of a candidate or list.

Tally sheet

The signed record of each polling station's result; a photo of it is the evidence that lets you challenge a count.

Parallel count

Your own sum of each polling station's results, built from the tally sheets, to compare with the official numbers.

Election silence

A period before the vote when some countries' laws ban campaigning or polls; the rules are national.

Political advertising

In the EU, a paid message, or one made on behalf of a political actor, that can influence an election, a referendum or legislation; it needs a label and a notice (Regulation (EU) 2024/900).

Coordinated inauthentic behavior

Fake or coordinated accounts used to mislead about who is speaking or how many there are; grounds for removal on every major platform.

UTM

Tags in a link that tell you which piece or channel each click or sign-up came from.

Special-category data

Under the GDPR, data revealing political opinions, religion, health or ethnic origin, among others; it can only be processed with explicit consent or a narrow exception.

Prebunking

Inoculating in advance: explaining a manipulation technique or a predictable attack before it reaches the audience.