Deep Canvassing: The durable effect that isn't a message

In brief

A single conversation at the door can leave a voter's attitude measurably shifted for months, but only by a small amount. The part that appears to work is hearing a concrete personal story about a member of a group; arguments alone showed no detectable effect, and adding non-judgemental listening did not enhance persuasion. The effect has limits: it faded after a counter message and by six months, and every field test so far has been American. The measured shift is around d = 0.08, which is real but modest.

How to use this

Before writing a canvass script, find a concrete personal story from or about a member of the group you want to move. Make that story the core of the conversation and drop the argument script. Do not add non-judgemental listening training expecting it to boost persuasion. Run many conversations and re-test; the shift from any single conversation is small. If you are funding or regulating a programme, keep it provisional, pilot it, and publish the results, including any that trim the claim. Ask whether each choice serves the persuasion you are trying to do or the conclusion you already preferred.

What the story is about

The founding experiment ran in real neighbourhoods in South Florida. Canvassers spoke with 501 people about transgender rights, and three months later their attitudes were still shifted, compared with people who got a different conversation (Broockman and Kalla, 2016). Whether an argument alone would do the same job was tested in three pre-registered field experiments, with the analysis plan fixed in advance. The first randomised one ingredient: whether the conversation included an exchange of personal narratives. Everything else stayed the same. Arguments on their own had no detectable effect on exclusionary immigration attitudes. Otherwise identical conversations that added the exchange durably reduced those attitudes for at least four months, at d = 0.08. It has since held across different stories, places and formats, and when run by organisations with little prior experience (Kalla and Broockman, 2020). An effect size weighs the average shift against how far apart people's answers were to start with, so a d of 0.08 is a very small share of that spread.

The next question is which part of that conversation does the work. Three field studies and a survey experiment took the narrative approach apart. Each switched one of three prompts on or off: asking the voter for their own analogies, inviting them to give the other side's perspective, and letting them hear a story about a member of the group. Dropping the first two did not weaken the result. Conversations built only around hearing that story durably reduced exclusionary attitudes, whichever way the story arrived, from a member of the group or from someone else recounting it (Kalla and Broockman, 2023). Read side by side, these designs point to one durable ingredient: the voter hearing a concrete story about a particular person. That reading is ours; the studies do not state it jointly.

Warm, non-judgemental listening is the other candidate. A pre-registered field experiment tested it directly. Trained canvassers held roughly ten-minute video conversations about unauthorised immigration with 1,485 US participants. Two things were randomised separately: whether the canvasser told a persuasive story about an undocumented immigrant, and whether they practised high-quality non-judgemental listening. The narrative worked, moving policy attitudes by d = 0.26 immediately, 0.25 on prejudice and about 0.20 at five weeks. The listening manipulation did not enhance persuasion. Adding it produced d = 0.06 and 0.07 at first, decaying to d = -0.02, with the interval around that long-term estimate never exceeding d = 0.11 (Santoro et al., 2025). Listening still raised liking and trust. The research on listening is unsettled, though: a 2026 review finds puzzles running through it (Itzchakov and Bodie, 2026).

The deep-canvassing effect is small in the company it keeps. A meta-analysis of prejudice-reduction field experiments, 69 separate interventions and 24,378 participants in all, puts the pooled effect at d = 0.51, with the studies disagreeing so much that the plausible range of results runs from -0.90 to 1.92, and reports that interventions work better with school and college groups than with adults (Hsieh, Faulkner and Wickes, 2022). Conversations at the door sit far below that, measured on adults, in the field, months later. Whether a shift that size turns into votes is untested: no published deep-canvassing experiment measures a counted vote, and no traceable published figure exists for the cost per mind changed.

Durability has limits. An independent team ran a pre-registered field experiment on jail decarceration and anti-carceral policy, with half the treatment conversations discussing anti-Black racism and half taking a race-absent approach. Both conditions raised support. The effects were still visible a week later but weakened after a counter message and by six months. Deep canvassing is not a panacea, the authors conclude, though it remains promising given how little else survives even a week (Brock-Petroshius and Gilens, 2025). The evidence base has a shape too: every field experiment is American, and the method arrived through progressive organising, promoted before the 2020 US election by groups aiming to move Trump voters toward progressive positions (Demetrious, 2022). No published field test by a conservative organisation, or on an issue where the direction is rightward, has appeared.

Put the pieces together. A single conversation at the door, in which a voter hears a concrete personal story about a member of a group, can leave that voter's attitude measurably shifted months later. The change is small. It is measured against people who received a different conversation, and it outlasted everything the argument-only version achieved, which was nothing detectable. Lasting change is real here, and it is modest.

So what

The durable part of persuasion is the story the voter receives. The argument does not do it, and the listening does not add it. That is a claim about what the voter gets from the conversation, not about how warm the canvasser seems. The effect is real, small, measurable months later, and bounded. It decays against a counter message (Brock-Petroshius and Gilens, 2025), and every field test so far has been American. Hold all of that together and the method earns a place in a campaign, not a pedestal.

For political parties

A party that wants this effect should build the canvass around a story. Have canvassers carry a concrete personal account from or about a member of the group the party wants to move, and cut the argument script and the listening-training module. Expect small shifts from each conversation, which means scale, repetition and re-testing rather than one successful pilot. Ask throughout: does this choice serve the persuasion the party is trying to do, or the conclusion it already preferred?

For government

Treat deep canvassing as a real but narrow method. The field experiments are American, they grew out of progressive organising, and none of them measures a counted vote or a cost per mind changed. Any rule, guidance or funding built on the method should stay modest and provisional for those reasons. Where a public body does test it, the honest description is a small, measurable attitude shift that decays, not a cure for prejudice. Pilot it, publish the result, and expect the next study to trim the claim again. A programme that only reports what flatters it stops being intelligence.

Case studies

Standing the method up is the part nobody publishes, and there is one account of it. In 2020, 230 canvassers working with seven community organisations across seven US locations held conversations with 6,869 voters, in three pre-registered experiments on immigration and transgender rights. Preparation took roughly two months, and six weeks of that went on iterating the conversation with voters outside the study, so the script was tested and revised before the experiments began. The account comes from the same team that showed an argument alone does nothing (Kalla and Broockman, 2020).

References

Brock-Petroshius, K. and Gilens, M. (2025) 'Can deep canvassing promote anti-carceral attitudes? A field experiment', Journal of Social Issues, 81(2)., e70012. Available at: https://doi.org/10.1111/josi.70012 (Accessed: 10 September 2026).

Broockman, D. and Kalla, J. (2016) 'Durably reducing transphobia: a field experiment on door-to-door canvassing', Science, 352(6282), pp. 220-224. Available at: https://doi.org/10.1126/science.aad9713 (Accessed: 10 September 2026).

Demetrious, K. (2022) 'Deep canvassing: persuasion, ethics, democracy and activist public relations', Public Relations Inquiry, 11(3), pp. 361-377. Available at: https://doi.org/10.1177/2046147x211033838 (Accessed: 10 September 2026).

Hsieh, W., Faulkner, N. and Wickes, R. (2022) 'What reduces prejudice in the real world? A meta-analysis of prejudice reduction field experiments', British Journal of Social Psychology, 61(3), pp. 689-710. Available at: https://doi.org/10.1111/bjso.12509 (Accessed: 10 September 2026).

Itzchakov, G. and Bodie, G.D. (2026) 'Puzzles of interpersonal listening: conflicting findings, theories, and future research', Social and Personality Psychology Compass, 20(5)., e70145. Available at: https://doi.org/10.1111/spc3.70145 (Accessed: 10 September 2026).

Kalla, J.L. and Broockman, D.E. (2020) 'Reducing exclusionary attitudes through interpersonal conversation: evidence from three field experiments', American Political Science Review, 114(2), pp. 410-425. Available at: https://doi.org/10.1017/s0003055419000923 (Accessed: 10 September 2026).

Kalla, J.L. and Broockman, D.E. (2023) 'Which narrative strategies durably reduce prejudice? Evidence from field and survey experiments supporting the efficacy of perspective-getting', American Journal of Political Science, 67(1), pp. 185-204. Available at: https://doi.org/10.1111/ajps.12657 (Accessed: 10 September 2026).

Santoro, E., Broockman, D.E., Kalla, J.L. and Porat, R. (2025) 'Listen for a change? A longitudinal field experiment on listening's potential to enhance persuasion', Proceedings of the National Academy of Sciences, 122(8), e2421982122. Available at: https://doi.org/10.1073/pnas.2421982122 (Accessed: 10 September 2026).