The TARES Test: Running Ethical Campaigns at Scale

One-to-one persuasion has a built-in check. The other person can interrupt, ask what you mean, notice your discomfort, and walk away. A campaign has none of that. Whether it is selling a product, a candidate, a vaccine, or a ballot measure, a campaign speaks to thousands or millions of people who cannot answer back, using messages that have been tested against alternatives and delivered to audiences selected for their likely response. The persuader knows a great deal about the audience; the audience knows almost nothing about the persuader's methods. That asymmetry is why campaign ethics cannot be left to good intentions. It needs a procedure.

The most usable one was published in 2001 by Sherry Baker and David Martinson in the Journal of Mass Media Ethics. They called it the TARES test, after its five principles, and wrote it for working professionals in advertising and public relations. This post sets it out, adds one refinement from the legal scholar Cass Sunstein, and then asks the practical question: does ethical persuasion work well enough to compete?

Baker and Martinson begin from the view that persuasion is a legitimate professional activity whose proximate goal, such as the sale or the vote, must serve some morally defensible end. Each principle is posed as a set of questions the persuader asks of themselves.

• Truthfulness of the message. This goes further than literal accuracy. Is the overall impression true? Has anything been omitted that the audience would need in order to decide well? A message assembled entirely from accurate statements can still deceive through selection and emphasis.
• Authenticity of the persuader. Do I believe this myself? Would I make the same case to people I care about? Would I be comfortable if my part in the campaign were public? Concealed sponsorship fails here, which is what makes astroturfing wrong even when its claims happen to be true.
• Respect for the persuadee. Is the audience treated as people capable of making their own decisions, or as a mechanism to be triggered? Would they still agree if they understood exactly what I was doing?
• Equity of the appeal. Is the contest fair? This principle looks at the gap in knowledge and power between the two sides and asks whether the appeal exploits a vulnerability: children, the recently bereaved, the financially desperate, people who cannot reasonably be expected to see through the technique.
• Social responsibility for the common good. If the campaign succeeds beyond expectations, is the wider community better or worse off? That includes side effects on public trust and on the information environment itself.

The principles are jointly necessary. A truthful message aimed at a vulnerable audience in a way that exploits them fails on equity. A respectful, fair appeal for a cause the persuader privately thinks is harmful fails on authenticity.

"Respect" can feel vague. Cass Sunstein's essay "Fifty Shades of Manipulation" offers a workable test: an attempt to influence people is manipulative to the extent that it does not sufficiently engage or appeal to their capacity for reflective, deliberative choice. The phrase "to the extent that" matters. Manipulation comes in degrees, and Sunstein does not claim that every emotional appeal or every nudge is illegitimate. A graphic health warning engages emotion but gives the viewer accurate, relevant information to reflect on. A countdown timer that resets when the page reloads gives the viewer nothing to reflect on; it works only if they do not think. Sunstein also suggests a publicity check that pairs naturally with TARES: if the technique would stop working, or would cause outrage, once disclosed to its targets, that is strong evidence that it bypasses their judgment.

Three features of modern campaigns put particular strain on the test.

Segmentation. When different audiences receive different messages, each message can pass a truthfulness check while the campaign as a whole does not, because it is telling groups incompatible things and relying on the fact that they will never compare notes. An audit has to look at the full set of variants side by side. Microtargeting on psychological traits or on moments of vulnerability is primarily an equity problem.

Optimization. Message testing is not unethical in itself; finding out which honest explanation people understand best is a service to them. The risk is that an optimizer told only to maximize clicks or donations will discover manipulation without anyone intending it. The subject line implying a personal debt, the fake deadline, and the pre-ticked recurring donation box all tend to win tests. Both major American parties' fundraising operations have drawn press criticism and regulatory attention for tactics of this kind, which suggests the cause is the incentive and not the ideology. See dark patterns. The remedy is to decide before testing which variants are off the table.

Distance. The people who design a campaign never see the individual who is misled by it. TARES works partly by forcing that person back into view.

The strongest argument for cutting corners is that everyone else does and that it works. The field evidence is more interesting than that.

Joshua Kalla and David Broockman analysed 49 field experiments on campaign contact in American general elections, covering canvassing, phone calls, mail, and advertising. Their best estimate of the average persuasive effect on candidate choice was zero. They found exceptions: contact made very early, primaries and ballot measures where voters lack a party cue, and cases where a campaign had invested heavily in finding persuadable voters or took an unusually distinctive position. For anyone tempted by manipulative tactics, the finding is sobering. Most mass persuasion, honest or otherwise, does far less than its practitioners and its critics assume, so the ethical cost of manipulation typically buys very little.

What did show durable effects in their other work is close to the opposite of a trick. In deep canvassing, a canvasser holds a conversation of around ten minutes, asks the voter about their own experiences, shares a relevant story, listens without judgment, and invites the voter to reflect. In a 2016 study in Science, Broockman and Kalla found that such conversations reduced prejudice toward transgender people and that the change was still measurable three months later. The study has a notable history: it was run after the same two researchers uncovered fraud in an earlier, widely publicized paper that had claimed a similar result. Their 2020 follow-up across several issues found that the non-judgmental exchange of narratives was the active ingredient, and that presenting arguments alone did not produce the effect. Deep canvassing passes TARES almost by construction. It is transparent, it engages the listener's reflection, and it could be described to the voter in full without losing its force.

A second line of evidence concerns moral reframing. Matthew Feinberg and Robb Willer found that people typically argue from their own moral values and that this is ineffective with those who hold different ones. In their experiments, conservatives became more supportive of same-sex marriage when the case was made in terms of patriotism and loyalty, and liberals became more supportive of military spending when the case was made in terms of fairness and opportunity for the disadvantaged. The effects were modest and were measured in survey experiments, not in live campaigns.

Reframing is a useful test case. Done well, it means finding the reasons within the audience's own values that genuinely support your position, which is a form of respect. Done cynically, it means voicing values you do not hold to win over people you look down on, and it fails on authenticity. The technique is the same. The test tells the two apart.

• T: Lay every variant of the message side by side. Are they consistent with each other? What would a well-informed critic say has been left out?
• A: Is the sponsor identified wherever the message appears? Would the team be comfortable seeing the strategy memo published?
• R: For each technique, write one sentence explaining it to the target. If the sentence would make the technique stop working, remove the technique.
• E: Who is most susceptible to this appeal, and why? Are any targeting criteria proxies for vulnerability? Are there exclusions for minors and for people in crisis?
• S: If every competitor copied this tactic, what would happen to public trust in the category?
• Optimization guardrails: List prohibited variants before testing begins, including false urgency, fake personalization, pre-ticked consent, and imitation of official notices, and have someone outside the growth team review the winning variants.

The same five questions work from the receiving end. When a campaign reaches you, ask what is missing, who is really speaking, whether the appeal would survive your understanding it, why you in particular were selected, and who benefits if it succeeds. For the manipulative counterparts of these methods, browse Frameworks and Defenses.