When a Health App Says You Are Safe, You May Take More Risks

In 1975 the economist Sam Peltzman showed that mandatory seat belts and crumple zones did not cut traffic deaths as much as expected. Drivers felt safer, so they drove faster. The pattern, now called the Peltzman Effect, has since turned up with bicycle helmets, sunscreen, and other safety measures. Sidney T. Anderson, associate professor of marketing, argues that the same dynamic is now built into how firms market AI health products.

Wearables, symptom checkers, coaching apps, and medication reminders are sold on precision, constant oversight, and reduced risk. That message is the problem. The paper proposes that the safety signals marketers attach to these products can encourage the riskier behavior they were meant to prevent.

The idea

The framework rests on a gap between three kinds of safety. Objective safety is what the product can actually detect and prevent. Claimed safety is what the brand asserts. Perceived safety is what the customer believes. Marketing widens the gap between claimed and objective safety by turning a narrow algorithmic reading into a broad promise of protection. Customers, who rarely have the expertise to judge the difference, act on the promise.

From there the paper lays out a three-stage pathway. First, AI safety cues lower perceived risk without changing actual risk. A person wearing a cardiac monitor is no less likely to have a cardiac event, only more likely to feel covered. Second, lower perceived risk expands what the customer feels licensed to do. That shows up as pushing past limits, treating an algorithm’s threshold as a floor rather than a target, and offloading self-monitoring onto the device. Third, that extra latitude produces harm: injury, delayed care, or lapsed treatment that offsets the benefit the product was designed to deliver.

The paper stresses that this is not a story about technology failing. The device can work exactly as designed. The risk comes from the interaction between safety-centric positioning and the customer’s recalibrated sense of danger.

Not all AI, and not all behaviors

Four product types carry different levels of compensatory risk. Predictive tools such as symptom checkers collapse a probability into a categorical all-clear. Conversational coaches deliver fluent, authoritative advice that people over-trust. Continuous monitors such as heart rate and glucose wearables emit a constant stream of normal readings, which the paper rates as the highest risk because the reassurance never stops. Reminder apps encourage cognitive offloading and complacency.

The behaviors at stake differ too. Preventive behaviors such as exercise and diet are prone to intensity escalation: a runner chases the mileage target on the dashboard while ignoring strain the device does not measure. Compliance behaviors such as medication adherence and recovery protocols are prone to lapses when the customer trusts the app to catch every miss.

The paper also identifies who is most exposed. People with high trust in algorithms, low AI literacy, and few clinical relationships to correct their assumptions are at the greatest risk. For them, algorithmic reassurance works like moral hazard in insurance. The perceived safety net lowers the incentive to avoid the behaviors that cause harm.

What it means for the firms selling these products

Anderson places responsibility for this dynamic inside marketing strategy, not just public policy. Regulatory risk compensation happens to whole populations through mandates. Marketing-induced risk compensation happens to individuals through voluntary adoption driven by brand promises. The firm that writes the promise owns the consequences.

The practical guidance follows from that. Communication should frame the product as a partner that supports the customer’s judgment rather than an authority that replaces it. Product design should build in prompts that keep the customer engaged, disclose uncertainty, and qualify reassuring readings. Escalation rules should be explicit and customer-facing, so the product’s protective scope is clear before anyone relies on it.

The paper is conceptual, and its nine propositions are meant as a structure for future testing rather than settled findings. The immediate message for product and brand teams is narrower and actionable: treat safety framing as a variable to test, not a fixed positioning decision.

What it means for managers

• Position AI health products as support, not protection. Language that hands risk ownership to the device invites people to relax their own vigilance.
• Design for informative friction. Periodic self-checks, visible uncertainty, and alerts that qualify a normal reading keep the user in the loop.
• Publish escalation thresholds. Tell customers exactly when the product hands off to a clinician, so they do not assume it covers more than it does.

Anderson, S. T. (2026). Marketing safety, producing risk: A Peltzman effect framework for AI in consumer health. Journal of Marketing Management. 10.1080/0267257X.2026.2730465