There is a point, somewhere between the seventh and the thirteenth impression, at which an advertisement stops selling a product and begins selling its absence. The consumer has seen the shoe. They have seen it seventeen times. They have Googled the shoe. They have put the shoe in their cart. They have removed the shoe from their cart. And now the shoe follows them across every website they visit, in rotations of creative that cycle with the desperation of someone who will not take a hint.
The shoe does not sell. The shoe retreats. And the retreat is not irrational — it is a predictable consequence of a system operating past its design threshold. Retargeting, at sufficient frequency and sufficient precision, produces the opposite of its intended effect. Marketing engineers call this the "frequency cap." Behavioral scientists call it reactance. The consumers who have stopped buying the shoe call it being followed.
What nobody had a name for was the second-order effect: the following itself becomes the reason for the abstention. The ghost purchase follows you after you decided against it, and the following is what makes you decide against it harder. We propose a name for this. The Precision Discount: the measurable depreciation of purchase intent that accompanies surveillance-level ad targeting precision.
"Consumers do not like to be watched. This is perceived as an invasion of privacy." — Wayne D. Hoyer, University of Texas at Austin, Psychology & Marketing, 2025
The canonical case is not a scandal. It is a grocery store. In 2012, a Target data scientist named Andrew Pole built a pregnancy prediction score using twenty-five products — unscented lotion, supplements containing calcium, magnesium, and zinc, cotton balls, washcloths — weighted and combined into a model that could identify pregnant customers with sufficient precision to send them coupons for cribs and diapers before their second trimester. The model worked. It worked so well that a Minneapolis father walked into his local Target, demanded to speak to a manager, and asked why his teenage daughter was receiving coupons for baby clothes. He did not yet know what Target knew.
Target's response was not to abandon the model. It was to obscure it. The company began interspersing baby product coupons with coupons for lawn mowers and hunting equipment, making the targeting less legible, less precise — less creepy. The pregnancy model remained. The visible evidence of its operation was dressed in camouflage.
Eight years later, in August 2022, Sephora paid $1.2 million to settle California's enforcement action under the CCPA. The company's offense: selling personal information via third-party advertising trackers without proper disclosure, and failing to honor Global Privacy Control browser signals — the explicit machine-readable expression of consumer opt-out preference. Sephora was not a rogue operator. It was a luxury cosmetics retailer with a sophisticated compliance department, operating a standard third-party tracking stack, doing what nearly every major DTC brand was doing. The settlement was a signal. Thirty percent of forty major retailers examined by Consumer Reports in 2025 continued serving retargeted advertisements after receiving valid opt-out signals, in potential violation of state privacy statutes that carry penalties of $100 to $750 per consumer per incident.
The system is aware of its problem. It has been aware of its problem for years. The awareness has not produced reform. It has produced better camouflage.
In 2025, researchers at the University of Texas at Austin and the University of Bern published a four-study, 2,500-participant examination of what they called the "creepiness phenomenon" in digital marketing. Their finding: personalized advertisements nearly doubled participants' reported feeling of being under surveillance compared to non-personalized control conditions. Ambiguity about data sources and the perception of intrusive surveillance together explained 75 percent of the emotional discomfort reported. Each one-point increase in consumer reactance on a seven-point scale reduced stated willingness to purchase by approximately half a point.
The researchers tested interventions. They presented participants with transparency disclosures. They offered monetary incentives. They associated the brand with charitable donations. They showed participants images of kittens. None of these interventions restored purchase intent to baseline levels once creepiness had been activated. The effect was, in their words, "robust and difficult to mitigate once triggered." Prevention, they concluded, was the only reliable strategy.
The underlying psychology is not new. Jack W. Brehm articulated Psychological Reactance Theory in 1966: when an individual's behavioral freedom is threatened or eliminated, they experience a motivational state aimed at restoring that freedom. The classic prediction is a boomerang effect — people do the opposite of what they are being led to do. Applied to retargeting, the mechanism is direct. The consumer who is followed across the web by an advertisement for a product they considered and rejected is experiencing a threat to their freedom to make that decision privately, without external pressure. The response is to decline harder.
Research by Youn and Kim at Computers in Human Behavior (2019) mapped this chain empirically across Facebook newsfeed advertising. Perceived autonomy reduction increased perceptions of ad intrusiveness. Intrusiveness perception triggered freedom threat. Freedom threat produced reactance. Reactance manifested as both negative cognition — actively discounting the brand — and behavioral avoidance — installing ad blockers, abandoning sessions, deliberately navigating away from served impressions.
The mathematics of this threshold have been documented at scale. Atlas DMT, analyzing 38 direct-response advertisers, found that the first online ad impression converts at 3.5 times the rate of the overall campaign average. Impressions one through three each deliver better than 100 percent higher conversion than the campaign mean. The fourth and fifth impressions still outperform the average. The sixth impression does not. After five total impressions, cost per acquisition is no longer profitable. One-third of all online ad impressions are served to users who have already seen the same creative ten or more times — pure waste, purchased and served and nobody converting.
The data from 5.8 billion ad impressions analyzed by researchers Evert de Haan and Steffen Försch, published in the International Journal of Research in Marketing (2018), confirmed the shape of the curve: higher frequency produces lower click-through rates, and shorter intervals between exposures produce lower CTRs still. The effects are stronger for campaigns with less creative diversity, for higher-spending brands, and for durable goods where the purchase decision involves deliberation. Sahni, Narayanan, and Kalyanam's field experiment with a major home-improvement retailer, published in the Journal of Marketing Research (2019), found that switching retargeting on produced 14.6 percent more website returns within four weeks — but 33 percent of that total effect occurred on the first day. The effect decays fast. The follow-up impressions are not producing proportionate returns. They are producing reactance.
"An inverted-U relationship between ad frequency and effectiveness exists: effectiveness first increases, then reaches a peak, and subsequently declines." — Todri, Ghose & Singh, Information Systems Research, 2020
The tipping point for personalization specifically has been isolated. Research published by Kim and Han in Behavioral Sciences (2025) confirmed that under conditions of high privacy concern, highly intrusive personalization — the kind that demonstrates surveillance-level knowledge of the consumer's behavior — performed significantly worse than moderate contextual personalization. The high-intrusion condition did not merely fail to outperform the moderate condition. It underperformed it, producing purchase intent scores no better than generic advertising and significantly worse than the contextual baseline.
The industry has quantified its own problem with unusual precision. Meta's conversion-optimized campaigns saw a 36.6 percent reduction in click-through rates following Apple's App Tracking Transparency framework rollout in 2021 — not because the ads got worse, but because the signal fidelity dropped and the algorithm was forced to guess. Firms with high iOS user dependence experienced an average 40.1 percent revenue decline attributable to the measurement disruption. This was not a privacy regulation. This was the consequences of a privacy choice presented to users at the operating system level, and the consequences were entirely predictable and were predicted.
The consequences were also, in a narrow performance-marketing sense, worse than the problem they were responding to. Apple ATT's prompt — the one that asked users if they wanted to allow apps to track them across other apps and websites — was opted out of by 80 to 85 percent of iOS users. The users who opted out were, disproportionately, the highest-value consumers: the ones with the most digital footprint, the most purchase intent, the most complex behavioral histories that made them valuable targets. Behavioral targeting had been most effective on the people most likely to refuse it.
GDPR enforcement has produced €5.88 billion in total fines since 2018, with €1.2 billion in the most recent twelve-month period. LinkedIn was fined €310 million for using the wrong legal basis for behavioral advertising. TikTok was fined €530 million for illegal data transfers to China. Google received €325 million for manipulating cookie consent interfaces. The enforcement actions have not ended behavioral targeting. They have required it to be better concealed.
The post-cookie world was supposed to be a crisis. For contextual targeting, it became an opportunity. The market for contextual advertising — placing ads based on page content rather than user history — is projected to grow from $205 million to $673.5 million by 2033. Not because advertisers abandoned the desire to reach the right person at the right moment, but because the alternative had become legally and technically untenable. Contextual targeting is older than behavioral targeting. It requires no tracking. It leaves no audit trail of the consumer's digital body. It is, in the industry's own framing, "creepy-proof."
The liminal space is the gap between the system that is breaking and the system that will replace it. Behavioral retargeting still works, in the narrow sense that it still produces measurable conversions. It works less well than it did, it costs more per conversion as third-party data sources atrophy, and it produces a measurable suppression effect in a identifiable subset of consumers — the ones for whom the precision of the targeting has crossed the threshold from relevant to intrusive. These consumers do not simply ignore the ad. They actively revise their assessment of the product and the brand upward in negative direction. The Precision Discount applies to their consideration of the category, not just the brand. They are less likely to buy the shoe. They are also less likely to buy any shoe.
The system is aware of this. The system has frequency caps it routinely ignores. The system has compliance departments and legal exposures and quarterly performance targets that do not reward the prevention of a negative outcome that was never measured. What gets measured is the conversion that occurred. What does not get measured is the conversion that would have occurred if the consumer had not been followed. The denominator is invisible.
Somewhere between impression one and impression seventeen, the retargeting ghost completes its work. The shoe was not bought. The consumer was not convinced. The attention was not converted. It was consumed. And the ghost moved on to the next shoe, the next cart, the next deliberation interrupted — a distribution layer for the unresolved.