Targeted advertising

• 1. Introduction

  ›

  Targeted advertising follows you. It watches what you search, where you click, how long you linger on a page, and then it follows you somewhere else to show you an ad for the thing you were just thinking about. The practice has a name: data-driven marketing. And it has quietly become the financial backbone of most of the internet.

  The question this documentary will explore is not simply how it works. It is what it costs. Not in dollars, but in privacy, autonomy, and trust. And it will ask whether the system that delivers ads with stunning precision actually knows you as well as it thinks it does.

• 2. How the Tracking Works

  ›

  When a user visits a website that runs on Google's display network, that site sends a cookie back to Google. That cookie carries information about the user: what they searched, where they are based on their IP address, and what pages they visited. Google then builds a profile, assigning the user to a category. Visit promotional-products sites often enough, and you get filed under promotional products.

  Cookies are the most common tool, but they are far from the only one. Web beacons, third-party ad-serving software, and traffic-logging systems all contribute. Data from a single visit to one website can be transmitted to more than 100 other websites, shared with business partners, advertisers, and other third parties. The collected data typically excludes names, email addresses, and phone numbers. But it does include IP addresses, MAC addresses, browser fingerprints, and device-specific alphanumeric IDs.

  Social media adds another layer. Facebook collects likes, view history, and geographic location. It lets advertisers narrow a campaign by gender, age, location, behavior, and interests. The platform only needs to look in one place, the user's profile, to find nearly everything an advertiser wants to know.

  Cable television has joined the same system. Cable box addresses can be cross-referenced against data brokers such as Acxiom, Equifax, and Experian, pulling in information about marriage, education, criminal records, and credit history. Political campaigns can also match viewer addresses against public records showing party affiliation and primary voting history.

• 3. The Many Forms of Targeting

  ›

  Demographic targeting was the first and most basic form used online. It segments an audience by gender, age, ethnicity, income, and parental status. All members of a segment share one trait, and the advertiser places a campaign only in front of that group.

  Psychographic targeting goes deeper. A study conducted jointly by the Entertainment Technology Center at the University of Southern California, the Hallmark Channel, and E-Poll Market Research found that a user's lifestyle is a better predictor of media use than demographic profile alone. Cohorts with similar demographics can have entirely different attitudes and habits. Psychographic segmentation uses personality, values, interests, and lifestyle to draw finer distinctions.

  Contextual targeting is the most straightforward method. An advertiser places an ad next to content that relates to the product. An article about buying a home runs alongside an insurance ad. An automated matching system reads the page's keywords and serves an appropriate ad, sometimes through pop-ups. The system can fail when it cannot distinguish between a positive and negative use of a keyword, placing contradictory or inappropriate ads beside sensitive content.

  Behavioral targeting tracks individual users rather than content. If a consumer frequently searches for plane ticket prices, the system recognizes this and begins showing airfare deals across unrelated sites, including on platforms like Facebook. Researchers also refer to this process as audience targeting. A 2006 study by BlueLithium, later acquired by Yahoo, tested behavioral targeting across 400 million ad impressions. Using nine behavioral categories such as shoppers or travelers, researchers found that behavioral targeting in context worked best for generating clicks, while behavioral targeting out of context produced the highest conversion rates.

  Location-based targeting uses IP addresses and mobile app data to place users geographically, down to ZIP code. Applications like Uber that request location access allow advertisers to reach users with ads for nearby restaurants or shops. The same location data has drawn the attention of intelligence agencies worldwide, who can use it to track personnel at sensitive sites such as military bases by purchasing feeds from commercial providers.

• 4. The Economics of Precision

  ›

  A 2009 study by the Network Advertising Initiative put a number on what targeting is worth. Targeted ads secured an average of 2.7 times as much revenue per ad as non-targeted run-of-network advertising. They also converted users who clicked into buyers at twice the rate of untargeted campaigns.

  Researchers Farahat and Bailey used a large-scale natural experiment on Yahoo to measure the true economic impact more precisely. Assuming a cost per 1,000 ad impressions of one dollar, they found that the marginal cost of a brand-related search from a general ad campaign was $15.65 per search, compared with $1.69 per search from a targeted campaign. For clicks, the gap was even wider: 72 cents per click without targeting versus 16 cents with it.

  Research from 2015 found that content marketing generated three times as many leads as traditional outbound marketing while costing 62% less. Separately, 86% of people skip television commercials, and 44% ignore direct mail, illustrating how much traditional advertising reaches the wrong audience.

  Academics Chen and Stallaert studied the economics of behavioral targeting from a publisher's perspective in a paper titled An Economic Analysis of Online Advertising Using Behavioral Targeting. Their 2014 analysis found that a publisher's revenue can double under behavioral targeting in some circumstances. But the gain is not guaranteed. When competition among advertisers is intense and valuations are low, targeting can actually reduce ad prices and hurt publisher revenue. They identified two competing effects: a competitive effect and a propensity effect. Which one dominates determines whether the publisher wins or loses.

• 5. The Privacy Reckoning

  ›

  A survey conducted by the Pew Internet and American Life Project between January 20 and the 19th of February 2012 found that 68% of Americans said they were not okay with targeted advertising because they did not want their online behavior tracked and analyzed.

  By early 2018, the average global ad-blocking rate had reached 27%. Greece led all countries, with more than 40% of internet users reporting they used ad-blocking software. Among technically oriented users the rate climbed to 58%.

  Privacy International, a UK-based registered charity, argued that any interception of web traffic should require explicit and informed consent from users. Canadian academics at the University of Ottawa Canadian Internet Policy and Public Interest Clinic formally demanded that the federal privacy commissioner investigate online profiling for targeted advertising. The European Commission, through Commissioner Meglena Kuneva, raised concerns about online data collection, user profiling, and behavioral targeting and signaled an intent to enforce existing regulation.

  In December 2010, the online tracking firm Quantcast agreed to pay $2.4 million to settle a class-action lawsuit over its use of so-called zombie cookies. These cookies, placed on partner sites including MTV, Hulu, and ESPN, would regenerate after deletion to continue tracking users. Facebook's earlier use of its Beacon tool to track users across the internet drew similar scrutiny. In October 2009, a survey by the University of Pennsylvania and the Berkeley Center for Law and Technology found that a large majority of US internet users rejected behavioral advertising outright.

  Critics also pointed to a structural problem: researchers had shown by 2009 that supposedly anonymized data could be used to re-identify real individuals. In March 2011, the online ad industry announced it would begin working with the Council of Better Business Bureaus to police itself and monitor how marketers track consumers.

• 6. When Targeting Fails

  ›

  Behavioral targeting sounds precise. The reality is more complicated. Research cited in the source found that when targeting by gender, the system's guess is accurate only 42% of the time. That figure is worse than a random guess. When targeting by both gender and age, the accuracy drops to 24%.

  The contextual ad-matching system carries its own failure mode. It cannot reliably distinguish between a positive and a negative mention of a keyword. A news article about a product recall could end up hosting an ad for the recalled product.

  Time targeting adds yet another variable. According to the Journal of Marketing, more than 1.8 billion users spent a minimum of 118 minutes daily on social media in 2016, and nearly 77% of those users interacted with content through likes, comments, and link clicks. Advertisers responded by trying to schedule ads for periods when working memory is most receptive. Research in chronopsychology found that working memory tends to be strongest in the morning, lowest in mid-afternoon, and moderate in the evening. Placing an ad at the wrong time, even with a perfect demographic match, may mean it is never retained.

  Retargeting, where ads follow a user who has already viewed a product, is considered a more effective technique. Store catalogs use a version of it by enrolling customers in email lists after a purchase. Online, the same items a user browsed can chase them across unrelated websites. But there is an acknowledged threshold: consumers who see the same ad too many times report feeling watched, and some begin to avoid the brand entirely.

• 7. The Microphone Debate

  ›

  Since at least the mid-2010s, a persistent theory has circulated among smartphone users: technology companies are using device microphones to listen to private conversations and serve targeted ads based on what was said. The theory is usually accompanied by personal accounts of an uncanny ad appearing after a conversation about an unrelated topic.

  Facebook denied the practice. Mark Zuckerberg denied it in congressional testimony. Google also denied using ambient sound or conversations to target advertising. Technology experts who investigated the claims described them as unproven and unlikely, pointing instead to the many non-audio methods companies already use to track interests and behavior.

  In December 2023, the publication 404 Media reported that Cox Media Group had been marketing a service called Active Listening to advertising professionals. The pitch claimed the service could access microphones in smartphones, smart televisions, and other devices to target ads. A promotional deck stated that the service targeted Google and Bing and identified Cox Media Group as a Google Premier Partner. Meta, Amazon, Google, and Microsoft all denied using the service. Google, after reviewing the situation, removed Cox Media Group from its Partners Program. Cox Media Group removed the material from its website and denied listening to any conversations.

  The episode did not resolve the underlying debate, but it confirmed that the line between data collection and surveillance is one that advertisers, regulators, and users are still negotiating.