“Advertisers target smartphone users from search history, location, purchases, browsing and social-graph data that already predict interests with striking accuracy.”

• The U.S. Federal Trade Commission's September 2024 staff report on nine major social media and video streaming companies found that the firms collected and indefinitely retained vast amounts of user and non-user data — including demographic, behavioral, location and inferred-interest data — and fed it into automated systems used for ad targeting, confirming the core mechanism the claim describes.
• The FTC's 2014 data broker study documented brokers compiling billions of data elements per consumer, including purchase transactions, web browsing, and location, and deriving thousands of behavioral/interest 'segments' sold for marketing — i.e., the exact inputs named in the claim.
• Peer-reviewed research (Kosinski, Stillwell & Graepel, PNAS 2013) showed that ordinary digital footprints (Facebook Likes — a social-graph signal) predicted sensitive attributes with high accuracy: 95% for race, 93% for gender, 88% for sexual orientation among men. A follow-up (Youyou et al., PNAS 2015) found model-based personality judgments from Likes outperformed judgments by friends and family, supporting 'striking accuracy' for interest/trait inference.
• The New York Times' 2019 'One Nation, Tracked' investigation analyzed a leaked dataset of 50 billion location pings from 12 million U.S. phones, collected by location-data firms for advertising/analytics, demonstrating that smartphone location targeting at scale is real and granular enough to identify individuals.
• The FTC sued data broker Kochava (2022) over the sale of precise geolocation data derived from mobile devices that could track people to reproductive health clinics, places of worship and shelters — regulatory confirmation that smartphone location data is traded in the ad ecosystem.
• The UK Competition and Markets Authority's 2020 digital advertising market study documented in detail how search history, browsing, purchase and profile data underpin personalised advertising, and estimated personalised ads command substantially higher prices than non-targeted ads.
• The Norwegian Consumer Council's 'Out of Control' (2020) technical report showed popular smartphone apps transmitting user identifiers, location, IP and behavioural data to dozens of adtech companies in real time.

• 'Striking accuracy' is the weakest part of the claim: field experiments by Neumann, Tucker & Whitfield (Marketing Science, 2019) found third-party consumer profiling was often poor — gender targeting was correct only about 42–72% of the time depending on vendor, and age+gender combinations fared worse, sometimes no better than chance.
• The FTC's own 2014 report and subsequent academic audits noted that broker-derived segments contain substantial errors and stale inferences; Pew (2019) found roughly half of U.S. Facebook users said the platform's listed interest categories did not accurately describe them (about 27% said the listings were not very/at all accurate, and many were surprised by them).
• Accuracy is uneven across data types: location and purchase data tend to be reliable, while inferred 'interests' and demographics from probabilistic cross-device graphs are frequently wrong, so a blanket 'striking accuracy' overstates average performance.
• The claim can imply a single, seamless profile; in practice data is fragmented across walled gardens (Google, Meta, Amazon, Apple) and brokers, and Apple's App Tracking Transparency (2021) plus GDPR/CCPA-style rules have measurably reduced cross-app identifier availability, degrading some targeting pipelines.
• Popular adjacent beliefs — e.g., that phone microphones secretly listen to conversations to serve ads — are not supported by the evidence; researchers (Northeastern's 'Panoptispy' study) found no covert audio exfiltration in a large app sample. The claim as written does not assert this, but it is often read that way.

88/100 — the lab’s own confidence in the score above, not a second reading of the claim.

U.S. Federal Trade Commissiongov · 95 · U.S. Federal Trade Commissiongov · 94 · U.S. Federal Trade Commissiongov · 93 · PNAS (Kosinski, Stillwell & Graepel 2013)journal · 92 · PNAS (Youyou, Kosinski & Stillwell 2015)journal · 90 · Marketing Science (Neumann, Tucker & Whitfield 2019)journal · 88 · The New York Timesnews · 85 · Norwegian Consumer Council (Forbrukerrådet)other · 80 · Northeastern University / ReCon (Panoptispy study)edu · 78

• Government and regulatory records show that the ad ecosystem uses the kinds of data named in the claim. The FTC’s data-broker report describes brokers collecting consumer data from purchases, online activity, social media, public records and other sources, then selling marketing segments and inferences to advertisers.
• Mobile-location targeting is well documented. The FTC’s InMarket action says a mobile advertising company collected precise location data through apps, used it to infer visits to sensitive and commercial places, and created audience segments for targeted advertising.
• Major ad platforms acknowledge use of search/activity/location signals for ad personalization. Google’s ad-technology disclosure says ads may be personalized using searches, videos watched, site/app activity, approximate location and information from advertisers.
• Scientific evidence supports the idea that digital traces can predict interests/traits with high accuracy. Kosinski et al. in PNAS found Facebook Likes predicted attributes such as political views, religion, ethnicity, sexual orientation and personality traits; Youyou et al. found computer judgments from Facebook Likes could outperform human acquaintances for personality prediction with enough Likes.
• There is experimental evidence that such inferred traits can improve ad effectiveness. Matz et al. in PNAS found psychologically matched Facebook ads produced higher clicks and conversions than mismatched ads.
• Adtech regulators have documented real-time bidding and profiling systems that transmit or use browsing/device/location/profile data to select ads, supporting the claim that smartphone users are targeted based on multi-source behavioral data.

• The phrase “with striking accuracy” is broadly true in some studies but overstates uniform accuracy. Pew Research found only 59% of Facebook users said the platform’s ad-interest categories reflected their real-life interests very or somewhat accurately, while 27% said they did not accurately reflect them.
• FTC reporting also cautions that data-broker inferences may be inaccurate, stale or unverifiable, and consumers often cannot see or correct the profiles used for marketing.
• Advertisers often do not receive raw search histories, purchase logs or full social graphs directly; they commonly buy platform-defined audiences, lookalike segments or data-broker categories. So the claim is more accurate as a description of the targeting ecosystem than of every individual advertiser’s direct access.
• Scientific studies showing high prediction accuracy usually concern specific datasets, populations and outcomes, such as Facebook Likes predicting traits, not guaranteed precise prediction of every smartphone user’s commercial interests in every context.

86/100 — the lab’s own confidence in the score above, not a second reading of the claim.

Federal Trade Commissiongov · 95 · Googleother · 78 · Proceedings of the National Academy of Sciencesjournal · 95 · Proceedings of the National Academy of Sciencesjournal · 94 · Proceedings of the National Academy of Sciencesjournal · 94 · UK Information Commissioner’s Officegov · 90 · Pew Research Centerother · 90

• The digital advertising ecosystem, dominated by companies like Meta (Facebook/Instagram) and Alphabet (Google), is fundamentally built on aggregating user data—including search queries, web browsing behavior, and purchase histories—to build highly specific predictive profiles for targeted advertising.
• Data brokers continuously harvest and sell Mobile Ad IDs (MAIDs) linked to precise geolocation data from smartphone apps, allowing advertisers to target users based on their physical movements and real-world behaviors.
• Academic research has repeatedly demonstrated that social graph data (such as social media 'likes' and friend networks) can predict highly intimate personal traits, interests, and demographic information with striking statistical accuracy.
• Advertisers use offline purchase data (often acquired through data partnerships with credit card networks and loyalty programs) and match it with online profiles using 'onboarding' services, enabling them to predict future buying intent accurately.

• While predictive algorithms are statistically powerful across large populations, they frequently make erroneous inferences about individual users, such as serving persistent ads for products a user has already purchased or misinterpreting a brief search as a long-term interest.
• Recent privacy initiatives, such as Apple's App Tracking Transparency (ATT) framework in iOS and stricter global privacy regulations (like the GDPR), have somewhat degraded the accuracy and availability of third-party browsing and location data.

95/100 — the lab’s own confidence in the score above, not a second reading of the claim.

Electronic Frontier Foundationother · 85 · Proceedings of the National Academy of Sciences (PNAS)journal · 95 · The New York Timesnews · 90

• Advertisers and data brokers demonstrably use smartphone location data linked to mobile-advertising IDs; the FTC said X-Mode/Outlogic sold precise location data to hundreds of clients for advertising and brand analytics.[1]
• Academic research documents that Facebook algorithmically infers users’ interests from their activity and offers advertisers targeting based on those inferred interests.[2]
• A controlled experiment found that knowing 18 or more randomly selected interests could enable highly specific ad targeting: 8 of 9 campaigns successfully reached the selected person.[3]
• These findings support the narrower proposition that behavioral, location, and social-graph signals are used for advertising inference and can sometimes be highly predictive.

• The wording overstates accuracy by saying the combined data already predict interests with “striking accuracy” without specifying a metric, population, platform, or prediction task.
• North Carolina State researchers found Facebook’s inferred-interest categories were inaccurate or irrelevant about 29.3% of the time among 146 participants, while a controlled account experiment found 33.22% inaccurate or irrelevant.[4]
• A peer-reviewed study reports an internal Facebook estimate that U.S. interest-category precision was only 41%, implying that more than half of targeted users could fall outside an advertiser’s intended audience.[2]
• The evidence does not establish that advertisers routinely access all five listed data types—search history, location, purchases, browsing, and social-graph data—for every smartphone user, nor that their joint predictions are consistently highly accurate.
• The strongest targeting result concerns a specially constructed experiment in which researchers supplied many known interests; it does not demonstrate ordinary advertising performance from passively collected smartphone data.

91/100 — the lab’s own confidence in the score above, not a second reading of the claim.

Federal Trade Commissiongov · 98 · Social Media + Societyjournal · 92 · arXivjournal · 75 · North Carolina State Universityedu · 90

• Ad platforms including Google, Meta and Amazon routinely use first-party and partner data covering search queries, device location, purchase records, browsing activity and social connections to build interest profiles and deliver targeted ads, as described in their own ad-product and privacy documentation and confirmed by regulators.
• Peer-reviewed work (Kosinski, Stillwell & Graepel, PNAS 2013) showed that digital records such as Facebook likes predict personality, political views and other attributes at levels that often exceed human judges, supporting the 'striking accuracy' phrasing for social-graph and behavioral data.
• Investigative reporting has documented commercial location-data brokers obtaining precise, persistent smartphone location histories that are sold for advertising and analytics.
• FTC examinations of data brokers and mobile advertising have described the aggregation of search, location, purchase and browsing signals into marketing profiles used to target consumers.

• Apple App Tracking Transparency (2021 onward) and the ongoing deprecation of third-party cookies have reduced cross-app and cross-site identifiers for large numbers of users, so some third-party interest predictions are less precise than they were a few years ago.
• Meta sharply restricted third-party access to social-graph data after 2018, meaning advertisers no longer obtain the same unrestricted friend-network signals that earlier studies used.
• Individual-level predictions remain probabilistic; accuracy is high in aggregate and for many traits but is not uniformly 'striking' for every person or every interest.
• Opt-outs, ad blockers, private browsing, VPNs and platform privacy controls degrade the completeness of the data available to many advertisers.

78/100 — the lab’s own confidence in the score above, not a second reading of the claim.

Proceedings of the National Academy of Sciencesjournal · 95 · The New York Timesnews · 88 · Federal Trade Commissiongov · 92 · The Guardiannews · 82 · Pew Research Centerother · 87