How Data Brokers Profile and Target You

In June 2021, a Catholic newsletter called The Pillar published a story that ended a priest’s career. The reporter had purchased commercially available mobile location data, the kind that advertising firms buy and sell every day. They used it to track the phone of a senior Catholic Church official to gay bars and to the home addresses of other men. The priest resigned within days.

He had never consented to any surveillance. His phone had simply been present in certain locations, and a location data firm had recorded those visits as part of its standard data collection. The data was legal to purchase. The purchase required no warrant. Anyone with a few thousand dollars and a business account could have bought the same dataset.

This is what the data broker profiling industry actually does. It goes beyond collecting your name and address. We cover the basics in the data broker economy and how data brokers find you. The profiling machinery runs inference engines that turn your mundane behavior into predictions about your health, finances, politics, and relationships. Those predictions are sold to insurers, lenders, campaign operations, and advertisers. You never see them. They affect consequential decisions about you regardless.

The short answer. Data brokers don’t just record facts about you. They build statistical models that predict things you never disclosed: a pregnancy, a chronic illness, financial stress, sexual orientation, political leanings. They do it from purchase patterns, location history, and your social connections. These “propensity scores” are sold to insurers, lenders, employers, and campaigns, and you can’t see or dispute them. Aggregators like Acxiom and LexisNexis sit at the center of this. Here is how the machinery works, and what you can do about it.

Jump to a section:

From facts to predictions: how the inference engine works

The shift that made data brokerage genuinely dangerous wasn’t the collection of data. It was what happened when statisticians started building predictive models on top of it.

The most documented early example is Target’s pregnancy prediction model, reported by the New York Times in 2012. Target’s data science team noticed that new parents are among the most valuable customers a retailer can acquire. A household with a new baby reorganizes its entire shopping behavior at once. The problem: competitors were also trying to reach those customers. Target wanted to identify pregnant customers before competitors could.

The model they built relied on a set of 25 products whose purchase pattern was statistically predictive of pregnancy. The list included unscented lotions, certain vitamin supplements, cotton balls in larger-than-normal quantities, and specific hand sanitizers. None of those products is pregnancy-related in any obvious way. But the combination, at the right time, was a reliable signal. Target could predict a pregnancy, often in the first trimester, before the customer had told anyone. In one widely reported case, a father complained to a Target store manager that his teenage daughter was receiving pregnancy-related coupons. He called back weeks later to apologize, having confirmed the coupons were accurate.

The mechanics are worth understanding because they generalize to everything data brokers do. You don’t need to disclose a health condition for a data broker to infer it. You need to purchase things, visit places, or search for information in a pattern that correlates with that condition in training data. Inferences about diabetes, cancer treatment, depression, financial stress, and relationship breakdown have all been documented in broker product descriptions or academic studies of the industry.

Credit card transaction data is particularly rich for this kind of inference. A 2016 study in Nature Human Behaviour used just 3 months of credit card transaction metadata: the merchant, amount, and date, with no descriptions of what was purchased. From that alone, researchers could infer a person’s age, sex, and relationship status with 90% accuracy. They could also re-identify individuals with 90% accuracy in an “anonymized” dataset of 1.1 million people. Data brokers have been doing this commercially for years with far larger datasets.

Propensity scores: the invisible ratings attached to your profile

Brokers and marketing analytics firms assign scores to individual consumer profiles. These are called propensity scores or intent scores. Each one represents a calculated probability that you’ll do something. That something might be to buy a house in the next six months, file for bankruptcy, switch insurance providers, get divorced, or develop a chronic illness.

These aren’t marketing curiosities. They are sold to insurers who use them to refine underwriting. They are sold to lenders who use them to decide whether to offer a loan and at what rate. They are sold to employers screening candidates and to landlords reviewing applicants. The Federal Trade Commission’s 2014 report, Data Brokers: A Call for Transparency and Accountability, documented broker product categories. Those categories included “financial stress indicators” and “ailment and affliction” lists. One list even grouped people as “rural and barely making it.”

None of this appears on your credit report, which is regulated. It lives in unregulated data broker products that you cannot see, dispute, or correct.

The FTC report noted that one broker sorted consumers into segments with revealing names. Examples: “Rural Everlasting” (older rural consumers with low incomes), “Ethnic Second-City Strugglers,” and “Rough Neck Families.” These categories had direct pricing implications for the products offered to people in them.

Lookalike modeling: targeted without ever interacting

You can be profiled and targeted by a company you have never heard of, based on behavior you’ve never shared with them. This is what lookalike modeling accomplishes.

The mechanism: a brand has 100,000 existing customers. It wants to find 500,000 more people who behave like its best customers. It takes those 100,000 profiles to a data broker or advertising platform and asks: who else looks like these people? The broker runs the profiles through their database, identifies the statistical characteristics that predict “best customer” behavior, and returns a list of 500,000 people who match those characteristics. The brand then targets them.

From your perspective, you start seeing ads for a product in a category you’ve recently shown interest in, from a brand you’ve never visited. You didn’t give that brand your data. You didn’t agree to any targeting. Your data was matched through a chain of brokers and intermediaries, and the targeting happened without any direct relationship.

Meta’s advertising system operates on this model explicitly — advertisers upload “Custom Audiences” (their existing customer lists), and Meta identifies “Lookalike Audiences” from its database. Other platforms offer the same capability. The underlying data that makes the match possible often comes from broker databases that Meta has licensed or integrated.

Location history as a health record

The priest story is not an isolated example of what location data can reveal.

Mobile advertising firms build what they call “panel” datasets: the 90-day or 12-month movement history of tens of millions of devices, tied to persistent device identifiers. The firms include Foursquare, X-Mode Social, and SafeGraph. Foursquare (now Foursquare City Guide) was once the dominant location data broker. Digital Envoy acquired X-Mode Social after Apple and Google dropped it from their app stores in 2020 for selling military data.

That movement history reads like a medical file. Picture a device that visits an oncology clinic three times in two months, stops at a pharmacy, and appears at a cancer support group. It is very likely associated with someone undergoing cancer treatment. Brokers have packaged and sold these inferences explicitly. SafeGraph came under public scrutiny after Vice/Motherboard reported in 2018 that it was selling data packages called “disease insights,” linking device movement to medical facilities.

The insurance application of this data is documented. Life insurers have used data sources including pharmacy purchase records, consumer buying history, gym membership status, and credit data — none disclosed in policy applications — to adjust underwriting. New York’s Department of Financial Services opened an inquiry into this practice in 2019 after reports that life insurers were considering social media and consumer data in underwriting decisions.

In the US, this is broadly legal because none of these inferences are technically medical records. They are consumer data inferences based on behavior that happens to correlate with health conditions.

Your own data profile being minimal or locked down doesn’t protect you as much as you might expect. The people around you compensate.

In 2013, researchers at the University of Cambridge published a paper showing that Facebook Likes alone could accurately predict sexual orientation, political affiliation, religion, intelligence, and personality traits. The median predictor was around 150 Likes. Cambridge Analytica extended this further. Its approach affected an estimated 87 million Facebook users, based on 270,000 people who took a quiz. By mapping the social connections of the quiz takers, the firm built models of people who had never interacted with the quiz at all.

The mechanics: if five of your close Facebook friends have a particular personality profile, the model predicts your personality profile with roughly 70% accuracy. You didn’t consent to the quiz. Your profile was inferred from your connections’ data.

Commercial data brokers use household modeling for the same purpose. If your partner has a detailed profile — rich with transaction data, location history, and financial behavior — brokers attribute similar characteristics to all adults at that address. You inherit their inferences even if your own digital footprint is minimal. This is documented in multiple academic analyses of broker data products and is referenced in the FTC’s broker investigations.

There is a further documented phenomenon sometimes called guilt by association: some alternative credit scoring products use social network connections as a factor, explicitly or through proxy variables. Being connected to people with poor credit histories or high-risk financial behavior affects some non-traditional credit scores even when your own credit history is unrelated. This is controversial and legally contested in the US under the Equal Credit Opportunity Act. But it happens in products that position themselves as offering credit access to people with thin credit files.

The genuinely dangerous inferences

Several categories of inference carry particular real-world risk because of how they can be used against the person profiled.

Political affiliation is inferred through a combination of voter registration records (public in most US states), magazine subscription history, donation records, geographic location, and even consumer purchases. Car ownership type correlates with political affiliation in predictive models used by political campaigns. Voter file analytics firms including Catalist and i360 maintain profiles on virtually every registered US voter, including modeled political scores, persuadability scores, and issue-level predicted positions.

Religious practice can be inferred from location data (regular attendance at a mosque, church, or synagogue), purchasing patterns (halal or kosher food purchases), and community membership records. In jurisdictions where religious minorities face discrimination or violence, this category is hazardous. In employment contexts in the US, using religion as a hiring factor is illegal — but using a proxy variable that correlates strongly with religion is a harder legal question.

Sexual orientation inference is documented and has caused direct harm, as the priest case shows. Grindr’s sale of user data — including precise location — to data brokers was the subject of a $11.7 million GDPR fine by Norway’s Datatilsynet in 2021, upheld on appeal. Grindr shared user data with advertising partners without obtaining valid consent. That data, in principle, was available to purchase through standard advertising channels.

Mental health and financial stress inferences are sold explicitly. The FTC’s 2014 report documented broker lists marketed for their utility in targeting “financially vulnerable” consumers — a description that maps to people likely to respond to predatory financial products. Payday loan advertisers, high-interest credit card issuers, and debt settlement firms are documented purchasers of these audience segments.

The products built from your profile

The profiling doesn’t stay inside the broker’s database. It becomes commercial products sold to specific buyer categories.

Insurance scores derived from consumer data are used by some auto and life insurers. Unlike your official credit score, these are unregulated in most US states, you cannot request them, and you cannot dispute inaccuracies in them. The FTC has examined this practice but comprehensive federal regulation hasn’t materialized.

Affordability segments are sold to lenders and “alternative financial services” — the industry term for payday lenders and similar products. The targeting logic is explicit: find people under financial stress who are likely to need short-term cash and are less likely to carefully evaluate interest rates. Regulatory filings from payday lending companies show they purchase audience data from commercial brokers as a standard marketing practice.

Political persuasion scores are the most granular political profiles in commercial data. These are sold to campaigns and issue advocacy organizations. They include predicted voting likelihood, issue priorities, persuadability score (how likely the person is to change their mind), and vulnerability to specific message types. The data behind them comes from voter files merged with commercial data broker profiles.

Price discrimination based on inferred profile is documented, though less clearly than the political targeting. Amazon tested different prices for identical products based on factors including device type and user history — documented in news reporting in the early 2020s. Retail pricing algorithms adjust dynamically based on signals that correlate with consumer willingness to pay. A user profile that includes income bracket, purchase history, and geographic location provides those signals. The extent to which this happens for individual consumers is difficult to verify because pricing tests are opaque, but the practice exists.

What you can actually do

None of this is entirely fixable. The system is large, entrenched, and legal in most jurisdictions. But several specific actions reduce the volume of data flowing into broker systems and the consequences of what they already hold.

Request your consumer file from the major brokers. Acxiom operates a consumer portal at aboutthedata.com where you can see and correct some of the data Acxiom holds. LexisNexis offers a consumer disclosure report. Equifax and TransUnion, as credit reporting agencies, are subject to Fair Credit Reporting Act requirements and must provide your report on request. These are not your full broker profiles — they’re a fraction of what’s held — but they make the system visible.

Opt out of data broker sales. Most data brokers accept opt-out requests, and many are required to under California’s CCPA and the DELETE Act. California’s DROP portal opened to consumers on January 1, 2026. Registered brokers must start processing requests from August 1, 2026; see how to file a DELETE Act request. Submitting opt-outs to the top 20 brokers reduces the most consequential data flows. Services automate this across 200+ brokers. Contacting 4,000 brokers one at a time is not realistic for most people.

Reduce data at source. Loyalty cards trade convenience for permanent transaction records. App location permissions feed location data brokers. Third-party cookies tracked your browsing history until browser changes began limiting them. Each data point you don’t generate isn’t available to be profiled. This is partial — you cannot opt out of the public records that form brokers’ foundation — but behavioral data is where the inference engine gets its signal.

Place a credit freeze. This doesn’t stop broker profiling generally. But it limits what lenders can see when making credit decisions, and it reduces the utility of some financial data products built from your profile.

File complaints when inferences cause material harm. Several regulators accept complaints about inferred-data uses that cause financial, employment, or housing harm. They include the FTC (US), the CFPB (US financial products), and national Data Protection Authorities in the EU and UK. Regulatory enforcement has been limited but it exists. The Grindr fine and a series of GDPR enforcement actions against adtech companies (including a €150 million fine against Meta in Ireland) came from complaints.

The inference engine already exists, and your profile is already inside it. The goal of these steps isn’t to disappear from it entirely, which isn’t achievable for most people. It’s to limit the most specific and sensitive inferences and reduce the data flowing in. The aim is to create friction between your profile and the buyers who want to use it against you.