Most privacy discussion concerns companies people interact with, where there is at least an account, a policy accepted, and a relationship that could be ended.
Data brokers have none of that. They assemble detailed profiles on people who have never heard of them, from sources those people did engage with, and sell the result to whoever will pay.
Where the material comes from
Almost none of it is obtained improperly, which is what makes the industry difficult to constrain. Each source is individually mundane.
| Source | Typically yields |
|---|---|
| Public records | Property, courts, licences, electoral rolls |
| Loyalty schemes | Purchase history tied to an identity |
| Application components | Location, device identifiers, usage |
| Warranty and competition forms | Voluntarily supplied demographics |
| Web tracking | Browsing interests over time |
| Purchased datasets | Everything another broker already holds |
The last row is what makes the industry self reinforcing. Brokers trade with each other, so a record entering anywhere tends to propagate, and there is no single origin to address.
Inference is the product
Raw records have limited value. What is sold is what can be inferred from them, and the inferences go considerably further than the inputs suggest.
Income bracket, likelihood of moving, health conditions, financial stress, religious affiliation, and political leaning are all commonly modelled. None of these are collected directly. They are predicted from purchase patterns, location history, and association, then attached to the profile as though they were facts.
Wrong inferences are sticky
A prediction is not a measurement, and a meaningful proportion of these attributes are simply incorrect.
The problem is that an error, once assigned, is sold onward and appears in other datasets as an established attribute rather than as an estimate. Correcting it at one broker does not reach the copies, and the individual usually cannot see the record at all, so there is no mechanism by which the error would be discovered.
This is the sharpest practical harm. Not that the industry knows too much, but that it confidently asserts things that are not true about people who cannot see or challenge the assertion.
Who is buying
Marketing is the visible market and not the most consequential one.
Employment and tenant screening services draw on these datasets, as do insurers assessing risk and lenders assessing creditworthiness, sometimes directly and sometimes through intermediaries that repackage the data as a score. Government agencies purchase access to records that would otherwise require legal process to obtain, which is a route around a protection rather than a use of one.
Removal, and its limits
Where the law provides a right to deletion, it is real and worth exercising, and it is laborious. Each broker must be approached separately, many require identity verification that involves supplying more data than they held, and there is no central mechanism covering the industry.
Removal is also temporary. Because the sources remain and brokers continue trading with each other, a profile deleted today is commonly reassembled from the same public records and the same application components within months.
Why this is hard to regulate
Rules aimed at the industry run into a structural difficulty. Each individual collection is lawful, each transfer is between businesses rather than involving the individual, and the resulting profile is derived rather than collected, which places it awkwardly within frameworks written around collection.
Jurisdiction compounds it. A broker can hold records on residents of a country without operating there, and enforcement against an entity with no local presence is slow where it is possible at all.
What reduces the supply
Since removal is a treadmill, the more durable measures act on the inputs. Declining loyalty schemes that tie purchases to an identity, restricting location access on applications that have no functional need for it, and minimising what is volunteered on forms all reduce what flows in.
Public records cannot be withdrawn and are the foundation of most profiles, so the objective is reduction of the commercial layer built on top rather than absence from the dataset.
Note: the distinguishing feature of this industry is the absence of a relationship. There is no account to close, no terms to decline, and usually no way to see what is held, which is why it is regulated as a market rather than addressed as a product choice.