The 'modeled you': the data they guessed, not gathered

By MercPrivacy · Published 2026-06-24 · Updated 2026-08-05

Income bands, household headcounts, propensity scores: much of your marketing profile was never collected. It was guessed. Why wrong guesses still sell, and how to starve the model.

Somewhere in a marketing database there is a version of you with a specific income band, a household headcount, a set of likely interests, and a score predicting whether you will answer an unknown number. You never told anyone most of it. Nobody had to collect it, because nobody collected it at all.

It was modeled. And the modeled you — accurate or not — is a product, priced and sold like any other.

Most people who worry about data brokers picture the gathered layer: addresses, phone numbers, relatives, records. That layer is real, and [what a data broker knows](https://mercprivacy.com/knowledge/what-a-data-broker-knows) walks through it. But much of the calling, mailing, and targeting runs on the inferred layer, and the inferred layer plays by stranger rules.

## Gathered versus guessed

The gathered layer is records. A deed exists. A number was ported. An address appeared on a filing. Gathered data can be wrong, but it is wrong the way a typo is wrong — there is a source to point at.

The guessed layer is different in kind. The common modeled attributes:

- **Income bands** — estimated from neighborhood, home value, occupation codes, and purchase patterns. Not your W-2. A guess wearing the clothes of a fact. - **Household composition** — inferred from shared addresses, surnames, and mail patterns. This is how a roommate becomes a spouse in somebody's database. - **Interests and propensities** — assembled from subscriptions, purchases, and lookalike logic: people who resemble you did this, so you probably will. - **Response scores** — the quiet ones. Likelihood to answer a call. Likelihood to refinance, donate, switch insurers. These exist so a list buyer can rank a million strangers by expected yield.

None of these fields required your participation. They are computed about you, from inputs you mostly cannot see. And the inputs are humbler than people imagine: a magazine subscription implies an interest and an income; a loyalty card timestamps your shopping; a charitable donation lands you on donor-modeling lists. None of it is dramatic, which is the point — the model is built from the exhaust of ordinary life, purchased in bulk.

## Why wrong guesses still sell

Here is the part that offends people's sense of fair play: the model does not have to be right about you to be worth money.

A list buyer is not purchasing truth. They are purchasing lift — a hit rate better than dialing at random. A model that misjudges you personally still narrows a million strangers into a denser, cheaper list, and your individual wrongness is rounding error in someone else's spreadsheet. Nobody downstream audits the guesses; the list is bought, dialed, and refreshed regardless.

> Nobody in this chain is paid to be right about you. They are paid to be slightly less wrong than random.

That explains the calls that make no sense — the solar pitch to the renter, the retirement rollover pitch to the twenty-five-year-old. The caller was not confused about you specifically. A guess sold, and a dialer did what dialers do. How those lists turn into ringing phones is its own story, told in [why the calls started](https://mercprivacy.com/knowledge/why-the-calls-started).

Your mail, incidentally, is the one visible surface of this hidden layer. The retirement seminar invitation, the medical alert postcard, the new-mover coupons arriving months after you moved — each is a model's guess about you, printed and stamped. People read junk mail as random. It is not random. It is a report on what the modeled you looks like this quarter, delivered to your door by the same industry that will not show it to you any other way.

## You cannot correct a model

A people-search listing at least gives you a record to point at: this address is wrong, suppress this entry. The modeled you offers no such handle.

- **You cannot see it.** Propensity scores and inferred segments do not appear on any page you can look up. There is no listing to dispute, because there is no listing. - **Correction rights reach stored data.** Several states let you correct inaccurate personal data a company holds, and that is worth having. But a model's output is regenerated from its inputs — correct the guess today and the same inputs rebuild it tomorrow. - **The market is wholesale.** These attributes trade between vendors and platforms with no consumer-facing counter, no complaints desk, no appeal.

There is one more reason correction is the wrong frame: the guesses are not stable enough to argue with. Scores refresh as new inputs arrive; segments get rebuilt when vendors retrain. The wrong income band you somehow corrected in the spring would be recomputed by fall. You would be disputing a photograph of a river.

Arguing with the model is not an available move. Changing what it eats is.

## Starving it

The inputs are the lever. Each of these reduces what the modeled you is built from:

1. **Suppress the identity spine.** The people-search and broker layer is what links your scattered records into one target. Fewer live listings means weaker linkage, and guesses that attach to a fuzzier, less sellable you. That suppression work is what [how it works](https://mercprivacy.com/how-it-works) describes. 2. **Cut the standing feeds.** Prescreened credit offers stop at [optoutprescreen.com](https://www.optoutprescreen.com); marketing-mail preferences live at [dmachoice.org](https://www.dmachoice.org). Both quietly remove dependable signals from the pipeline. 3. **Stop volunteering fresh inputs.** Warranty cards, quote forms, giveaway entries, surveys — each is a labeled, dated data point sold into exactly this market. The freshest signals in the system are the ones people hand over themselves. Where a form is unavoidable, fill only the required fields; an optional phone-number box is optional for you and valuable for them, which tells you everything about whether to fill it.

None of this deletes the modeled you. It degrades it — staler inputs, weaker linkage, lower confidence — until you slide down the ranked lists that decide whose phone rings this quarter.

## Where we fit

The modeled you shrinks as the gathered you does; that is the practical good news. We run the suppression and the upkeep that make it happen. Ask [Stephanie](https://mercprivacy.com/stephanie) anything — she answers instantly and free — or call (830) 587-5011.