Skip to content
← Back to blog
Security·November 16, 2026·9 min read

How do you test that a CCPA opt-out actually stops ad trackers? What the Healthline complaint found after a "triple opt-out"

California tested Healthline's opt-outs and still found 118 ad cookies afterward. The complaint reads like a test plan: opt-out, page titles, vendor contracts.

The strangest sentence in California's complaint against Healthline Media is not about the law. It is about a test. After a reader used all three of Healthline's opt-out mechanisms, the Attorney General's investigators say they still found 118 cookies tied to third-party advertising companies, plus transmissions to dozens of ad companies that included the title of the article being read. The company had built the opt-outs. Nobody had checked that they worked.

That is the engineering lesson, and it applies well beyond ad tech. If your product has an opt-out, a consent flag or a "do not use for decisions" switch, the regulator's first move is to flip it and watch the network. This post turns the complaint into a test plan. It is a builder's read of a public filing, not legal advice.

What did California allege against Healthline?

On 1 July 2025 the California Attorney General filed People v. Healthline Media, LLC in San Francisco Superior Court (case CGC-25-626794) and announced a $1.55 million civil penalty settlement the same day. The complaint alleges four things: Healthline kept sending data to advertising third parties after readers opted out, it shared article titles that suggested a reader had already been diagnosed with a serious illness, its advertising contracts lacked the terms the CCPA requires, and its cookie banner did not disable the advertising cookies it said it would.

The Attorney General's office says it tested Healthline's opt-out mechanisms in the fall of 2023 and found they did not work correctly. The complaint says about 6.5 million Californians visited Healthline each month, and that Healthline told investigators roughly 65,000 Californians had opted out, mostly through the Global Privacy Control (GPC), a browser signal that tells every site the visitor opts out of sale and sharing.

What did the opt-out test actually find?

Paragraphs 14 to 18 of the complaint describe a short, mechanical investigation. Investigators turned on all three mechanisms: the "Do Not Sell or Share" form, the GPC signal, and the cookie banner's advertising checkbox. Then they loaded pages and looked.

  • 118 cookies associated with third-party advertising companies were still placed, including ones used to track a person across sites.
  • Transmissions to dozens of ad companies still carried unique identifier cookies, and some carried the article title.
  • 82 pixels or tags tied to advertising companies were still present on a single page.
  • A cookie-sync pixel, which the vendor's own documentation says matches your cookies to partner cookies and device IDs, still fired.
  • A universal identifier in local storage, not a cookie at all, was still there.

Notice what a cookie-only check would have missed: the local-storage identifier and the pixels that fire without setting a cookie. The investigators checked cookies, network requests, tags and storage. Your test should check the same four.

Healthline's explanation, per paragraph 28, was twofold. One opt-out mechanism was misconfigured, and a privacy compliance vendor "may not have properly identified and blocked all relevant online trackers" after detecting an opt-out. Pixels the investigators saw had been installed by advertising vendors, not by Healthline's engineers. The company then did an extensive manual review and now disables trackers directly in response to an opt-out.

Why did an article title become a legal problem?

Paragraph 21 says some titles strongly suggested a diagnosis, for example "Newly Diagnosed with HIV? Important Things to Know." Healthline sent those titles to ad vendors next to a unique identifier. The Attorney General alleges that violated the purpose limitation principle in Civil Code section 1798.100(c), read with section 7002(b) of the CCPA regulations, which ties permitted uses to what the consumer would reasonably expect.

The complaint's reasoning is worth reading as an engineer. Healthline's privacy policy mentioned targeted advertising briefly. It never mentioned sharing article titles, and a reader cannot see a title leave the page in the background. So the company could not show that readers expected it. The complaint also walks through how the data could travel: a data broker could add "reader viewed an article about Crohn's disease" to a profile and sell that inference onward.

Here is my opinion, and I would defend it on a call: the page title is the most dangerous field in most web products, because nobody classifies it as personal data. A title, a URL path, a search query, a filename, a job-posting slug. Each can say something about a person once a stable identifier travels with it. If your team has a data inventory that lists "email" and "IP address" but not "document title," it has a gap the Healthline complaint puts squarely in view.

What did the contract finding say?

Paragraphs 24 to 27 are the quietest and, for a software vendor, the most useful. Healthline had signed an industry contractual framework and assumed its ad partners had too. Investigators compared the recipients of the data against the framework's public signatory list, and several recipients were not on it. For those, they read the individual contracts. One allowed use for any "internal use" benefiting the recipient. Another said data would be processed "for the purposes contemplated" in the agreement but never listed those purposes.

Section 1798.100(d) requires a written contract that names the limited and specified purposes for which the data may be used. The complaint adds a second consequence. Healthline passed an opt-out signal (the U.S. Privacy String) to recipients, but often did not address it in the contract. Under section 1798.135(g), a business is protected when it communicates an opt-out only if it has no reason to believe the recipient will misuse the data. The Attorney General's position is that silence in the contract, plus contracts that claimed broad rights, gave Healthline that reason.

Healthline's fix, in the complaint's words, was to disable all sales and sharing through trackers for third parties without compliant contracts. That is a deployable rule: a tag may fire only if its vendor has a contract row that passes review. Our post on vendor contract requirements for ADMT covers the same terms from the service-provider side.

A worked example: one regression test, three assertions

Picture a 24-person Series A company with a California-facing web app, a consent platform and a tag manager. It is not a publisher, but it runs analytics, session replay and two ad pixels. Here is the test the Healthline complaint suggests, run in CI against staging on every release.

  1. 1.Opt-out holds. Launch a headless browser with the GPC signal on (the Sec-GPC request header and the navigator.globalPrivacyControl property). Load the ten most-visited page templates and the logged-in dashboard. Record every network request, every cookie, every local-storage and session-storage key. Fail the build if any request goes to a domain on your advertising-vendor list, or any new identifier appears after the first paint.
  2. 2.Payloads are clean. For every request to a third party, assert the URL, referrer and body do not contain the page title, the path segments of sensitive sections, or search terms. Keep an allowlist of fields per vendor, not a blocklist.
  3. 3.Every domain has a contract. Take the set of third-party domains from step 1's network log (run once without GPC so you see everything) and check it against your vendor register. A domain with no row, or a row with no reviewed contract, fails the build.

The first assertion is what California's investigators did by hand in 2023. The third is what Healthline did after the fact. Build it before an investigator does it for you. The CPPA's earlier decisions point the same way: our read of the Honda and Todd Snyder orders covers opt-out buttons that quietly failed, and the GM settlement covers collecting more than the purpose justified.

Does this matter if you are not an ad publisher?

Yes, with a limit. The Healthline case is about targeted advertising, not about automated decisionmaking technology (ADMT). Advertising targeting is not one of the five "significant decision" categories in section 7001(ddd), so it will not trigger the ADMT obligations that start on 1 January 2027. The failure pattern, though, is the same one that will hurt ADMT teams: a control exists on paper, nobody tests it end to end, and a vendor sits between your switch and the data. An ADMT opt-out that your vendor's scoring service ignores fails exactly the way Healthline's GPC handling did. You can check whether your product is in ADMT scope with the ADMT checker.

Here is where our help is the wrong answer. If you run a marketing site with a tag manager and no automated decisions about people, you do not need custom engineering. Your consent platform vendor, a one-day tag audit and a written contract review will do it, and a services firm adds cost without adding anything. Custom work earns its keep when the opt-out has to reach a model, a queue or a vendor API you built, which is the case for ADMT compliance engineering. If that is your situation, tell us what the system decides and we will say whether it is in scope.

Sources

Written by Zain Ali

Frequently asked questions.

The California Attorney General's complaint, filed 1 July 2025, alleges Healthline kept sending readers' data to advertising companies after they opted out. Investigators who used all three opt-out mechanisms still found 118 third-party advertising cookies and transmissions to dozens of ad companies. The complaint says one mechanism was misconfigured and that Healthline had not tested whether it worked.

The complaint alleges that article titles such as "Newly Diagnosed with HIV? Important Things to Know" suggested a reader had a diagnosis, and Healthline sent them to ad vendors with a unique identifier. The Attorney General says that broke the purpose limitation principle in Civil Code section 1798.100(c) and regulation section 7002(b), because readers would not reasonably expect it and the privacy policy never mentioned sharing titles.

The complaint says several ad-vendor contracts did not name the limited and specified purposes that Civil Code section 1798.100(d) requires. One allowed any "internal use" benefiting the recipient, and another referred to "purposes contemplated" without listing them. The complaint also says many contracts did not address the opt-out signal Healthline sent, which weakened its safe harbor under section 1798.135(g).

No. Targeted advertising is not one of the significant-decision categories in section 7001(ddd) of the CCPA regulations, so it does not trigger the ADMT obligations that apply from 1 January 2027. The case matters to ADMT teams as a pattern: a control that exists but was never tested end to end, with a vendor between the switch and the data.

Mirror the Attorney General's method described in the July 2025 complaint: turn the opt-out on, then inspect cookies, network requests, tags and browser storage for advertising vendors. Run it on every release, and also check that no page title or sensitive URL path reaches a third party. Add a check that every third-party domain has a reviewed contract.