How credit cards are classified
Google's publisher policies place credit-related products and services, banking products and services, and some financial planning and management services in a single restricted group. The examples given are credit cards and loans, bank and checking accounts, and debt management products. That grouping matters more than most advertisers expect: a balance-transfer offer, a secured card, and a debt-consolidation program inherit the same targeting constraints, because the rules are written for the category rather than for a single product page. When a campaign is rejected, the cause is usually the audience definition, not the headline. It is also common for one team to write the offer copy while another builds the audience, which is how a compliant ad ends up attached to a non-compliant segment.
The financial-hardship signals you cannot buy
Under Google's personalized advertising rules, inferred or actual sensitive information cannot be used to select or target ads. The sensitive list explicitly covers negative financial status information and other details about a person's financial situation, including data indicating a low credit rating or a high debt burden. In practice this rules out much of what a card issuer's CRM hands over: delinquency flags, utilization-based hardship models, credit-building segments built from scored files, and lookalikes seeded from those lists. It does not matter whether the segment arrives from a first-party file or an outside vendor — the restriction attaches to the targeting, not to the source. Readers often assume that user consent clears the way; the policy does condition audience-data use on holding all necessary rights, but it separately bars the targeting itself, so the two tests are not interchangeable. Note also how broadly "negative financial status" can be read: it is not limited to bankruptcy or default, and modeled hardship scores are inference rather than fact.
The US and Canada overlay
For audiences in the United States and Canada, personalized ads for credit — along with housing and employment — may not be targeted on gender, age, parental status, marital status, or ZIP code. The list is narrower than many compliance teams assume and broader than most media plans reflect. A ZIP-level split around affluent neighborhoods, or an age band used as a proxy for the early-career applicant, both fall inside the prohibition once delivery is personalized. The overlay is easy to miss because geo-targeting tools rarely flag the credit category on their own. Scope matters here: this is a personalization rule, not a ban on the audience data existing. Contextual placement, where the page's subject rather than the viewer's profile selects the ad, is governed differently and remains a workable route.
What still works, and on what conditions
Three routes survive. Contextual targeting on content — a page about rebuilding credit or comparing annual fees — carries no reliance on personal financial data. Interest-based advertising remains available if the ad discloses that it is interest-based, for example through the AdChoices icon, and every party in the chain follows an accepted industry code such as the Digital Advertising Alliance's self-regulatory principles. Audience-data use, including cookie lists, requires that the advertiser or publisher hold all rights necessary to that data, and that a privacy policy disclose the collection, sharing, and use connected with Google's products, covering cookies, web beacons, and IP addresses. Each route moves the burden from segment quality to documentation, which is why a well-built compliance file is often the difference between a campaign that scales and one that stalls.
Where offers and claims break
The graver category is concrete promises that cannot be honored. Free or cash offers, implausibly cheap offers, and promises to provide loans appear among Google's egregious examples, as does a specific promise outside the advertiser's control, such as opening a named bank's high-yield account "with no credit check." For card marketing, the practical warning is that unqualified approval language sits close to that line. Traffic-source rules add a second layer: whatever link or ad sent the visitor must accurately describe the destination, must not promise an offer that is absent or hard to find, and must land users on a page that closely matches what an organic visitor sees. Publisher-side content rules separately bar misleading statements, deceptive promotion, and get-rich-quick framing, and they require that paid promotional material not exceed the publisher's own content on the screen.
A practical pre-launch review
Five checks catch most problems before spend:
- Trace each audience segment to its origin and remove anything built on hardship, score, or debt-load signals.
- Strip gender, age, parental status, marital status, and ZIP layers from credit line items serving US and Canadian users.
- Rewrite approval and rate claims as conditional statements tied to disclosed terms.
- Confirm the landing page presents the offer the ad describes, with comparable prominence.
- Log where the interest-based disclosure appears and which self-regulatory code the chain follows.
Run the checks again whenever the creative or the offer terms change, not only at launch.
What this reading cannot settle
Google revises these policies and places responsibility for current compliance on publishers and advertisers, so a review completed last year proves nothing about this quarter. Enforcement goes beyond rejected ads: Google states it may stop serving ads to a site, disable an account, or require changes to how its services are implemented. This analysis covers Google's published rules only. Credit advertising in the US also sits under separate consumer-finance law and state requirements that the policy text does not address, and audience eligibility calls belong with a compliance or legal reviewer rather than a media buyer. Treat the checks above as a screening layer, not a clearance.
The first change to make
Rebuild the credit card plan around page context and disclosed offers, then re-open demographic and hardship layers only where someone can point to specific policy text that permits them, and date-stamp that evidence for the next review.
研究の知恵