plainstamp

HUD FHEO AI housing advertising builder's guide

Rule: us-hud-fheo-ai-housing-advertising-2024. Source: HUD/OFHEO press release, May 2, 2024 — links to the digital-advertising guidance PDF (companion to the tenant-screening guidance). Statutory framework: Fair Housing Act, 42 U.S.C. §§ 3604(c), 3605, 3613, 3614, 3617; HUD disparate-impact rule, 24 CFR § 100.500. Audience: digital advertising platforms running housing-related ad inventory (Meta, Google, Amazon, programmatic ad-tech), and housing advertisers (housing providers, real-estate agencies, mortgage lenders, tenant-screening services) buying that inventory. Severity: mandatory.

What HUD did

On May 2, 2024 the U.S. Department of Housing and Urban Development released two guidance documents through its Office of Fair Housing and Equal Opportunity. This guide covers the second one — "Guidance on Application of the Fair Housing Act to the Advertising of Housing, Credit, and Other Real Estate-Related Transactions through Digital Platforms". Its sibling on tenant screening is covered in a separate guide.

The advertising guidance addresses AI / algorithmic systems used by digital advertising platforms to target housing-related advertising. It does not create a new statute. It is HUD's official position on how the existing Fair Housing Act applies to a category of automated ad targeting that has grown rapidly since the 2022 Meta-DOJ housing-advertising settlement — and which HUD viewed as needing affirmative supervisory clarification.

The bottom-line rule: the Fair Housing Act's disparate-impact framework (codified at 24 CFR § 100.500) applies to algorithmic ad targeting just as it applies to human-curated audience segmentation. A targeting algorithm that produces a disparate impact on a protected class is unlawful unless the platform can show (a) the algorithm is necessary to achieve a substantial, legitimate, nondiscriminatory advertising-business interest AND (b) no less-discriminatory alternative is available.

Two parallel sets of obligations

The guidance reaches both sides of the housing-ad market:

Platforms. AI / algorithmic ad-targeting systems running housing-related inventory. Includes Meta, Google, Amazon, and the programmatic ad-tech ecosystem (DSPs, SSPs, ad networks, retargeting vendors). The platform's targeting algorithm IS the regulated artifact.

Advertisers. Housing providers (apartment owners, property management), real-estate agencies, mortgage lenders, tenant-screening services, and any party purchasing housing-related ad inventory. The advertiser's targeting choices are the regulated artifact.

These obligations are independent — both apply simultaneously to the same ad. A platform's compliant infrastructure does not absolve the advertiser of liability for the targeting parameters it specifies, and an advertiser's clean targeting choices do not absolve the platform of liability for an algorithm with proxy-based disparate impact.

The five required elements

The plainstamp us-hud-fheo-ai-housing-advertising-2024 rule encodes five required elements, each derived from the May 2024 guidance.

1. No protected-class proxies in targeting

Algorithmic ad-targeting systems for housing-related inventory must not use protected-class characteristics directly or via proxies. The directly-named characteristics (race, color, religion, sex, national origin, disability, familial status) are an obvious floor. The harder enforcement question is proxies — features correlated with protected classes that produce equivalent disparate impact.

Common identified proxies:

Platforms must test their available targeting parameters against the platform's own user-population data and remediate identified proxies before housing inventory ships against them.

2. Audience-segmentation disparate-impact testing

Platforms must test their audience-segmentation algorithms for disparate impact under the same three-step framework that applies to other algorithmic decisions:

  1. Detect. Quantitative testing for disproportionate adverse effect (e.g., disparate-impact ratio relative to the platform's overall user composition; odds ratios; marginal-effect tests). Plus qualitative review.
  2. Identify legitimate rationale. A specific advertising-business interest the algorithm serves. "Maximize ad-relevance" is too generic; the rationale must be specific and traceable.
  3. Search for less-discriminatory alternatives. Alternative algorithms, alternative seed populations, alternative segmentation features. If a less-discriminatory alternative exists that serves the same legitimate rationale, the platform must adopt it.

This testing must occur before deployment AND at regular intervals (annual minimum; quarterly is the de facto standard among large platforms post-Meta-settlement).

3. Advertiser targeting controls

Platforms must provide controls that suppress fair-housing-risky targeting parameters in housing-related campaigns. Concretely:

4. Content moderation for protected-class language

Beyond targeting, platforms must moderate the content of housing-related ads for explicit protected-class language. § 3604(c) prohibits indicating a preference based on protected class — language that does so is unlawful even if the targeting is clean.

Examples of flagged content:

Platforms can use AI-based content moderation but retain liability for false negatives. The platform's moderation must run before ad approval; the platform must give advertisers a specific basis for any rejection (so they can revise rather than guess).

5. Advertiser-side targeting liability

The advertiser's targeting choices are independently liable, regardless of what the platform's automation made appear acceptable.

Practical implications:

Housing advertisers must implement internal controls: pre-launch review of targeting parameters by a compliance reviewer, documented exclusion of protected-class proxies, periodic training, retention of the rationale for any targeting choice.

Common failure patterns

After ~12 months of HUD enforcement and follow-up since the May 2024 guidance:

  1. Lookalike audience without seed audit. Advertiser uploads a customer list as a lookalike seed without auditing the seed for protected-class composition. Algorithm produces a lookalike that inherits the seed's demographic skew. Both platform AND advertiser liable.
  2. "Detailed targeting" parameters not flagged for housing inventory. Platform's housing-ad detection misses ads with subtle housing context (e.g., "near transit" without an explicit housing keyword); detailed targeting parameters that should have been suppressed remain available. Platform liable for the false-negative.
  3. AI-generated ad creative slipping moderation. Generative-AI ad creative includes protected-class indications that bypass keyword-based moderation. Platform's content moderation must adapt to generative-AI output patterns.
  4. Cross-platform targeting via off-platform identity graphs. Advertiser uses a third-party identity graph to retarget across platforms; the identity graph carries protected-class proxies that neither platform's housing-ad workflow detects. Advertiser liable; platform's contributory liability is fact-specific.
  5. No audit cadence. Platform conducts a one-time pre-launch audit and treats the algorithm as static; algorithm drifts as user-population shifts. Periodic-retest obligation reaches all live housing-targeting algorithms.

Stacking with adjacent regimes

Minimum-viable-compliance checklist

For a digital advertising platform running any housing-related ad inventory:

For a housing advertiser purchasing digital inventory:

Sample disclosures

Plain-language platform-side fair-housing-advertising disclosure (≤200 words)

[Platform] complies with the Fair Housing Act (42 U.S.C. § 3604(c)) and HUD/OFHEO guidance dated May 2, 2024 in operating its housing-related advertising inventory.

Housing-detected ad campaigns are routed through a restricted-targeting workflow that suppresses targeting parameters posing fair-housing risk. Ad creative is screened for protected-class language before approval. Audience-segmentation algorithms are audited annually for disparate impact under 24 CFR § 100.500.

Advertisers retain responsibility for the targeting choices they specify. Advertisers cannot direct [platform] to use protected-class proxies; selecting any such proxy in a housing campaign is the advertiser's liability under the Fair Housing Act.

If you believe a housing ad delivered through [platform] was discriminatory, file a complaint with HUD at [hud.gov complaint URL] or with [platform fair-housing contact].

Formal-language advertiser-side targeting disclosure (legal-counsel-grade)

FAIR HOUSING ADVERTISING TARGETING DISCLOSURE

Pursuant to the Fair Housing Act (42 U.S.C. §§ 3604(c), 3605, 3617) and HUD/OFHEO guidance dated May 2, 2024, [advertiser name] discloses the following with respect to housing-related advertising campaigns:

  1. Targeting parameters specified by [advertiser]: [enumerated]. Targeting parameters explicitly excluded as protected-class proxies: [enumerated]. Compliance reviewer: [name and date].
  2. Audience seed (if lookalike): [seed source]; protected-class-composition audit conducted [date] using methodology [description]; result: [pass/fail with documentation reference].
  3. Third-party data enrichment (if any): [vendor]; vendor's representation of protected-class-proxy controls: [link or document]; [advertiser]'s independent audit: [date / result].
  4. [Advertiser] retains primary responsibility for the targeting choices specified in this campaign regardless of [platform]'s automation. Vendor or platform automation is not a defense.
  5. Records of this campaign (targeting parameters, ad creative versions, performance) are retained for [retention period] consistent with state law.

Authoritative sources

Disclaimer

Not legal advice. plainstamp surfaces the published text of real regulations and the May 2024 HUD guidance, with citation back to HUD's published source. For high-stakes housing-advertising deployments, verify against the cited HUD documents and consult counsel licensed in your jurisdiction. The disparate-impact analysis under 24 CFR § 100.500 is fact-intensive and the right answer depends on the platform's user composition, the algorithm's specific configuration, and the available market alternatives.