HUD Publishes Notice on the Use of Technology for Housing-Related Advertisement
By: Garit Frye, Summer Policy Intern
On April 29th, HUD published a notice titled “Guidance on Application of the Fair Housing Act to the Advertising of Housing, Credit, and Other Real Estate-Related Transactions through Digital Platforms.” The notice contains information about the potential for discrimination and liability issues under the Fair Housing Act when using automated systems, such as algorithms and Artificial Intelligence (AI), for housing-related advertisement, as well as recommendations for housing providers on how to avoid those issues.
Discriminatory Advertising Under the Fair Housing Act
The notice first provides some general information about liability under the Fair Housing Act. It defines intentional discrimination as the use of a protected characteristic, or a proxy, as the basis for housing-related decisions. However, it notes an action can still be liable under the Act if it has a discriminatory effect (regardless of intent) and is unnecessary to achieve a non-discriminatory interest or could be achieved by a less discriminatory alternative. It also notes that housing providers can be held liable under the Act for actions that produce a discriminatory effect, regardless of whether it was deliberate or not, how much they did or did not know about the action, if it was committed by a third party (whether they had an actual or apparent relationship with), or even if it was committed by an automated system or AI.
The Fair Housing Act also explicitly prohibits discriminatory statements and advertising practices. The actions prohibited include:
- Doing any housing-related advertisements that indicate a preference, limitation, or discrimination based on a protected characteristic or their proxy;
- Running advertisements in areas that deny certain groups from receiving information about housing opportunities; and
- Refusing to publish housing-related advertisements or requiring different charges or terms because of protected characteristics.
Potential Liability Issues and Advertising Technology
The notice lists a few technological tools used in advertising that pose potential liability issues under the Fair Housing Act, including audience categorization, custom and mirror audience tools, and algorithmic delivery functions.
Audience categorization is when advertisers sort their audience based on some characteristics or activities. HUD states there are a few ways this can cause discriminatory effects and liability issues, including:
- Sorting potential audiences based on protected characteristics or their proxies;
- Excluding or limiting protected class groups from accessing information about housing opportunities; and
- Showing different content to different groups based on protected characteristics.
Custom audience tools are when advertisers upload a list of people’s identifying information to create a “source list,” and then advertisements are shown to that list. A source list can also be created by adding people who perform a certain action (e.g., visiting a particular website) being tracked by an advertiser or ad platform. A mirror audience tool is related to custom audience tools, as they try to find users similar to the ones on a source list. Both tools could be discriminatory if based on a protected characteristic or a close proxy. Also, even basing the lists on actions could potentially be discriminatory if most of the people performing that action are predominantly from one protected class.
Algorithmic delivery functions use machine learning or AI to consider a wide range of factors to determine what ad(s) to deliver to a consumer at a particular moment. There are a few ways these functions can be discriminatory when running housing-related advertisements, including:
- Using protected characteristics or their proxies as one of the factors;
- Excluding some protected groups by choosing not to deliver ads to them;
- Concentrate delivery to a protected group; and
- Delivering different versions of an ad to different audiences based on protected characteristics or their proxies.
There are also some components of algorithmic delivery functions that could lead to the above actions, including differential pricing and level of confidence. Differential pricing is where the algorithms choose whether to deliver an ad based on how much the advertiser is paying, as they identify some groups as being “worth” more for certain ads. Levels of confidence are just when the algorithm is more or less confident in its prediction that consumers will interact with an ad. Both components can lead to protected groups being shown ads less frequently, thus causing a discriminatory effect.
Recommendations
To avoid discriminatory effects and potential liability issues under the Fair Housing Act when using technology-based advertising tools, HUD recommends that housing providers:
- Utilize ad platforms that are taking steps to manage the risk of discriminatory delivery of housing-related ads;
- Before using an ad platform, obtain the necessary information and disclosures regarding how they mitigate the risk of discriminatory delivery;
- Follow ad platform instructions to ensure that housing-related advertisements are identified as such, so the platform can appropriately handle them;
- Carefully consider the composition of the source lists used for custom and mirror audience tools for housing-related ads;
- Make considered use of any tools provided by the ad platform for evaluating the project demographics of a target audience;
- Monitor the outcomes of advertising campaigns for housing-related ads to identify and mitigate discriminatory outcomes; and
- Do not utilize categorization tools that segment audiences based on protected characteristics or their proxies.
To read the full notice, see here.