AI tenant screening and Fair Housing for property managers
What property managers need to know about AI tenant screening: Fair Housing Act disparate impact, HUD guidance status, FCRA notices and practical controls.
By the Ceety Systems teamUpdated 7 min read
Key takeaways
- Using an algorithm or a screening vendor does not move Fair Housing Act responsibility away from the housing provider.
- HUD's 2024 AI guidance is no longer current HUD policy as of September 2026, but the statute, court precedent and the FCRA still apply.
- If a consumer report played any part in a denial or worse terms, the FCRA requires an adverse action notice.
- Explainable criteria, individualized review, outcome testing and vendor due diligence are the core controls at any portfolio size.
AI tenant screening is legal, but it does not change who is responsible. If a screening score or automated recommendation leads you to deny an applicant, the Fair Housing Act and the Fair Credit Reporting Act (FCRA) apply to that decision the same way they would if a person made it by hand. The practical answer is to know what your screening tool looks at, keep a human in the loop for borderline and negative outcomes, test results for uneven effects, and send the notices the FCRA requires.
This is general information, not legal advice. Fair housing law also varies by state and city, so check the rules where your properties are.
What does the Fair Housing Act prohibit in tenant screening?
The Fair Housing Act makes it unlawful to refuse to rent, or to set different terms, because of race, color, religion, sex, disability (the statute says "handicap"), familial status or national origin. That covers intentional discrimination, such as treating applicants differently by group.
It also covers disparate impact: a neutral-looking policy that falls more heavily on a protected group without a sufficient justification. In 2015, the Supreme Court held in Texas Department of Housing and Community Affairs v. Inclusive Communities Project that disparate impact claims can be brought under the Act, while also stressing limits, including that a statistical disparity alone is not enough without a link to a specific policy.
For screening, that is the risk to manage. A model that leans heavily on eviction filings, criminal records or thin credit files can produce different outcomes by race, national origin or disability even if no one intended it.
What is the status of HUD's 2024 guidance on AI screening and advertising?
On April 29, 2024, HUD's Office of Fair Housing and Equal Opportunity issued two documents: guidance on applying the Fair Housing Act to the screening of rental applicants, and guidance on housing advertising through digital platforms. HUD announced them together in May 2024.
As of September 2026, the position has changed:
- The digital platform advertising guidance was withdrawn. A notice published in the Federal Register on April 6, 2026 lists it among eight guidance documents withdrawn effective September 17, 2025, and says those documents should not be relied upon.
- The tenant screening guidance is not named in that notice, but it is no longer posted on HUD's main site and is available only on HUD's archive site. Do not treat it as current HUD policy.
- HUD has proposed removing its disparate impact regulation. A January 14, 2026 proposed rule would remove HUD's discriminatory effects rule and leave the question to the courts. A supplemental proposal in August 2026 reopened comments to October 9, 2026. As of this writing, neither is final.
What has not changed: the statute, the Supreme Court's 2015 decision, and the right of applicants to sue. The withdrawal notice itself says that conduct not complying with the Act remains subject to enforcement, and that complainants may file a civil action within two years. State and local fair housing agencies also enforce their own laws.
So the planning question is not "which HUD document applies," but "can we explain and defend every screening criterion we use."
What does the FCRA require when screening leads to a denial?
A tenant screening report is usually a consumer report under the FCRA, and the screening company is usually a consumer reporting agency. According to the Federal Trade Commission's guidance for landlords, Using consumer reports: what landlords need to know:
- Adverse action includes more than denial. Requiring a co-signer, a deposit not required of others, a larger deposit or higher rent based on the report all count.
- The notice must include the name, address and phone number of the reporting agency; a statement that the agency did not make the decision and cannot explain it; and the applicant's right to dispute the information and get a free copy of the report if they ask within 60 days.
- The notice may be written, electronic or oral, though the FTC recommends writing as best practice because it creates a record.
- Use reports only for housing purposes, as you certify to the agency, and dispose of them securely.
AI raises the stakes on accuracy. A screening product that matches records by name alone, or scores applicants on data they cannot see, produces denials that are hard to explain and easy to dispute.
Practical controls for AI tenant screening
These controls track the statute, the case law and the FCRA rather than any one agency document.
1. Use criteria you can explain
Write down every factor the tool uses and why it predicts whether someone will pay rent and follow the lease. If the vendor cannot tell you which factors drive a score, you cannot defend it. Avoid proxies with weak links to tenancy, such as old eviction filings that did not result in a judgment, or arrests without convictions.
2. Allow individualized assessment
Give applicants a way to explain a negative item and to submit context, such as a resolved debt, a medical collection, or evidence that a record belongs to someone else. Consider requests for reasonable accommodation from applicants with disabilities through this process.
3. Keep a human in the loop
Let the tool recommend; have a trained person decide denials, conditional approvals and borderline cases. Record who reviewed the file and why, so the decision can be reconstructed later.
4. Test for disparate outcomes
Periodically compare approval, denial and conditional-approval rates across groups where you can lawfully do so, and look at which criteria drive the gaps. When a criterion produces a large difference, ask whether a less discriminatory alternative would serve the same purpose.
5. Do vendor due diligence
Before signing, and at renewal, ask the screening vendor:
- Which data sources and match rules do you use, and how do you handle common names?
- Which factors drive the score, and can you give applicant-level reasons?
- Have you tested outcomes for disparate impact, and will you share the method?
- How do disputes work, and how fast are corrections made?
- Can we switch off specific criteria, such as criminal history or eviction filings, to match our policy and local law?
6. Keep the paperwork
Keep a written screening policy, the criteria configured in the tool, adverse action notices, dispute records and test results. If a complaint arrives, this is your evidence.
How does this differ for a small landlord and a large operator?
The obligations are largely the same. The difference is in how much you build and monitor.
| Small landlord or independent manager | Large operator or REIT | |
|---|---|---|
| Screening tool | Off-the-shelf report and score from one vendor | Vendor models plus in-house rules, often across several states |
| Criteria | One written policy, reviewed yearly | Policy by market, reflecting state and local rules |
| Human review | Owner or manager reviews every denial | Trained review team with documented reasons |
| Outcome testing | Periodic check of denials and their reasons | Regular statistical testing, reviewed by counsel |
| Vendor oversight | Questionnaire at signing and renewal | Contract terms for testing, audit rights and data correction |
| Records | Folder per applicant | Case management with retention rules |
A small landlord does not need a data science team. The minimum is a written policy, reading every report before deciding, sending proper adverse action notices and asking the vendor the questions above.
A large operator should treat screening as a governed AI system: an inventory of models and rules, testing before changes go live, monitoring after, and clear ownership. Frameworks such as the NIST AI Risk Management Framework help structure that work. Our real estate page describes how we build screening and leasing workflows with that kind of review and evidence built in.
Frequently asked questions
Is it legal to use AI for tenant screening?
Yes, in general, provided the criteria do not discriminate on a protected basis and do not create an unjustified disparate impact, and you meet FCRA duties. State and local laws may add limits, for example on criminal history or source of income, so check the rules where your properties are.
Does HUD's 2024 tenant screening guidance still apply?
As of September 2026, it should not be treated as current HUD policy. HUD formally withdrew its related digital advertising guidance, and the screening guidance now appears only on HUD's archive site. The Fair Housing Act and the FCRA still apply regardless.
If the vendor's algorithm made the decision, is the vendor responsible instead of us?
Not instead of you. The housing provider makes the rental decision, and using a third-party tool does not remove fair housing or FCRA responsibility. Vendors can have their own obligations, which is why contract terms and due diligence matter.
Do we need to tell applicants that AI was used?
The main federal duty is the FCRA adverse action notice whenever a consumer report contributed to a denial or worse terms. State and local rules on automated decisions are changing, so check the requirements where your properties are, and consider telling applicants how screening works even where it is not required.
How often should we test screening outcomes?
It depends on volume and risk. A small landlord might review denials and their reasons each year; a large operator should test on a regular schedule and whenever a model or criterion changes.