ARLA Propertymark AI Housing Warnings: What Letting Agents Must Do

ARLA Propertymark's AI warnings put letting agents on notice

ARLA Propertymark has warned its members that using artificial intelligence to write property listings, screen tenants or handle compliance paperwork does not transfer legal responsibility away from the agent. The trade body's guidance makes clear that under the Consumer Protection from Unfair Trading Regulations 2008, the Property Misdescriptions Act legacy rules and the UK GDPR, the agent remains the data controller and the party answerable to Trading Standards. An AI tool that invents a school catchment area or misstates an EPC rating creates a misleading commercial practice, and the fine lands on the agency, not the software vendor.

The warning matters because AI adoption across lettings has moved quickly. Portals including Rightmove and Zoopla now offer AI-assisted listing tools, referencing firms advertise automated decisioning, and a growing number of agencies use chatbots for tenant enquiries. ARLA Propertymark's position is that these tools are permitted, but only with human oversight, documented checks and clear disclosure where automated decisions affect applicants.

Where the liability actually sits

Three areas carry the most risk for letting agents.

1. Property descriptions and marketing

An AI model trained on generic housing data will happily describe a flat as having "ample storage" or being "moments from the station". If that storage is a single cupboard or the station is a 25-minute walk, the listing breaches the CPRs. The Competitions and Markets Authority has pursued estate and letting agents for misleading descriptions, and it treats AI-generated copy the same as copy written by a negotiator. Every listing needs a named human who has verified the claims against the property.

2. Tenant referencing and automated decisions

UK GDPR Article 22 gives applicants the right not to be subject to a decision based solely on automated processing that produces legal or similarly significant effects. A referencing system that auto-declines an applicant on a credit score threshold, with no human review, sits squarely in that territory. ARLA Propertymark's guidance points agents towards keeping a human in the decision loop and being able to explain, on request, how a decision was reached. That means retaining the underlying data and the reasoning, not just the outcome.

3. Data protection and training data

Agents feeding tenancy applications, ID documents or deposit information into a third-party AI tool need a lawful basis and a data processing agreement. Several AI vendors train their models on submitted data by default. Under the Data Protection Act 2018, passing tenant data to a processor without a written contract and without telling the applicant is a breach. ARLA Propertymark has advised members to check vendor terms line by line and to record which tools touch personal data.

What the warnings mean in practice

The practical effect is that AI becomes a drafting and triage tool, not a decision-maker. Agents who get this right tend to follow a consistent pattern.

• Human sign-off on every listing. A named staff member checks square footage, EPC rating, council tax band and transport claims before publication.
• Documented referencing decisions. Automated flags trigger manual review, and the review is logged with a timestamp and reviewer name.
• Vendor due diligence. A data processing agreement is in place, training-on-customer-data is switched off, and the vendor's sub-processors are identified.
• Staff training. Negotiators and property managers know which tools are approved and what they are not allowed to paste into a public AI chatbot.
• Audit trail. Records show what the AI produced, what the human changed and why.

That last point is where most agencies are weakest. When Trading Standards or the Information Commissioner's Office asks how a decision was made, an agency that can produce a clean audit trail resolves the query quickly. One that cannot faces a longer investigation and a higher chance of a penalty. Building that trail into daily workflows is easier with property inventory software that timestamps edits, stores evidence and keeps a version history against each property record.

The Renters' Rights Act context

The warnings land alongside the Renters' Rights Act, which received Royal Assent in 2025 and phases in through 2026. The Act abolishes Section 21 no-fault evictions, moves assured shorthold tenancies to periodic tenancies, introduces a Decent Homes Standard for the private rented sector and creates a new ombudsman for landlords and agents. Each of these changes increases the volume of evidence an agent must hold: possession grounds under Section 8 need documented proof, Decent Homes compliance needs dated inspection records, and ombudsman complaints need a full paper trail.

AI can help manage that volume, but only if the underlying records are accurate. An AI tool summarising a property file is only as good as the inventory, safety certificates and inspection notes it reads. ARLA Propertymark's warnings effectively say the same thing: automate the admin, verify the facts.

Compliance checklist for letting agents

Agents reviewing their AI use against the ARLA Propertymark guidance should work through the following.

• List every AI tool in use, including those embedded in CRM, portal and referencing systems.
• Confirm a lawful basis and a signed data processing agreement for each tool that handles personal data.
• Disable model training on customer data where the vendor allows it.
• Put a named human reviewer against every published listing and every automated referencing decision.
• Keep records of AI output, human edits and the reasons for any override.
• Train staff on what can and cannot be entered into public AI tools.
• Review the position annually, or sooner if the vendor changes its terms.

ARLA Propertymark members can access the trade body's full guidance and its compliance helpline. Non-members can still follow the same principles, and the regulator applies the same standards regardless of membership.

What happens next

The Information Commissioner's Office has signalled that it will look closely at automated decision-making in housing, and the new property ombudsman will have powers to require evidence from agents. Agencies that treat AI as a shortcut around compliance will find the cost of that shortcut rising. Those that treat it as a drafting aid with a human sign-off layer will find the administrative burden of the Renters' Rights Act more manageable than they expected.

The immediate action for any letting agency is a one-page AI register: what tools you use, what data they touch, who reviews the output and where the evidence is stored. That document is the first thing a regulator will ask for.