Key takeaways
- Staff are already using AI: around 36% of UK workers report using it[1], and 44% of UK adults used an AI chatbot at least monthly in 2024[2]. A policy brings that use into the open.
- The core rule is about data: no client personal data, ID documents, financial details or confidential deal information in public AI tools. Approved business tools need a proper processor contract[3].
- Every AI output is a draft. A named person checks facts, law and tone before anything reaches a client, a portal or a tenant.
- Keep humans in charge of decisions about people: 91% of UK adults are concerned about important decisions being made by computers without human input[4].
Why agencies need an AI policy now
AI writing and summarising tools have spread faster than most agencies' policies. Negotiators use them to draft property descriptions, property managers to tidy repair emails, administrators to summarise long documents. The ONS estimates that around 36% of UK workers report using AI[1], and smaller firms are catching up: 28% of businesses with fewer than ten employees reported using at least one AI technology, against 49% of the largest[5].
Source: Department for Science, Innovation and Technology (GOV.UK), 2024[2]
Show the data
| Item | Value |
|---|---|
| At least once a month | 44% |
| At least once a week | 23% |
Most of this use is helpful. The risks come from what goes in and what comes out: client data pasted into tools the agency does not control, and confident but wrong text published without checking. A short, practical policy lets staff keep the benefits while protecting clients and the business.
The data rules: a traffic light
| Information | Public AI tools (personal or free accounts) | Approved business tools | Notes |
|---|---|---|---|
| Generic text: blog ideas, a template letter with no names | Allowed | Allowed | Check the output before use |
| Property features from published details | Allowed if already public | Allowed | No owner names or security details |
| Client names, phone numbers, emails, addresses | Never | Only if the tool and use are approved | Personal data under UK GDPR |
| ID documents, bank details, references, payslips | Never | Never, unless a specific approved process | High risk for fraud and data breaches |
| Offers, chain details, negotiating positions | Never | Only if approved | Confidential to clients |
| Health, vulnerability or criminal information | Never | Never without a specific assessment | Special care required |
| Passwords, system screenshots | Never | Never | Security risk |
Make the default simple: if a document contains a client's name or anything they told you in confidence, it does not go into a public tool.
Approving business AI tools
Some AI tools are built for business use, with contracts that cover data protection. Before staff use one with client data, someone senior should check it. When the provider processes personal data on your behalf, UK GDPR Article 28 requires a written contract and processors that provide sufficient guarantees, with terms covering documented instructions, confidentiality, security, sub-processors, assistance, deletion and audits[3].
- Does the provider use your data to train its models, and can you switch that off?
- Where is data stored and processed, and are international transfers covered?
- How long is data kept, and can you delete it?
- Who at the provider can see your data?
- Does it support individual log-ins, multi-factor sign-in and access controls?
- Can you export or delete everything if you stop using it?
Keep a list of approved tools and what each is approved for. Anything not on the list is treated as a public tool.
What staff may and may not use AI for
Allowed, with checking:
- Drafting property descriptions from the agent's own notes and published facts.
- Rewriting letters and emails for clarity and tone, with personal details removed or in an approved tool.
- Summarising guidance, legislation or long internal documents for the team.
- Preparing training materials, scripts and checklists.
- Brainstorming marketing ideas, social posts and headlines.
Not allowed:
- Making or recommending decisions about tenants, such as referencing outcomes or which applicant gets a property.
- Generating fake reviews, testimonials or images that misrepresent a property.
- Altering photos in a way that hides defects or changes what a buyer or tenant would see.
- Producing legal notices, tenancy agreements or anti-money laundering decisions without qualified human review.
- Impersonating a real person in calls or messages.
Checking AI output before it reaches a client
Facts
Check every measurement, feature, tenure detail, council tax band and price against your records. AI tools invent plausible details.
Material information
Make sure nothing material is omitted or overstated. The government plans guidance on the quality of information in property listings to help agents meet their legal responsibilities[6].
Law
Confirm that any statement about rights, notices or deadlines is current. AI tools may describe rules that have changed, such as tenancy law in England since May 2026.
Fairness
Remove wording that could discriminate, such as phrases implying a preference for certain tenants or excluding people on benefits or with children.
Tone
Read it aloud. It should sound like your agency, not a generic brochure.
Sign-off
The person who publishes or sends it is responsible for it, whatever tool produced the draft.
Customer-facing AI
Some agencies use AI to answer calls, chat on websites or reply to portal enquiries. Customers deserve to know when they are dealing with an automated system, and they should always be able to reach a person. Public attitudes are cautious: 91% of UK adults are concerned about important decisions being made by computers without human input[4]. The same principles apply whether the tool drafts an email or speaks to a caller.
- Tell callers and website visitors when they are speaking to an AI assistant.
- Limit what the tool can do to clear tasks, such as answering property questions, taking viewing requests and passing details on.
- Make handover to a person quick and reliable, with full notes so the customer does not repeat themselves.
- Review a sample of conversations regularly for accuracy and tone.
If you use an AI Employee such as Ava for calls, the policy should record what it handles, what it passes to staff and who reviews its conversations.
Security: AI-assisted fraud
Criminals use AI too. Phishing was the most common type of cyber attack on UK businesses in 2025/26, experienced by 38% of businesses[7], and AI makes fraudulent emails more convincing. Voice and video imitation adds a new risk for agencies handling client money and completion funds. Remind staff that a convincing message or call from a "director", "solicitor" or "landlord" asking for an urgent payment or bank detail change must still be verified through a known number, every time.
AI in recruitment and staff decisions
The policy should cover the agency's own use of AI as an employer, not just client work. Tools that screen CVs, score interview answers or analyse staff performance can bake in unfairness, and the agency remains responsible under equality law for the decisions it makes.
- Do not let an AI tool reject candidates automatically. A person should review every application against the role's criteria.
- If you use AI to draft job adverts, check them for wording that could deter applicants on grounds of age, sex, disability or other protected characteristics.
- Never feed staff health information, grievance details or disciplinary records into a public tool.
- Tell candidates and employees if AI tools are used in a process that affects them, and how a person reviews the outcome.
The same caution applies to tenant selection. Referencing decisions belong to people applying consistent, lawful criteria, with the reasoning recorded.
Who is responsible for what
Spread responsibility clearly so the policy does not become nobody's job:
- The owner or a director approves the policy and any new business AI tool.
- The data protection lead keeps the approved tools list, checks contracts and handles incidents.
- Branch and lettings managers make sure their teams are trained and follow the checking steps.
- Every member of staff is responsible for what they put into a tool and for anything they send or publish.
Write the names, not just the roles, on the policy's first page, and update them when people change jobs.
A one-page AI policy outline
- Purpose: why the agency allows AI tools and what it wants to protect.
- Scope: all staff, contractors and outsourced teams, on any device.
- Approved tools: the list, what each is for and who approves additions.
- Data rules: the traffic light above, in plain language.
- Permitted and prohibited uses, with agency examples.
- Checking: the person who sends or publishes is responsible, using the checklist.
- Customer-facing AI: disclosure and human handover.
- Incidents: report any mistaken upload of client data the same day to the data protection lead.
- Training: at induction and when the policy changes.
- Review: every six months, because tools and risks change quickly.
Add it to the staff handbook, and make sure outsourced and offshore teams follow it too.
Training and review
A policy only works if people understand it. Run a 30-minute session using real examples: a property description with an invented feature, a tenant letter that gives the wrong notice period, a request to paste a reference into a chatbot. Ask the team which tools they actually use; the answers often surprise managers. Review the policy every six months and after any incident, and update the approved tools list as contracts change.
Example: the description that invented a garden
Example: a negotiator in a fictional agency uses an approved tool to draft details for a ground-floor flat from her notes. The draft describes "private south-facing garden", but the flat has shared access to a communal lawn. The checking step catches it before publication. At the next team meeting the manager uses it as the training example, and the agency adds "outside space and parking" to the checklist of facts that must be verified against the instruction notes.
Frequently asked questions
Can estate agents use AI to write property descriptions?
Yes, as long as the draft is based on accurate notes, contains no invented features, and a named person checks every fact and the overall impression before publication.
Is it a data breach to paste a tenant's details into a chatbot?
It can be. Client personal data should only go into tools the agency has approved under a proper processor contract. Report any mistaken upload to the data protection lead the same day.
Should we tell clients we use AI?
Tell customers when they are interacting directly with an AI assistant, and give them an easy route to a person. For drafting tools used internally, your privacy notice should cover any processing of personal data.
References
Office for National Statistics (2026). Artificial intelligence in UK businesses: 2023 to 2026 (opens in a new tab) ↑
Department for Science, Innovation and Technology (GOV.UK) (2024). Public attitudes to data and AI: Tracker survey (Wave 4) report (opens in a new tab) ↑
legislation.gov.uk (The National Archives) (2026). UK GDPR Article 28: Processor (opens in a new tab) ↑
Department for Science, Innovation and Technology (GOV.UK) (2024). Public attitudes to data and AI: Tracker survey (Wave 4) report (opens in a new tab) ↑
Office for National Statistics (2026). Artificial intelligence in UK businesses: 2023 to 2026 (opens in a new tab) ↑
Ministry of Housing, Communities and Local Government (GOV.UK) (2026). Home buying and selling reform roadmap (opens in a new tab) ↑
Department for Science, Innovation and Technology (GOV.UK) (2026). Cyber security breaches survey 2025/2026 (opens in a new tab) ↑