AI policy
Last updated: 17 November 2026
How our AI works for clients, in short: it drafts, a person approves, and nothing goes out on its own until the client switches that on. This page is a summary of the AI use terms in our client agreement (version 2.2, 30 October 2026).
What the AI does
Our systems use AI models from other companies to read, sort, summarise and draft. They answer enquiries and take the actions a client has allowed: logging an enquiry, sending matching listings, sending an approved follow-up, answering a call or booking a viewing. The providers are on the subprocessors page.
AI can be wrong
The same question can get different answers, and an answer can be wrong, incomplete, out of date or made up. That is a known limit of the technology. It is why the rules below exist.
Drafts first
For the first month after a system goes live, everything it would send or do outside the client's business waits for a named person at the client to say yes. Every message, report and quote. Nothing goes out without that yes.
Sending on its own is off until the client switches it on
After the first month, the client chooses for each kind of output:
- it always needs approval;
- it needs approval for the first few of its kind, then runs on its own; or
- it runs on its own and is reported afterwards.
Every system stays on "always needs approval" until a named approver at the client changes it. The change is logged.
What always needs a person
Whatever the setting, our systems will not do these without a person's approval:
- agree or change a price, rent, fee, commission or discount;
- commit the client to a cost above the limit in their agreement;
- sign, send for signature or agree any contract, heads of terms, tenancy or lease;
- send anything to a new list of contacts for the first time;
- make any decision that has a legal or similarly significant effect on a person.
Clients agree not to use our systems for decisions like tenant referencing, credit, affordability or eligibility without a separate written agreement that covers human review.
Telling people it is AI
- An AI receptionist or website chat says at the start that the caller or visitor is talking to an AI assistant, and offers a person where the client's handover rules allow.
- Where calls are recorded or transcribed, callers hear a notice first.
- A message that AI drafted and a person approved is sent in the client's name. It does not need to say AI drafted it, unless the law or a platform's rules require it.
Client data and AI training
- We don't use client data to train or fine-tune any AI model, ours or anyone else's, unless the client agrees in writing for a specific purpose.
- We set up the AI providers we use so that, as far as their settings allow, they don't train their models on client data.
- We do use client data to deliver, test, fix, secure and monitor the client's own system.
- We keep general lessons about building better systems, such as which prompt structure works for a lettings enquiry. They contain no client data and can't identify anyone.
Pausing a system
A client can pause any system at any time from the portal. We can pause one without notice if it is sending wrong, harmful or unlawful content, if there is a security risk, or if a platform's rules require it. We tell the client straight away and explain why.
Hidden instructions
Emails, documents and web pages can carry hidden instructions written to manipulate AI. We design our systems to limit the harm: outputs go through the approval rules above, each system has only the permissions it needs, actions are logged, and price and spending actions always need a person.
Outreach
- The client decides who is contacted, approves the messages and keeps the do-not-contact list.
- Every marketing message says who it is from and gives a way to opt out. Opt-outs our systems receive are applied within 2 working days.
- Our systems do not make automated marketing calls with a recorded or synthetic voice.
When models change
AI providers change and retire models. We may move a system to a different model or provider, as long as it still passes its acceptance tests, the protection of client data does not go down, and any new subprocessor is announced first.
Records
The portal keeps a record of every approval, rejection, send, pause and setting change for each system. Clients can export it.
What we don't promise
We promise to build and run each system as agreed, with reasonable skill and care. We don't promise a business result such as a number of leads, a reply rate, revenue or a search ranking. Our systems don't give legal, financial, tax or valuation advice.
