Key takeaways
- Your firm is the controller of call data; the AI supplier is normally your processor and needs a proper contract.
- Tell callers early that they are speaking to an automated system and that the call is recorded, and point to your privacy notice.
- Carry out an impact assessment where the processing is likely to be high risk, and write one anyway as good practice.
- Set retention periods for recordings, transcripts and summaries, and make sure the supplier deletes on time.
- Follow-up marketing texts and calls fall under PECR, separate from service messages.
Why this is a data protection question
An AI system answering your phone collects personal data on every call: a name, number, address, details of the home, sometimes who lives there and whether someone is elderly or unwell. Most systems also record the call and produce a transcript and summary. That is processing under UK GDPR, and the same rules apply whether a person or a machine takes the call. The difference is scale and automation: everything is recorded, stored and searchable by default.
Callers care about this. In a 2024 government survey, around nine in ten adults saw risks in data being sold on for profit (93%), data being hacked or stolen (92%) and important decisions being made without human input (91%)[1].
This guide is general information for trade firms, not legal advice. Data protection law has been amended in places since 2018, including by the Data (Use and Access) Act 2025, so check current guidance or take advice for your situation.
Who is responsible for what
Your firm decides why calls are taken and what happens to the information, so you are the controller. The AI supplier processes the data on your behalf, which normally makes it a processor. UK GDPR requires processing by a processor to be governed by a contract that sets out, among other things, that the processor acts only on your documented instructions, keeps the data secure, does not engage another processor without your written authorisation, and deletes or returns the data when the service ends[2].
- Ask the supplier for its data processing terms before you sign, and read them.
- Check which sub-processors it uses, such as hosting and speech services, and where they are.
- Check whether call data is used to train or improve the supplier's systems, and whether you can switch that off.
- Check how the supplier will help you answer a customer who asks for a copy of their recording or for it to be deleted.
Telling callers what is happening
When you collect personal data from someone, you must tell them, at the time, who you are and how to contact you, why you are processing the data and on what lawful basis, who will receive it, and how long it will be kept, along with their rights[3]. On a phone call that is usually done in two layers: a short greeting, and a fuller privacy notice on your website.
Example greeting: “Thanks for calling Northside Heating. You're speaking to our automated assistant, and calls are recorded so we can deal with your job. Our privacy notice is on our website. How can I help?” (Example firm name.)
- Say it is automated in the first sentence, not halfway through.
- Mention recording before the caller gives details.
- Offer a way to reach a person, or explain when a person will call back.
- Update the website privacy notice to cover AI call handling: what is recorded, who processes it, retention and rights.
Do you need an impact assessment?
Where a type of processing, in particular using new technologies, is likely to result in a high risk to people's rights and freedoms, the controller must carry out a data protection impact assessment before the processing starts[4]. The law lists examples of high-risk processing, including systematic and extensive evaluation of people based on automated processing on which significant decisions are based[4]. A typical AI answering system for a heating firm, which takes down jobs and books visits, may not reach that threshold, but AI used to make decisions about people, record health details at scale or monitor staff could.
Even when one is not required, a short written assessment is the best way to show you thought it through. The law describes what an assessment contains: a description of the processing and its purposes, an assessment of necessity and proportionality, an assessment of the risks, and the measures to address them[4].
Describe the processing
Which calls, what data, recordings and transcripts, who sees them, which supplier and sub-processors.
State the purpose and necessity
Answering calls when nobody else can, taking down jobs, booking visits, routing urgent calls.
Identify risks
Callers not realising it is automated, health details recorded unnecessarily, recordings kept too long, unauthorised access.
Set measures
Clear greeting, minimal questions, retention limits, access controls, a route to a person.
Sign and review
Date it, keep it, and review when you change supplier or add features.
Retention: how long to keep what
You must tell callers how long their data will be kept, or the criteria used to decide[3], so decide before you go live. Different records need different periods.
| Record | Why you keep it | Approach |
|---|---|---|
| Call recording | Resolving disputes about what was said, checking quality | Short period you set, then deleted automatically |
| Transcript | Searching calls, training staff on scripts | Same as or shorter than recordings |
| Job summary | Doing the work | Moves into the job record and follows its retention |
| Calls with no job | None after a short check | Delete on the shortest cycle |
| Gas smell and emergency calls | Evidence of what advice was given | Consider a longer, stated period |
Check the supplier's settings match your policy, and that deletion actually happens at the supplier and its sub-processors, not just in your dashboard.
Health and vulnerability details
Heating firms often need to know whether a household includes someone elderly, disabled or unwell, because it changes how urgent a no-heat call is. Health information is special category data and needs extra care. Ask only what you need to prioritise the job, such as whether anyone in the home is vulnerable to cold, rather than inviting a medical history. Make sure the summary records the priority flag rather than the caller's full explanation, and limit who can listen to recordings.
Follow-up texts and calls
Messages about the caller's job, such as a booking confirmation, are service messages. Promotional follow-ups are marketing. Direct-marketing emails and texts to individuals need prior consent unless the soft opt-in applies to existing customers[5]. Unsolicited marketing calls must not be made to numbers registered with the Telephone Preference Service, once registered for 28 days, unless the subscriber has said they do not object[6]. If your AI tool can send texts or make calls, keep marketing switched off unless you have the right basis.
Requests from callers
Once calls are recorded and transcribed, people will occasionally ask for a copy, or ask for a recording to be deleted. Plan how you will handle these before the first one arrives.
- Know how to find a caller's recordings and transcripts by phone number and date.
- Name the person who handles requests and a backup for holidays.
- Check the caller's identity before sending anything; a recording of someone's home details is sensitive.
- Remember that a call may include other people's information, such as a tenant talking about a landlord.
- Keep a short log of each request and what you did.
Worked example: a firm gets ready
Example: a fictional heating firm plans to use AI to answer evening and weekend calls. Before going live, the owner reads the supplier's processor terms and confirms that call data is not used for training. He writes a two-page assessment, sets recordings and transcripts to delete after a short fixed period and job summaries to move into the job system, and changes the greeting so callers hear that the call is automated and recorded. The website privacy notice gets a new section on call handling. The AI asks one question about vulnerability, whether anyone in the home is especially at risk from the cold, and records only a yes or no.
Checklist before you go live
- Processor contract signed, sub-processors listed, training use understood.
- Greeting says automated and recorded; route to a person available.
- Privacy notice updated.
- Impact assessment written and dated.
- Retention periods set and tested at the supplier.
- Access to recordings limited to named staff with two-step sign-in.
- Marketing messages off unless consent or soft opt-in applies.
- A process for requests to access or delete recordings.
For the recording rules themselves, see call recording law for trades; for choosing a tool, the AI phone tool checklist.
Frequently asked questions
Do I have to tell callers they are speaking to AI?
You must tell callers how their data is used, and being upfront that the call is automated and recorded is the clearest way to do that. Say it at the start of the call.
Is the AI supplier responsible for data protection?
Your firm is usually the controller and stays responsible. The supplier is normally a processor and must work under a contract with the terms UK GDPR requires.
Do I need a DPIA for an AI receptionist?
One is required where processing is likely to be high risk. Many simple call-answering set-ups may not reach that threshold, but writing a short assessment is good practice either way.
How long can I keep call recordings?
As long as you need them for a stated purpose, and no longer. Set a period, tell callers in your privacy notice, and make sure the supplier deletes on schedule.
References
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) (2016). UK GDPR Article 28: Processor (opens in a new tab) ↑
legislation.gov.uk (The National Archives) (2016). UK GDPR Article 13: Information to be provided where personal data are collected from the data subject (opens in a new tab) ↑
legislation.gov.uk (The National Archives) (2016). UK GDPR Article 35: Data protection impact assessment (opens in a new tab) ↑
legislation.gov.uk (The National Archives) (2003). The Privacy and Electronic Communications (EC Directive) Regulations 2003, regulation 22 (opens in a new tab) ↑
legislation.gov.uk (The National Archives) (2003). The Privacy and Electronic Communications (EC Directive) Regulations 2003, regulation 21 (opens in a new tab) ↑