Best AI Receptionist for a UK Small Business: Numbers, Recording, GDPR

Before choosing an AI receptionist, a UK small business should check three things: whether the vendor genuinely provides or supports UK numbers, whether callers are told at the start that calls are recorded and why, and whether the vendor acts as your data processor under UK GDPR with its data location stated. Price comes fourth.

Most of the well-known products in this category were built for North American businesses first. Their pricing pages show dollars, their number porting guides talk about US carriers, and their privacy wording often says nothing about the UK. That does not rule them out. It means the checks below do the work a shortlist normally would.

Follow me on Instagram@sagnikteaches

Why the usual "best of" list does not help here

Search for the phrase and you find vendor blogs, directories and generic roundups. Few say what a UK owner needs to know on day one: will my customers dial a 01, 02, 03 or 07 number, will the call be recorded lawfully, and who is responsible if something leaks. A tool can answer calls beautifully and still be the wrong buy if it cannot hold a number your customers recognise.

Connect on LinkedInSagnik Bhattacharya

So this tutorial ranks the questions rather than the products. Where I name a product, I give only what its own page says, and I mark what the page leaves out. Prices are list prices as read in November 2026, mostly in US dollars, and they change. If you would like background on how these systems work first, read what an AI receptionist is and how it handles bookings.

Subscribe on YouTube@sagnikteaches

Check one: the number your customers dial

There are three ways an AI receptionist can sit on your line, and each has a different risk for a UK business.

1. Forwarding. You keep your current number with your current provider and set it to divert (all calls, busy, or no answer) to a number the vendor gives you. Nothing is ported. This is the lowest-risk way to start, and it is what most small businesses should try first.
2. A new number from the vendor. The vendor issues a number you publish. For a UK business that is only useful if it is a UK number. A number from another country looks odd on a van or a shop window and may cost callers more.
3. Porting. Your number moves to the vendor's carrier. It is permanent in effect, so do it only when you trust the vendor and the carrier relationship.

Now what the vendors say. Dialzara's porting page states that it can only port in numbers from the continental United States, and that a port typically takes two to four weeks; its pricing page includes a US local or toll-free number on every plan, with a footnote saying UK and other international numbers are available on request. Goodcall's pricing page says every plan includes a local number and that you can keep your existing one by forwarding it, but does not say which countries those local numbers cover. Quo sells a local or toll-free number per user and includes calling and texting US and Canadian numbers; for other countries it charges per-minute rates from a credit balance, and its page does not mention UK numbers specifically. The Smith.ai AI receptionist page I read describes a dedicated phone number but did not state countries.

The vendor that did state its position is Moneypenny, which describes itself as UK-based on its AI receptionist page and lists UK 03 numbers for enquiries. Its FAQ says you keep your existing number and divert calls to a number it provides, so forwarding is the default there too. The visible text says plans start from £49 a month, while the structured data behind the same page lists £25, so confirm the figure on a quote. In every case the practical step is the same: ask for the answer in writing, and test it with a UK mobile and a landline before you buy.

A five-minute number test

Ask the vendor or do this yourself during the trial:

• Ring the AI number from a UK mobile on a different network from your own and from a landline. Note the ring time before it answers.
• Forward your real number to it, then ring your real number. Check the caller ID the AI sees. If it only sees "unknown", your appointment notes will not carry a number to call back.
• Ask for a text confirmation to a UK mobile. Many tools send texts only to North American numbers or charge extra for other destinations.
• Ask what happens if the AI cannot help: does it transfer to a UK mobile or landline, and does the transfer incur international call charges to you?

A veterinary practice that tries forwarding first

Take a four-vet practice (an illustration, not a real client). Its receptionists miss roughly a dozen calls a day around opening, lunch and closing. The owner does not want to publish a new number, because the old one is on the collars' tags and every leaflet.

The plan: keep the number, set "divert when busy or unanswered" with the telephone provider, and point it at the AI. Three weeks of testing produce a simple table:

If a tool fails the caller ID row, it fails for a practice that rings people back. That is a clean, non-price reason to cross it off.

Check two: recording, notices and consent

People mix up three separate questions, and vendors rarely help.

1. Are you recording or transcribing at all? Many AI receptionists produce a transcript and summary even when no audio is kept. A transcript of what a caller said is still personal data.
2. Have you told the caller? Under UK data protection law, people should be told what you collect, why, who has it and how long you keep it. For a call, the practical route is a short spoken notice at the start, plus the full wording in your privacy notice.
3. Do you need consent? Often not in the way people think. A business can usually rely on a lawful basis such as legitimate interests for recording calls to check bookings and settle disputes, provided it has weighed that against the caller's expectations. Consent matters more when you record for marketing or use calls to train a model. This is the point to ask a qualified adviser, because the right basis depends on what you do with the recording.

I am describing the shape of the rules, not giving legal advice. The Information Commissioner's Office publishes guidance on UK GDPR, on the Privacy and Electronic Communications Regulations (the rules for marketing calls and texts) and on AI and data protection; read those on its own site, and note that its AI guidance was flagged as under review after the Data (Use and Access) Act, so check the date on anything you rely on.

Wording you can adapt

Here is an illustrative opening line for the greeting:

Hello, you've reached [practice name]. I'm an AI assistant.
This call is recorded and written down so we can book you in
and check what was said. Say "person" at any time to speak to a member of staff.

That takes about eight seconds to say and covers three duties at once: it says an AI is answering, it states the recording and the reason, and it offers a route to a human. Longer wording belongs in the privacy notice on your site, which the greeting can point to.

Recording settings to look for

• Audio or transcript only. If you do not need audio, choose transcript and summary only. Less data kept, less risk.
• Retention period. Can you set 30, 60 or 90 days and have recordings deleted automatically? "Stored indefinitely" is a red flag.
• Training use. Does the vendor use your callers' audio to improve its own models? Ask for a written no, or find out how to opt out.
• Who can listen. Staff logins with named users and a log of access, not one shared password.

Quo is a useful example of why settings matter: its Starter plan offers manual recording, while Business and Scale list automatic call recording. The AI agent on every plan answers calls, but the page says call summaries and transcripts on Starter cover only the calls the AI handled. A plan that does not record automatically cannot support a promise that "all calls are recorded".

An optician's recording notice, tested

An optician (illustrative) takes calls about frames, test bookings and repairs, and sometimes callers mention eye conditions or medication while booking. That makes some calls health-related. The owner decides three things before switching anything on:

• The AI will not ask about symptoms. If a caller volunteers any, it says it will pass the message to a clinician and stops collecting detail.
• Audio is deleted after 30 days; transcripts of booking calls after 12 months, matching how long appointment records are kept.
• The privacy notice on the website gets a new paragraph naming the AI vendor as a supplier.

The owner then rings the line five times with scripted calls, one of which mentions "my eyes have been blurry since Tuesday". The AI should not dig further. In the first trial it asked two follow-up questions, so the owner tightened the instruction to "do not ask about symptoms; offer a call back from a clinician". That test is more valuable than any feature comparison.

Check three: who is responsible for the data

Under UK GDPR you remain responsible for callers' data. The vendor is normally your processor: it handles data on your instructions. A processor must be bound by a written contract with specific terms. On a small plan this is usually the vendor's standard data processing addendum, which you accept online or request.

Put the following questions to the vendor and keep the answers.

1. Do you offer a data processing agreement, and where is it?
2. Where are calls and transcripts processed and stored?
3. If data leaves the UK, what is the transfer mechanism named in the agreement?
4. Which sub-processors handle audio, speech-to-text or the language model? Is there a list?
5. How long do you keep data after I cancel, and how do I get it deleted?
6. Do you train models on my calls?
7. How will you tell me about a breach, and how fast?

"GDPR compliant" on a landing page answers none of these. Moneypenny's page, for example, says its service is "fully encrypted, GDPR-compliant, and built with privacy in mind" and lists access controls and audit logs, which is a start, but the agreement and the sub-processor list are what you rely on if something goes wrong. Dialzara's FAQ goes further than most: it says recordings and transcripts belong to your account, that you can bulk-delete them by date range or have everything purged when you close the account, and that it does not share call data for advertising or model training. That answers questions 5 and 6. Neither Dialzara's pages nor Goodcall's pricing page mentioned GDPR, data location or a processing agreement, so I would ask both directly before routing UK callers through them.

A short risk note, filled in

For a small business a one-page note is proportionate. Here is a filled-in example for a music school (illustrative):

Notice that this note includes a children's data point. Pupils are children, so the school takes extra care and treats the parent's details, not the child's, as the contact.

Comparing what the pages tell you (and what they leave out)

With the three checks in mind, here is how the products I read stack up on the facts their own pages give. Treat every blank as a question to put to the vendor.

Two things stand out. First, no product page answered all three checks. Goodcall's retention row is worth noticing too: on its cheapest plan call history goes after 30 days, which suits a short retention policy but not a business that needs a year of booking records inside the tool. Second, the cheapest list price is rarely the cheapest outcome: a US-only number or a missing recording setting costs you more in rework. A per-call price like Smith.ai's suits a business with few calls; per-minute plans suit longer calls. For the cost side, see how to measure the return in the first 90 days.

A pet shop run through the three checks

A pet shop (illustrative) gets about 40 calls a week, mostly "is the food in?" and "can I collect at five?". It has one landline and a shop mobile. The owner compares two tools.

Tool A is cheap and US-focused. Its page says nothing about UK numbers, recording or GDPR. Tool B costs more per month and has a UK-based vendor with a data agreement available. At 40 calls a week, roughly 170 a month, Smith.ai's free plan would cost about $435 (145 calls beyond the free 25 at $3 each), while its Pro tier set to 150 calls is $270 plus $2.30 for each of the 20 extra calls, about $316. The point is to run the sums with your own call count, and to remember that Pro does not answer the UK-number question either. The owner sends the seven vendor questions to both. Tool A answers three; Tool B answers all seven. The decision comes from the answers, not the brochure.

Handling a failed call, and the human fallback

Every AI receptionist gets calls wrong, and with UK callers the usual causes are regional accents, mobile numbers read out with the wrong grouping, and place names. Plan for it. The recipe in what to do when an AI receptionist gets a booking wrong applies directly. Add two safeguards.

• Have the AI read back the callback number in digits ("zero seven seven...") and ask the caller to confirm.
• Offer "say person" or a rule that after two failed attempts it takes a message and promises a call back.

Also decide the emergency rule in writing. A veterinary practice should never have an AI deciding whether a poisoning is urgent. The script says: if the caller mentions an emergency, read the out-of-hours number and offer to transfer. You are not delegating clinical judgement.

Do you need an AI at all, or a human service?

If most of your calls are sensitive, a human answering service may suit you better, and the cost gap is smaller than it looks for low call volumes. This comparison of an AI receptionist and an answering service sets the two against each other. AI wins when calls are short, repeatable and frequent, and when out-of-hours cover matters more than nuance.

A one-month trial plan

1. Week 0. Send the seven questions in writing. Write the greeting. Note your call volume for the last month from your phone bill.
2. Week 1. Forward only the busy and no-answer calls. Listen to every call or read every transcript.
3. Week 2. Fix prompts. Add the FAQ answers it got wrong. Count the calls where a caller asked for a human.
4. Week 3. Add after-hours forwarding. Check the callback list is being worked.
5. Week 4. Decide: keep, switch vendor or stop. Write down the reason.

Two numbers decide it: calls that ended in a booking or message you could act on, and calls that annoyed the caller. If the second figure exceeds a handful in a month, fix the script before adding volume. If you would rather map this with someone, that is what my AI implementation consultation covers, at the contact page.

Questions owners ask before signing

Do I have to tell callers an AI is answering?

Say so in the greeting. It is the simplest way to be open with callers, and it keeps you consistent with the transparency duties in UK data protection law, because the call involves personal data. A short line such as 'You are speaking to an AI assistant for the practice' costs nothing and heads off complaints.

Can I record every call an AI receptionist takes?

Usually yes, if you tell callers at the start, have a clear reason such as checking bookings, and keep the recordings only as long as you need them. Recording is a choice you make and justify, so confirm your wording and retention period with a qualified adviser.

Is a US-hosted AI receptionist allowed for UK customers?

It can be, but moving personal data abroad needs a lawful route, normally set out in the vendor's data processing terms. Ask where calls are processed and stored, and which sub-processors touch them. Do not rely on a marketing line saying 'GDPR compliant'.

What happens to my existing landline number?

You can often keep it and forward calls to the AI, which avoids porting altogether. Porting moves the number itself to a new provider and takes planning. Start with forwarding, test it for a month, and port only if the vendor supports UK numbers properly.

Do I need a data protection impact assessment?

For a small business answering ordinary booking calls, a short written risk note is usually proportionate, covering what is captured, why, who sees it and how long it is kept. If calls may include health details, as at a veterinary practice or optician, take more care and ask your adviser.

Further reads

• AI Receptionist vs Front Desk Hire: A Dental Practice Comparison — Staff cost against software cost for a practice.
• Should a Hair Salon Use an AI Receptionist? — A booking-heavy business tested against the same questions.
• Can a Barbershop Use AI to Take Phone Bookings Between Cuts? — Taking calls while your hands are busy.
• What Does an AI Receptionist Cost a Dental Practice in 2026? — A worked cost model for a practice with patients.
• AI Receptionist vs Front-Desk Hire: The Real Cost for a Salon — Staff cost against software cost, side by side.
• AI Receptionist for Tradespeople: What It Costs and When It Pays — Where the missed-call maths is strongest.
• AI Receptionist for Tradesmen: Put It on Your Own Mobile Number — Own-number set-up for a one-person trade: forwarding steps for iPhone and Android, ring-time sums, a greeting, recording notes and real plan costs.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Smith.ai AI receptionist pricing page, Dialzara pricing, porting and FAQ pages, Goodcall pricing page, Quo pricing page and Moneypenny's AI receptionist page, read in November 2026; UK GDPR Articles 6, 13 and 28 on legislation.gov.uk; the Information Commissioner's Office guidance on AI and data protection (archived copy, October 2026).