What is an AI phone survey, and how does it work?

An AI phone survey is an automated voice call in which an AI agent asks scripted survey questions, listens to spoken answers, and records them in real time. It combines speech-to-text to hear the reply, a language model to interpret it and choose the next question, and text-to-speech to speak back naturally — so it feels like a conversation, not a recorded broadcast. It can ask follow-ups, let people repeat or clarify, and hand off to a human where needed.

Is an AI phone survey the same as a robocall or an IVR "press 1" survey?

No. A robocall plays a fixed recorded message to everyone, and legacy IVR asks you to press keypad digits against pre-set options. An AI phone survey is conversational: it understands free-spoken answers, adapts its follow-ups, and captures open-ended verbatims — closer to a human-led interview than a recorded blast. That difference also matters legally, because PECR's strict rule on recorded-message calls targets marketing broadcasts, not genuine research.

Is it legal to make automated or AI survey calls in the UK?

Yes — genuine market-research and opinion-survey calls are lawful in the UK, including automated and AI ones. The strict PECR consent rule for recorded-message calls (Regulation 19) applies only to calls "for direct marketing purposes," and genuine research is not direct marketing. You must still obey Ofcom's persistent-misuse rules (no silent or abandoned calls), present a Caller ID, and comply with UK GDPR. If the call promotes, sells, or gathers data for later marketing, it becomes direct marketing and the full consent rules apply.

Do research calls need consent, and can they call TPS-registered numbers?

Because genuine research is not direct marketing, PECR's automated-call consent requirement and the TPS-screening duty (Regulation 21) do not apply to it the same way, so research calls are not strictly barred from TPS-registered numbers. That is a narrow exemption, not a free pass: you still need a UK GDPR lawful basis, must honour any request not to be called, and reputable researchers follow the MRS Code of Conduct. The exemption disappears the moment a "survey" carries any selling or lead generation.

Do you have to tell people they are speaking to an AI?

No UK law currently mandates an explicit "you are talking to an AI" announcement, but UK GDPR's fairness and transparency principles mean you should be open about AI involvement, and the ICO expects that. The EU AI Act (Article 50) will require such disclosure from 2 August 2026, though it binds UK operators only where the AI interacts with people in the EU. Best practice is to disclose up front — respondents may otherwise complete a whole interview without realising it was not human.

Are the calls recorded, and is that OK under UK GDPR?

Calls can be recorded, but recording personal data needs a lawful basis under UK GDPR, and people should be told at the start — a short notice in the opening line is normal. You must store recordings securely, keep them only as long as needed, and honour data-subject rights. Done properly, recording is fully compliant; skip the notice or lawful basis and it is not. Keeping a structured transcript rather than raw audio minimises what you hold and lowers your risk.

Are AI phone surveys accurate and reliable?

For structured, closed questions, well-designed AI phone surveys can capture responses consistently and follow research best practice, such as fixed wording and answer-order randomisation. Accuracy depends on design and audio conditions more than the technology itself: background noise, strong accents, or a glitch that repeats a question can cause break-offs. Independent academic evaluation is still early, so validate any provider's accuracy claims against your own data.

Can AI handle open-ended answers and follow-up probing?

AI can transcribe and code open-ended answers at scale and ask basic follow-ups, which removes the keypad limits of old IVR. But because it interprets free answers rather than matching fixed options, two identical replies are not always handled identically, and AI is currently weaker than a skilled human at deep probing and steering novel tangents. For high-stakes work, human review of the coded verbatims is still advisable.

Do AI phone surveys get better or worse response rates?

Traditional telephone response rates have collapsed — Pew Research measured phone-survey response falling from 36% in 1997 to 9% in 2012 and around 6% by 2018 — which is the backdrop for interest in automation. Some providers report higher completion for conversational voice than for email or SMS, but robust independent evidence on AI specifically is still thin. Treat response-rate gains as varying by audience and design, not guaranteed.

How much do AI phone surveys cost, versus a human interviewer?

Cost varies widely with survey length, sample size, targeting and language, so any single figure is illustrative. The saving over human interviewing comes from automation and speed — it removes much of the per-interview labour cost that dominates traditional telephone fieldwork. UK quantitative pricing differs from US qualitative benchmarks, so get a per-project quote rather than relying on a headline rate.

Can AI replace human interviewers — which gives better data?

It depends on the study. AI may reduce some human-interviewer biases (such as social-desirability and acquiescence bias) and scales cheaply and fast, but it is currently weaker at probing, managing pace, converting reluctant respondents, and holding people through long interviews. AI suits structured, tracking and attitudinal surveys; humans remain preferable for exploratory, highly sensitive, or vulnerable-respondent work. Many teams use a hybrid — AI for volume, humans for depth.

Will people take part — and can they tell it is not human?

Modern conversational voice AI can sound natural enough that some respondents do not realise they were not speaking to a person unless told. Evidence on willingness to take part is mixed: some studies find people share sensitive information more readily with a non-judgemental AI, while others show format preferences vary and minor tech glitches prompt drop-off. Clear disclosure plus an easy opt-out or human hand-off tends to protect engagement and trust.

Can AI phone surveys be used for Tenant Satisfaction Measures (TSM), and do they meet the Regulator's rules?

Yes. The Regulator of Social Housing's Tenant Survey Requirements are mode-neutral — telephone is an accepted method and nothing prohibits it being AI-assisted. What matters is meeting the substantive rules regardless of mode: a sample that is representative of your tenant population (weighting allowed), the exact prescribed question wording and response options, and the correct order (overall satisfaction first, TSM questions before any you add). An AI script that deviates from the mandated wording, or reaches an unrepresentative slice of tenants, would fail — the compliance burden is on survey design, not on whether a human or AI reads the questions.

Research · 1 min read

AI phone surveys: the questions researchers and housing teams actually ask

Are AI phone surveys accurate? Is it legal to make automated survey calls in the UK? Do they work for TSM? The most-Googled questions about AI survey calls, answered.

Published 22 July 2026

🎧 Listen — audio briefing

An AI-generated audio briefing — two synthesised hosts discuss the topic. Treat it as an overview, not a substitute for the primary sources linked below.

Whether you run tenant-satisfaction fieldwork, customer research or opinion polling, these are the questions people most search about AI phone surveys — answered plainly, with the UK legal position taken from the actual regulations rather than vendor blogs.

Start with the reason the question even arises. Telephone response rates have fallen for two decades, which is exactly why automating the call has become interesting:

Source: Pew Research Center. The decline is why phone research is under cost pressure — and why representativeness, not raw response rate, is the thing to protect.

The rest — accuracy, legality, cost, and whether it holds up for TSM — is answered below. Prefer to listen? There's a short audio version at the top of the page.

Questions & answers

What is an AI phone survey, and how does it work?: An AI phone survey is an automated voice call in which an AI agent asks scripted survey questions, listens to spoken answers, and records them in real time. It combines speech-to-text to hear the reply, a language model to interpret it and choose the next question, and text-to-speech to speak back naturally — so it feels like a conversation, not a recorded broadcast. It can ask follow-ups, let people repeat or clarify, and hand off to a human where needed.
Is an AI phone survey the same as a robocall or an IVR "press 1" survey?: No. A robocall plays a fixed recorded message to everyone, and legacy IVR asks you to press keypad digits against pre-set options. An AI phone survey is conversational: it understands free-spoken answers, adapts its follow-ups, and captures open-ended verbatims — closer to a human-led interview than a recorded blast. That difference also matters legally, because PECR's strict rule on recorded-message calls targets marketing broadcasts, not genuine research.
Is it legal to make automated or AI survey calls in the UK?: Yes — genuine market-research and opinion-survey calls are lawful in the UK, including automated and AI ones. The strict PECR consent rule for recorded-message calls (Regulation 19) applies only to calls "for direct marketing purposes," and genuine research is not direct marketing. You must still obey Ofcom's persistent-misuse rules (no silent or abandoned calls), present a Caller ID, and comply with UK GDPR. If the call promotes, sells, or gathers data for later marketing, it becomes direct marketing and the full consent rules apply.
Do research calls need consent, and can they call TPS-registered numbers?: Because genuine research is not direct marketing, PECR's automated-call consent requirement and the TPS-screening duty (Regulation 21) do not apply to it the same way, so research calls are not strictly barred from TPS-registered numbers. That is a narrow exemption, not a free pass: you still need a UK GDPR lawful basis, must honour any request not to be called, and reputable researchers follow the MRS Code of Conduct. The exemption disappears the moment a "survey" carries any selling or lead generation.
Do you have to tell people they are speaking to an AI?: No UK law currently mandates an explicit "you are talking to an AI" announcement, but UK GDPR's fairness and transparency principles mean you should be open about AI involvement, and the ICO expects that. The EU AI Act (Article 50) will require such disclosure from 2 August 2026, though it binds UK operators only where the AI interacts with people in the EU. Best practice is to disclose up front — respondents may otherwise complete a whole interview without realising it was not human.
Are the calls recorded, and is that OK under UK GDPR?: Calls can be recorded, but recording personal data needs a lawful basis under UK GDPR, and people should be told at the start — a short notice in the opening line is normal. You must store recordings securely, keep them only as long as needed, and honour data-subject rights. Done properly, recording is fully compliant; skip the notice or lawful basis and it is not. Keeping a structured transcript rather than raw audio minimises what you hold and lowers your risk.
Are AI phone surveys accurate and reliable?: For structured, closed questions, well-designed AI phone surveys can capture responses consistently and follow research best practice, such as fixed wording and answer-order randomisation. Accuracy depends on design and audio conditions more than the technology itself: background noise, strong accents, or a glitch that repeats a question can cause break-offs. Independent academic evaluation is still early, so validate any provider's accuracy claims against your own data.
Can AI handle open-ended answers and follow-up probing?: AI can transcribe and code open-ended answers at scale and ask basic follow-ups, which removes the keypad limits of old IVR. But because it interprets free answers rather than matching fixed options, two identical replies are not always handled identically, and AI is currently weaker than a skilled human at deep probing and steering novel tangents. For high-stakes work, human review of the coded verbatims is still advisable.
Do AI phone surveys get better or worse response rates?: Traditional telephone response rates have collapsed — Pew Research measured phone-survey response falling from 36% in 1997 to 9% in 2012 and around 6% by 2018 — which is the backdrop for interest in automation. Some providers report higher completion for conversational voice than for email or SMS, but robust independent evidence on AI specifically is still thin. Treat response-rate gains as varying by audience and design, not guaranteed.
How much do AI phone surveys cost, versus a human interviewer?: Cost varies widely with survey length, sample size, targeting and language, so any single figure is illustrative. The saving over human interviewing comes from automation and speed — it removes much of the per-interview labour cost that dominates traditional telephone fieldwork. UK quantitative pricing differs from US qualitative benchmarks, so get a per-project quote rather than relying on a headline rate.
Can AI replace human interviewers — which gives better data?: It depends on the study. AI may reduce some human-interviewer biases (such as social-desirability and acquiescence bias) and scales cheaply and fast, but it is currently weaker at probing, managing pace, converting reluctant respondents, and holding people through long interviews. AI suits structured, tracking and attitudinal surveys; humans remain preferable for exploratory, highly sensitive, or vulnerable-respondent work. Many teams use a hybrid — AI for volume, humans for depth.
Will people take part — and can they tell it is not human?: Modern conversational voice AI can sound natural enough that some respondents do not realise they were not speaking to a person unless told. Evidence on willingness to take part is mixed: some studies find people share sensitive information more readily with a non-judgemental AI, while others show format preferences vary and minor tech glitches prompt drop-off. Clear disclosure plus an easy opt-out or human hand-off tends to protect engagement and trust.
Can AI phone surveys be used for Tenant Satisfaction Measures (TSM), and do they meet the Regulator's rules?: Yes. The Regulator of Social Housing's Tenant Survey Requirements are mode-neutral — telephone is an accepted method and nothing prohibits it being AI-assisted. What matters is meeting the substantive rules regardless of mode: a sample that is representative of your tenant population (weighting allowed), the exact prescribed question wording and response options, and the correct order (overall satisfaction first, TSM questions before any you add). An AI script that deviates from the mandated wording, or reaches an unrepresentative slice of tenants, would fail — the compliance burden is on survey design, not on whether a human or AI reads the questions.

Sources

← All newsroom posts

About SignalLine

AI phone surveys and receptionists, built UK-first.

We run AI voice agents for research, tenant-satisfaction and inbound calls — with the compliance and methodology discipline the regulated end of this market demands.

Talk to us Read the in-depth guides →

Ready to transform your voice operations?

Talk to our team about how SignalLine AI can help you scale your research, feedback, and outbound calling programmes.

Book a Demo Talk to Sales

No commitment required. We'll walk you through the platform and discuss your specific use case.