TeleTalker records calls, generates transcripts, and speaks with an artificial voice. Each of those creates legal and ethical duties around disclosure, consent, and personal data. This guide is the one every operator should read before going live.
When TeleTalker answers or places a call, it captures a person's voice, what they say, their phone number, and often their name, appointment details, or account information. In most legal systems, that combination is personal data, and recording it is processing of that data.
Processing personal data carries three baseline obligations that run through everything else in this guide:
Tell people it's an AI, tell them it's recorded, have a reason to record, keep the data only as long as you need it, and be ready to hand it over or delete it on request. The rest of this guide shows how TeleTalker helps you do each of those.
The single most effective compliance step is an automatic voice prompt at the start of every call that tells the caller they've reached an AI and that the call may be recorded. It satisfies AI-disclosure and recording-notice duties in one sentence, in the moment it matters, before any personal data is exchanged.
Set this as your agent's default opening line. TeleTalker plays it automatically on inbound answer and at the start of outbound calls:
It names the AI ("handled by an AI assistant"), discloses recording ("may be recorded"), states a purpose ("for service quality"), and stays warm and short so it doesn't feel like a legal warning. Adapt the purpose to match what's true for you — if you record for training, say training.
Because TeleTalker answers in 30+ languages, the greeting should play in the language the agent will speak. Keep a translated version of the same sentence for each language you serve, and make sure the disclosure and recording notice survive translation — don't drop them to save words.
Never let an agent begin taking bookings or answering questions before the notice has played. If you shorten the greeting for style, the AI-disclosure and recording clause are the two parts that must stay.
A voice this natural is exactly why disclosure matters. When callers can't tell they're speaking to software, letting them assume they've reached a human is deceptive, and in a growing number of places it's unlawful. Disclose clearly and early.
In your agent instructions, add a boundary such as: "If the caller asks whether you are a human or an AI, clearly confirm you are an AI assistant. Never claim to be a person."
TeleTalker records calls and stores audio, and the transcripts generated from those calls may also be processed by ElevenLabs. Callers need to know recording is happening, and in some places you need their agreement to it. The spoken notice covers the "knowing"; consent rules vary (see per-country notes below).
Jurisdictions fall into two broad camps for recording a conversation:
| Model | What it means | Practical effect |
|---|---|---|
| One-party | Only one participant needs to be aware/consent. | Your notice is generally enough; recording can continue. |
| All-party (two-party) | Every participant must be informed and agree. | Play the notice, and give the caller a real chance to object or hang up before recording is used. |
For all-party regions, configure the agent so that after the greeting, a caller who objects to recording is offered an alternative (a callback from a human, or continuing without storing audio if your configuration allows). Log that choice.
When TeleTalker dials out, you're initiating contact. Make sure you have a lawful reason to call that person, that the recording notice plays before anything else, and that you respect do-not-call and calling-hour rules in their region.
Be accurate with your callers and honest with yourself about where call data actually goes. TeleTalker keeps recordings and transcripts on the device, and the voice engine provider processes call audio and transcripts in the cloud to generate speech and text.
| Data | Where it lives | Notes |
|---|---|---|
| Call audio recording | On the device by default | Stored so you can review calls; treat as personal data. |
| Transcripts & structured notes | On the device; also processed by ElevenLabs | Text of what was said; may be used by the provider per their terms, including for service operation and, depending on settings, model training. |
| Phone number & contact name | On the device / your logs | Directly identifies a person. |
| Anything the caller shares | In the transcript | Could include health, financial, or other sensitive details — minimize what you ask for. |
Don't tell callers their data "never leaves the device." That isn't accurate for an AI voice service: audio and transcripts are processed by ElevenLabs to make the assistant work. State plainly that a third-party AI voice provider processes call audio and transcripts, and link to that provider's privacy terms.
The less you collect, the less you have to protect. Configure agents to ask only for what the task needs. A booking needs a name, number, and time — not a date of birth or a card number spoken aloud. If you don't need it, don't capture it.
"We keep everything forever" is not a retention policy — it's a liability. Decide how long recordings and transcripts stay, write it down, and enforce it. A short, defensible retention window reduces risk and respects callers.
If you keep recordings for service quality, they don't need to live for years. If a law in your sector requires a specific retention period, follow that — and delete when it expires.
People can ask for a copy of the data you hold about them and ask you to erase it. Build a simple, repeatable way to respond, and be ready to act on both the device and with your voice provider.
Publish a contact (email is fine) where callers can request access or deletion, and commit to a response time. A monitored inbox plus the steps above is enough for most small operators.
A privacy policy is where you write down, in public, what the spoken notice says in brief. Host it somewhere stable (your website, or a page under your teletalker.app presence) and reference it if a caller wants detail.
Plain language beats legalese. Callers are more likely to trust a short, honest policy than a wall of boilerplate. Update it whenever your recording, retention, or provider setup changes.
This is orientation, not legal advice. Recording and AI-disclosure rules differ by country and sometimes by state or province. Check your local rules before you launch, especially if you record or call across regions.
Under GDPR, call audio and transcripts are personal data. You need a lawful basis, clear notice, retention limits, and must honor access/erasure rights. AI-transparency expectations are rising — disclose the AI clearly.
Notice + lawful basisUK GDPR and the ICO's guidance mirror the EU: inform callers, limit retention, and support data-subject rights. Recording notice at the start of the call is standard practice.
Notice + lawful basisRecording consent is set state by state. Many are one-party, but several (including California, Florida, and others) require all-party consent. Play the notice everywhere, and treat calls as all-party when in doubt.
Check state — often all-partyPIPEDA treats call recordings as personal information: identify your purpose, get appropriate consent, and limit collection. Federal telemarketing rules also apply to outbound calling.
Purpose + consentData-protection laws are newer and vary by emirate and free zone. Disclosure of recording and a clear purpose are expected; verify local telecom rules before outbound campaigns.
Disclose + verify locallyDefault to the strictest habit: play the AI-and-recording notice, keep a lawful reason, minimize what you collect, set a retention limit, and support deletion. It travels well.
Strictest-habit defaultIf your caller and your device are in different countries, more than one set of rules can apply. For outbound campaigns across regions, confirm consent and calling-hour rules for the caller's location, not just yours.
Some earlier guides and FAQs contained figures and privacy claims that don't match the accurate product facts. Use these corrected values everywhere, and prefer them if you see a conflict.
| Topic | Old / inconsistent | Correct |
|---|---|---|
| Call rate | ~$0.85/min; $0.45/mo | $0.05 / minute, pay-as-you-go |
| Credit value | Varied | $10 = 200 minutes · credits never expire |
| Privacy framing | "Audio never leaves device / not in cloud" | Calls are recorded; audio + transcripts are processed by ElevenLabs |
| Recording | Implied minimal/none | Calls are recorded and transcribed by default |
Telling callers their data stays entirely on-device when it is in fact processed by a third-party provider is itself a transparency failure. Accurate disclosure is the compliant choice — and the honest one.