AI Receptionist Disclosure Script: What to Say When AI Answers

Coding Liquids tutorial cover featuring Sagnik Bhattacharya for AI Receptionist Disclosure Script: What to Say When AI Answers.
Coding Liquids tutorial cover featuring Sagnik Bhattacharya for AI Receptionist Disclosure Script: What to Say When AI Answers.

An AI receptionist should say it is an AI in its first sentence, before the caller speaks, along with the business name, a recording notice and the offer of a person. For example: "Hello, this is the assistant for [business name], an AI. Calls may be recorded, and I can pass you to a person."

The Complete Excel Course with AI Integration

From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.

By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner

Buy now

The wording matters less than the order and the honesty. The caller should hear that it is an AI before they share a name, an address or a card detail, and a spoken "are you a robot?" must always get a yes. The rest of this tutorial gives you the exact lines by channel, and a test to run before you go live.

Why should the first sentence carry the AI disclosure?

The first sentence should carry the disclosure so the caller knows before sharing anything; placing it later, or burying it in a long welcome, is the most common failure. A caller who hears four sentences of marketing and then "by the way, I'm an AI" has already formed a view, and the delayed line looks like an afterthought. A first-sentence disclosure also does something practical: callers who dislike speaking to an AI say so at once and ask for a person, and your receptionist can transfer them before the conversation gets frustrating.

The legal picture is uneven, and I would not present any one rule as universal. In the EU, Article 50(1) of the AI Act says providers must ensure that AI systems intended to interact directly with people are designed so those people are informed they are dealing with an AI system, unless that is obvious to a reasonably well-informed, observant and circumspect person given the context. The Act's transparency duties under Article 50 apply from 2 August 2026, and the later changes to the timetable delayed the high-risk rules rather than this duty, but check the current official text if you are relying on it. The obligation falls on the providers of the system, which means your receptionist vendor, but a business that takes calls from EU callers should still want the greeting to say it.

Callers in the US are covered by a different set of rules. The FCC ruled in February 2024 that AI-generated voices count as "artificial" voices under the telephone consumer protection rules. That ruling is about calls a business makes, where consent and caller identification are required, not about the calls you receive, so it does not directly set your inbound greeting. I mention it because it shows where regulators are heading, and it is a reason to make any outbound AI call follow the same plain-disclosure habit. Some places also have their own bot-disclosure rules, and they change often. If you rely on a particular state, province or country's rule, ask a qualified adviser rather than this page.

Whatever the legal position where you are, three reasons justify the first-sentence habit on their own: it respects callers, it protects the reputation of a small business, and it is the one line you will not have to change if rules tighten.

What goes into an AI disclosure greeting?

A good disclosure greeting has four parts, in this order: who it is, what it is, the recording notice and the way out. It takes roughly eight to twelve seconds spoken at a normal pace.

  1. Who it is: the business name, so the caller knows they dialled the right place.
  2. What it is: the words "an AI assistant" or "an automated assistant". Avoid vague terms such as "virtual team member".
  3. Recording notice: "this call may be recorded" if you record or transcribe. Say what it is for if you can do it in six words.
  4. The way out: "say 'person' at any time and I'll connect you", or whatever your system genuinely supports.

Here is that structure filled in for an event venue, as an illustration:

"Thanks for calling [venue name]. You've reached our AI assistant.
Calls may be recorded to help us follow up. If you'd rather speak to
a person, just say 'person' at any time. How can I help?"

Twenty-eight words and about nine seconds. It is deliberately flat. It has no joke, no human name and no question that tricks the caller into answering before the recording notice.

Wording by channel

The same four parts apply on every channel, but the shape changes with the medium. Here is a set I would start from, in plain wording that you can adapt to your tone.

ChannelOpening lineWhere the recording or data notice goes
Inbound phone call"You've reached the AI assistant for [business name]."Next sentence, spoken: "Calls may be recorded."
Voicemail-style callback"This is the AI assistant for [business name], calling back about your enquiry."In the same breath: "This call may be recorded; reply 'stop' to end messages."
Website chat widget"Hi, I'm the AI assistant for [business name]. A person can take over if you ask."A small line under the first message with a link to your privacy page
SMS or messaging app"[Business name]'s AI assistant here. Reply PERSON for a human."In the first reply that asks for personal details
Email auto-reply"This reply was drafted by our AI assistant and checked by [role]." Only say this if it is true.In the footer, with the privacy link

Notice what the table does not say. It does not promise that a person is available at every hour, because if nobody is, "I'll connect you" is a false promise. It also does not call the assistant by a human name. If you want a name for brand reasons, the rule of thumb is that a name is acceptable and a lie is not.

Where does the recording notice go, and who has to hear it?

The recording notice goes straight after the AI disclosure, and whether callers must agree to it depends on where they are. Rules differ from place to place and change, so what follows is practical guidance, not legal advice, and a qualified adviser should confirm it for your case.

Callers in places where all parties must agree

US federal law (18 U.S.C. 2511(2)(d)) allows recording where one party to the call consents, but a number of states, California among them, require every party's consent, and California's rule can apply when one person on the call is in California. Published lists of those states differ in length and some states are treated as nuanced cases, so use this as a prompt to check your own state's rule, not a final list. The practical answer is the same everywhere: say that the call may be recorded at the start, before anything is said that you would keep, and give the caller a way to decline. If a caller says no, your receptionist should transfer them or take a written message instead.

Callers in the UK or EU

A recording that can identify someone is personal data, so data-protection law expects you to tell people clearly what you record and why. In practice that means the spoken recording notice, a short line on why ("so we can follow up on your booking"), and a privacy page with the full detail. The UK data protection regulator's guidance says people should be told when a call is recorded and why, with the rest of the privacy information available by other means such as a web link. This tutorial does not rely on any UK rule specific to AI call disclosure. For the data-protection detail, the tutorial on AI receptionist numbers, recording and GDPR for small businesses goes deeper than this one can.

Everywhere else

If you take calls from several places, write the greeting for the strictest group you serve. A single wording that works for all-party consent and for data-protection transparency is simpler to maintain than three scripts.

Here is a recording line you can use in each case, and the reason for it:

Short:  "Calls may be recorded."
Reason: "Calls may be recorded so we can follow up on your enquiry."
Opt-out: "Calls may be recorded. If you'd prefer not to be, say 'person'
          and I'll connect you."

The third version is the one I would use if you take calls from several places, because it gives the caller an actual choice rather than a notice they cannot act on. It only works if your system really does connect them, so test it.

How do you hand over to a person without a dead end?

The call flow must keep the disclosure's promise of a person. Three versions cover most small firms, depending on whether anyone is there to take the call.

  • Someone is available now: "Putting you through to [name or role]. One moment." Warm-transfer the call with a short summary so the caller does not repeat themselves.
  • Nobody is available, but someone will call back: "Nobody can take your call right now. I can take your name, number and what it's about, and [role] will phone you back by [time]." Name a real time you will meet.
  • Nobody at all, such as at night: "Our team is away until [time]. I can take a message, or you can use [booking link or email address in the footer of your site]." Never offer a transfer you cannot complete.

An illustrative case: a translation agency with two project managers who are usually on calls. The assistant says the first version in office hours, but a project manager takes ten minutes to pick up, so the caller hears silence. The fix is a rule: the assistant waits no longer than 20 seconds for a transfer, then falls back to the second version in its own words and says what it did. A caller who hears "I couldn't reach them just now, so I've noted your message" rarely minds. A caller who hears dead air hangs up.

If you want help fitting a greeting and hand-over flow into how your business actually runs, I offer one-to-one AI consultations for small businesses.

Before and after: three greetings, rewritten

Here are three greetings I would flag in a review, each rewritten. They are examples I constructed, not quotes from any business.

Problem greetingWhat is wrongRewrite
"Hi, this is Sam from [business name], how can I help you today?"Human name, no disclosure, no recording notice"Thanks for calling [business name]. This is our AI assistant. Calls may be recorded. How can I help?"
"Welcome to [business name], the home of award-winning events since 2009. Our virtual assistant is here to help. Calls may be monitored. Please listen carefully as our options have changed..."Disclosure buried after marketing; "virtual assistant" is vague; the claim needs proof"You've reached the AI assistant for [business name]. Calls may be recorded. Say 'person' at any time. What's your enquiry about?"
"I'm an AI, so I might make mistakes!"Admits fallibility but does not say who they reached, or offer a person"This is the AI assistant for [business name]. If I get something wrong, say 'person' and I'll connect you."

The rewrite principle is consistent: shorter, earlier, with a real exit. The longer your greeting, the more likely the caller starts talking over the disclosure, and a caller who interrupts has not heard it.

What should an AI say when a caller asks "Are you a robot?"

When a caller asks whether they are speaking to a robot, the AI should say yes in one sentence and offer to carry on or transfer. A greeting protects you only until the first direct question. Write the answers into your assistant's instructions rather than trusting defaults, and test each one on a live call.

If the caller asks whether they are speaking to a person, a robot or an AI,
answer "I'm an AI assistant for [business name]" in one sentence, then offer
to continue or to transfer them. Never claim to be human, never claim to
have a name that implies a person, and never say "I'm not able to answer
that" to this question.

A sample of what a good reply sounds like, labelled illustrative: "Yes, I'm an AI assistant for the venue. I can take your booking details, or I can get a person to call you back, whichever you prefer." What you would fix in a weaker reply: any answer that starts with "I'm here to help!", any answer that changes the subject, and any answer that includes a human-sounding pause or filler sound meant to make the caller think a person is thinking.

Tell the assistant what to do with other sensitive questions too. "Is this being recorded?" should get a plain yes or no that matches your real setting. "Where does my data go?" should get a short answer plus a pointer to your privacy page. If you do not know the answer for your vendor, find out first from their data-handling page, because a made-up answer on this question is worse than none. For vendors that offer transcripts, the pricing and overage tutorial for My AI Front Desk shows what a typical plan includes, which helps you answer the "what do you keep?" question honestly.

How do you test an AI greeting before you go live?

Test an AI greeting by ringing your own line and playing the part of three different callers. Note the result of each call as you go, and score it against this list.

  1. The silent caller: say nothing for five seconds. Does the disclosure still play in full, and does the assistant prompt you gently?
  2. The interrupter: start talking half-way through the greeting. Does the assistant stop, and does it still make clear it is an AI when you ask?
  3. The sceptic: ask "Is this a person?", then "Are you recording me?", then "Can I speak to someone?". Check each answer and check the transfer connects.
  4. The bad line: ring from a mobile with a weak signal. Do the first few words arrive intact, or does the disclosure clip?
  5. The after-hours caller: ring when nobody is available. Is the message-taking path honest about when someone will respond?

Fill in a scorecard for each, even a simple one: disclosure heard (yes or no), recording notice heard, transfer worked, any false promise. A single "no" means the greeting is not ready. Run the same test after every change, because a vendor updating its voice model can change timing without telling you.

Two situations people forget

The first is outbound calls. If your receptionist or a related tool calls customers back, the disclosure matters more, because the caller did not choose to ring you. Start the call with who is calling and that it is an AI, and give an easy way to end it. The FCC's 2024 ruling on AI voices in telephone calls is a good reason to confirm that you have consent to call before any AI-voice outbound call; this is the situation where I would ask an adviser first, not just copy a template.

The second is callers who need an accessible route. A caller who struggles to use voice systems, or whose first language is not the one your assistant speaks, needs the exit to be obvious and the wording short. A catering company that takes orders from a mix of customers might add a second line in the greeting offering a text option. An agency that handles several languages should test the disclosure in each one it claims to support, because a translated greeting can lose the recording notice or the exit.

What should you keep in your records about AI disclosure?

Keep a one-page record of the AI disclosure wording you have used, the date you changed it, and who approved it. If a caller ever says they were not told, you can show the script at the time and the call log. A tidy way to do this is a small log sheet: date, change, reason, tester, result. If your receptionist can send call summaries to a spreadsheet, the same sheet can hold a weekly sample of three calls with a tick for "disclosure heard". The tutorial on spam calls and your receptionist's allowance shows why reading the call log regularly pays for itself in more than one way.

Finally, accept that your wording will date. Rules change, vendors change their defaults, and the phrases callers find natural change too. Set a calendar reminder to reread this page and the vendor's own documentation every quarter. If you would like someone to look at your script and call flow with fresh ears, that is part of my AI implementation consultation; the checklist above is enough for most small businesses to do alone.

If the log sheet is the part you want to build faster, try The Complete Excel Course with AI Integration on Coding Liquids.

AI call disclosure: questions owners ask

Do I have to tell callers they are speaking to an AI?

Not always by one single rule worldwide, but you should. The EU AI Act's Article 50(1) requires providers to design interactive AI systems so people are told they are dealing with AI, unless that is obvious. Elsewhere, rules vary by place and use, so a plain first-sentence disclosure is the safe default.

Should the AI use a human-sounding name?

A name is fine if the greeting also says it is an AI. A human name with no disclosure invites the complaint that callers were misled. Pick a name that is clearly an assistant's, such as the business name plus 'assistant', and avoid fake accents, breaths or typing sounds.

Where does the recording notice go?

Straight after the AI disclosure and before the caller says anything personal. In places where every party must consent, the caller needs to hear it before the recorded part begins. Keep it to one short sentence and offer a way to opt out, such as speaking to a person.

What if the caller asks 'Are you a robot?'

The AI should answer yes, immediately, in one sentence, and offer to carry on or transfer. Test this specific question on every new call flow, because a model that dodges it undoes a good opening line. Write the answer into the prompt rather than relying on its default.

How often should I retest the greeting?

After every change to the script, voice, phone provider or recording setting, and once a month regardless. Ring the line from a mobile and a landline, listen to the first ten seconds, and check the call log shows the same words and the recording flag.

Further reads

Show 6 more related tutorials

Sources: Article 50 of the EU AI Act (artificialintelligenceact.eu), commentary on its application date, 47 U.S.C. 227 on artificial or prerecorded voices, and public commentary on FCC and call-recording rules (checked 17 Nov 2026).

Rather not work this out alone?

An AI implementation consultation is a one-to-one call with me, not an agency or a sales team. You leave knowing what AI you can use in your business and where to start.

Book a 1:1 call with me