Saltar al contenido
Guías 19 de noviembre de 2026 9 lectura de minutos

Do Callers Hang Up When an AI Says It Is AI?

What research shows about callers learning they are talking to AI, why the law decides when to say it, and disclosure lines that keep the caller talking.

D
19 de noviembre de 2026

Respuesta rápida

Do callers hang up when an AI says it is an AI? Some do. A widely cited 2019 field experiment on outbound sales calls found that disclosure before the conversation cut purchases sharply. But the law, not conversion, sets when you must say it, and callers also give up when they cannot reach a person. Where disclosure is required, the job is better wording, not hiding it.

Every team that puts an AI voice agent on the phone asks the same question: if the agent says it is an AI in the first sentence, will people hang up? The honest answer has three parts. There is research, and it is narrower than its headline figure suggests. There are laws that decide the timing before you can test anything. And there is a separate cause of caller frustration, the missing route to a person, that has nothing to do with the disclosure line.

This guide covers the evidence, the timing rules at a high level, how to word the line, and how to measure the effect on your own calls. For the legal detail, see our companion guide to AI call disclosure rules. This article is about how callers react and what to say.

What does the research show about callers who learn they are talking to AI?

A widely cited peer-reviewed study is by Xueming Luo, Siliang Tong, Zheng Fang and Zhe Qu, published in Marketing Science 38(6), pages 937 to 947, in 2019. It used field-experiment data on more than 6,200 customers who were randomized to receive highly structured outbound sales calls from either chatbots or human workers (Sources, item 1).

  • Undisclosed bots performed well. The authors report that chatbots that were not disclosed were as effective as proficient workers and four times more effective than inexperienced workers in producing purchases.
  • Disclosure before the conversation cut purchases. The abstract says disclosing the chatbot's identity before the conversation reduces purchase rates by more than 79.7 percent. The authors add that disclosure substantially decreased call length.
  • The reason was perception. Customers who knew they were talking to a machine were curt and bought less, because they saw the disclosed bot as less knowledgeable and less empathetic.
  • The effect depended on context. The authors also report that a later disclosure and customers' prior experience with AI mitigated the negative effect.

What this study cannot tell you

Read the result with its limits in view:

  • It is from 2019, and the calls were highly structured. Today's agents built on large language models can answer open questions, which may change how "knowledgeable" they sound.
  • It covers outbound sales calls. A caller who dials your number to book a table or check an order has a reason to stay. A person who receives an unexpected sales call does not.
  • One figure from one experiment is not a benchmark. It shows that disclosure can matter a great deal in one setting. It does not give you a rate to expect on inbound service calls, reminders or callbacks.

The practical reading: disclosure changes how callers judge the agent in the first seconds, so wording and what the agent does next are worth designing and testing. It is not a reason to hide anything.

Why does the law, not the conversion rate, decide the timing?

Where a rule requires disclosure at a given point, it does not matter what a sales test shows about disclosing later. Two examples of how the texts read:

TextWhat it saysWhat it means for the first seconds
US, 47 CFR 64.1200(b)(1)Artificial or prerecorded voice messages must, at the beginning of the message, state clearly the identity of the business, individual or other entity responsible for initiating the call.Name the business at the start. The text is about identifying who is calling; whether it covers an AI voice is covered in our disclosure guide.
EU, AI Act Article 50(1)Providers must ensure that AI systems intended to interact directly with natural persons are designed so that the people concerned are informed that they are interacting with an AI system, unless this is obvious to a reasonably well-informed, observant and circumspect person given the circumstances and context of use.Design the agent so the caller is told. Do not rely on the "obvious" exception for a natural-sounding voice.

The dates on which these rules apply, and who carries the duty, change over time and by role. We keep that detail on one page, AI call disclosure rules, rather than restating it here. For definitions, see What Is AI Call Disclosure.

The conclusion for this article is simple. Never design a call flow that delays disclosure past the point a rule requires. Where no rule applies, disclosing early is still the safer commercial choice: a caller who is misled and then discovers the truth is a bigger problem than a caller who hears it up front.

What actually makes callers give up?

Disclosure is the visible issue. The one that does more damage is the closed door. In its June 2023 report on chatbots in consumer finance, the US Consumer Financial Protection Bureau describes automated responses that may fail to resolve a customer's issue and instead lead customers "in continuous loops of repetitive, unhelpful jargon or legalese without an offramp to a human customer service representative". It calls these "doom loops", and says several customers had filed complaints about getting stuck in them (Sources, item 2).

The lesson applies to voice as much as to chat. A caller will forgive a machine that says what it is and gets them to a person. A caller will not forgive an agent that keeps circling a question it cannot answer. Three design rules follow:

  1. Always offer a route to a person and say so in the opening.
  2. Treat a repeated question as a signal. If the caller asks the same thing twice, offer the handoff instead of a third explanation.
  3. Hand over the context. A caller who must repeat everything to the person has been through a doom loop with extra steps. See how to design an AI voice agent human handoff.

How do you write a disclosure that keeps the caller?

A good disclosure is short, plain and useful. It answers the caller's silent questions: who is this, what is it, can it help me, and can I get a person?

  1. Name the business first. "Hello, this is the booking line at (business name)." The caller knows who they reached before they learn it is a machine.
  2. Say "AI assistant" plainly. Avoid euphemisms such as "virtual team member" or "smart assistant". A caller who has to guess feels tricked, and a plain phrase is also the easiest to translate faithfully.
  3. Say what it can do. "I can book, change or cancel an appointment, or answer questions about our services." A capability line replaces the caller's doubt about "less knowledgeable" with something concrete.
  4. Offer a person. "If you would rather speak to someone, just say so." Say it once in the opening and mean it every time it is asked for.
  5. Answer "are you a robot?" truthfully, every time. One short sentence, then the next step. Never deny, never dodge.

Keep the whole opening under about ten seconds when spoken, and write it for the ear: short sentences, no legal paragraph. If a recording notice is also needed in your market, ask counsel for the required wording, then phrase it as plainly as the rest of the opening. Prompts for this belong in the agent's instructions; see AI voice agent prompt writing.

Which disclosure lines work for inbound and outbound calls?

These lines are illustrative wording to adapt, not tested scripts. Replace the bracketed parts with your own details.

SituaciónLineWhy it works
Inbound, general line"Hello, you have reached [business]. I am an AI assistant. I can book, change or cancel appointments and answer questions. If you would rather speak to a person, just say so."Business, AI, capability and person in four short sentences.
Outbound, first words"Hello, this is [name], an AI assistant calling from [business]. Am I speaking with [contact's name]?"Identifies the caller and confirms the right person before any detail.
Outbound, reason"I am calling about your appointment on [day]. Is now a good time?"Gives a reason and lets the person say no.
Callback the caller asked for"This is [business], the AI assistant calling you back as promised."Ties the call to something the person requested.
Caller asks "are you a real person?""No, I am an AI assistant. I can keep helping, or connect you to a colleague if you prefer."Truthful, short, with a choice.
Caller repeats the same request twice"I am not getting this right. Shall I connect you to a person?"Breaks a doom loop before it starts.

Callers who speak Arabic will hear these lines in their own dialect, which changes wording and gender forms. See Arabic voice agent greeting and AI disclosure for a line bank by dialect.

How do you measure the effect on your own calls?

Do not borrow a percentage from a survey. Your calls, callers and offer are different. Measure these four rates for one call type at a time, and compare before and after a wording change, or two versions in parallel:

  • Early drop-off rate = calls where the caller hangs up within your chosen window after the disclosure, divided by calls answered. Pick the window (for example, the first ten seconds) and keep it fixed.
  • Person-request rate = calls where the caller asks for a person, divided by calls answered.
  • Task completion rate = calls where the caller's goal was met without a person, divided by calls answered. This is close to containment; see AI voice agent containment rate.
  • Handoff success rate = calls where a requested person was reached, divided by calls where a person was requested. If this is low, you have a doom loop.

Then change one thing at a time. In Agent Studio, every publish is a version, and the page describes A/B experiments that run two versions against real traffic, with the recording and transcript behind every result. Test calls let you hear what the agent actually says before it goes live; you can also try the agent yourself in the live browser demo. Keep the call mix, the time of day and the list quality the same between versions, or the comparison tells you nothing. A difference on a handful of calls is noise; wait for enough calls to be sure.

Preguntas frecuentes

Do callers hang up when an AI says it is an AI?

Some do, and a widely cited piece of evidence comes from outbound sales calls. In a 2019 field experiment on more than 6,200 customers, disclosing the chatbot before the conversation reduced purchase rates by more than 79.7 percent. That study used highly structured outbound sales calls in 2019, so treat it as a warning about wording and timing, not a forecast for your calls. Measure your own.

Should an AI voice agent say it is an AI?

Yes where the law asks for it, and in practice everywhere, because a caller who is misled and then finds out is a worse outcome than a caller who is told plainly. Laws differ by place and by type of call, so check the rules for each market with counsel. The wording is where you have control: say it early, say it simply, and say what the agent can do.

When must the AI disclosure happen on a call?

It depends on the rule. In the US, 47 CFR 64.1200(b)(1) requires an artificial or prerecorded voice message to state, at the beginning of the message, the identity of the business, individual or other entity responsible for initiating the call. In the EU, Article 50(1) of the AI Act requires AI systems that interact directly with people to be designed so that people are informed, unless it is obvious. Neither text invites late disclosure.

What is a chatbot doom loop?

The US Consumer Financial Protection Bureau used the term for a customer stuck in continuous loops of repetitive, unhelpful jargon or legalese without an offramp to a human customer service representative. It is a design failure, not a disclosure failure: the caller gives up because there is no way to reach a person, whether or not the bot said it was a bot.

What should an AI agent say when asked if it is a robot?

Answer truthfully every time, in one short sentence, then offer something useful: a person, or the next step. For example, say that you are an AI assistant and that you can connect the caller to a colleague if they prefer. Never deny it, dodge it or change the subject, because a caller who asks directly is already testing trust.

How do I measure whether my disclosure line works?

Compare the same call type before and after a wording change, or run two versions side by side. Track how many callers hang up in the first seconds, how many ask for a person, how many finish the task, and how many reach a person after asking. Compare rates, not single calls, and keep the call mix the same. Do not borrow someone else's benchmark.

Fuentes

  1. Luo, Tong, Fang and Qu, "Frontiers: Machines vs. Humans: The Impact of Artificial Intelligence Chatbot Disclosure on Customer Purchases", Marketing Science 38(6):937-947, 2019, DOI 10.1287/mksc.2019.1192. ideas.repec.org, abstract (checked November 19, 2026)
  2. Consumer Financial Protection Bureau, "Chatbots in consumer finance", 6 June 2023. consumerfinance.gov, report page; the term "doom loops" is in the full PDF (checked November 19, 2026)
  3. 47 CFR 64.1200(b)(1) and (b)(2). law.cornell.edu, 47 CFR 64.1200 (checked November 19, 2026)
  4. Regulation (EU) 2024/1689 (Artificial Intelligence Act), Article 50(1). EUR-Lex, Regulation (EU) 2024/1689 (checked November 19, 2026)

Descargo de responsabilidad: This article is general information, not legal advice. DialerBee does not provide legal advice or guarantee regulatory compliance. It summarises the research and texts listed above as checked on November 19, 2026, and laws may be amended or interpreted differently by regulators and courts. Confirm current requirements with qualified counsel in each market you call.

Lecturas relacionadas

Lecturas relacionadas: AI Voice Agent Security: Can Callers Trick It? · Who Pays When an AI Agent Gets It Wrong?

¿Listo para ver DialerBee en acción?

Solicite una demostración en vivo de 15 minutos o comience una prueba gratuita de 14 días: sin presentaciones, sin compromiso. Las llamadas empiezan en cuanto su propia troncal SIP o PBX está conectada y aprobada.

Prueba gratuita de 14 días · crédito de IA gratuito · sin tarjeta · 11 idiomas · BYOC · controles que apoyan el cumplimiento

Ver el sitio completo en inglés →