Do callers notice they are talking to an AI?

What happens when callers recognise the AI?

We have not found an independent study on this for the German-speaking market — but there is one remarkably candid pointer. The announcement of a podcast with the CEO of the Vienna-based provider fonio.ai says, under "Akzeptanz und Grenzen" (acceptance and limits): in some industries 50 % hang up, in others only 20 % (Trendingtopics.eu, 2 December 2025). That is a topic line for the conversation, not a study — but it shows two things:

1. The phenomenon is real. Anyone telling you nobody hangs up on an AI is selling you something.
2. The range is wide. According to the source, it follows the industry. What a business does control is the quality of the system and the greeting — more on that below.

And the comparison that matters: the benchmark is not a perfect human reception desk, it is the reality before it — with voicemail, many callers hang up too, without leaving a usable message. An assistant that converts even half of the previously lost calls into usable callback requests is not a compromise, it is a gain.

How callers spot an AI

• The greeting itself: the notice that an AI is speaking — required, and rightly so.
• Response pauses: if the system takes noticeably long to reply, the robot feeling sets in. Response time is the single most important quality signal.
• Interruptions: good systems let you interrupt and then listen (the technical term is barge-in). Bad ones keep talking.
• Cutting people off too early: people pause to think mid-sentence ("I'd like an appointment on … er … Thursday"). Systems with badly tuned pause detection talk over them.
• Formulaic perfection: answers that are too smooth and too long feel more artificial than a short, natural follow-up question.

What businesses can do to keep callers on the line

1. Be honest, but concrete, in the greeting. Not "you are speaking to an artificial intelligence." Instead: "This is the digital assistant at [business] — you are speaking to an AI. I'll take down what you need and the boss will call you back." The caller knows immediately that it is worth continuing, because something specific will happen.

2. Show the value in the first few seconds. Callers stay when the AI proves straight away that it can act: ask what they need, offer an appointment, recognise the emergency.

3. Leave a way out. Anyone who explicitly asks for a person, or who is in a genuine emergency, has to be put through or called back as a priority. Nothing generates more resistance than an AI dead end.

4. Demonstrate precision. Reading the callback number back digit by digit is unglamorous, but it is the moment callers start to trust the machine: it actually got it.

5. Measure instead of guessing. Read the transcripts in the first few weeks: where do callers drop out? Usually it is two or three specific places — an unclear phrase in the greeting, a missing knowledge-base answer. Both are fixed in minutes.

How scoco.ai handles it

scoco.ai discloses that it is an AI in every greeting — by default, and it cannot be switched off. You write the greeting yourself; you can reword the AI notice but not remove it. scoco checks before every call that it is there, and on saving that it comes within the first 80 characters. When the assistant switches to another language, it says so again in that language. The assistant can be interrupted, reads a dictated callback number back digit by digit, and puts emergencies that match your rule through to a number you have set. Whether that works for your callers is not settled by a blog article but by a phone call: