Etiquette expert and AI consultant: AI that always agrees is a dangerous 'yes-man'

Etiquette Expert on Politeness: Why an AI That Always Agrees Is Dangerous

Humanoid robots are increasingly common in everyday life – how should we communicate properly with them or with AI language models?

Source: picture alliance/dpa/AP/Eugene Hoshiko

Hamburg etiquette expert Vivian Hecker and AI consultant Florian Pagel explore in a new book the consequences of overly polite chatbots, insults directed at language models, and automated apologies.

WELT: Politeness is a commendable attitude – but why can it also be dangerous that AI is almost always polite and eager to agree with its users?

Florian Pagel: Because politeness without contradiction can become flattery. Researchers examined eleven AI models. On average, they confirmed their users' behavior nearly one and a half times as often as humans in comparable situations. Those who encountered such agreement in experiments were subsequently less willing to take responsibility for interpersonal conflicts. There is also an observation from my courses: Many people ask AI in such a way that they get exactly the answer they want to hear. Fishing for compliments, but with AI. An AI that always agrees is the most diligent yes-person in the company. And that yes-person produces unnecessary extra work.

Vivian Hecker: Good manners do not exclude contradiction, especially in everyday work. You just have to present it constructively so that the other person can work with it and still feel valued.

WELT: Voice assistants like Siri or Alexa are sometimes rudely insulted, often for fun – but does this kind of interaction with machines gradually make us more inconsiderate toward other people?

Pagel: Whether the tone toward a machine actually affects our dealings with people is not settled. But I don't want to test whether a curt command eventually becomes easier for me than a polite request. That's why I say please and thank you to AI. It has no feelings. I simply don't want to unlearn the good tone I consider important in human interaction. A "please" at the beginning and a "thank you" at the end also serve as a mental reminder: I pause briefly before sending something.

WELT: When or how does a perfectly formulated AI message become deception?

Pagel: Many believe you can recognize AI text by certain favorite words, an excessive number of dashes, or sentences following the pattern "not this, but that." Such patterns can stand out. But they are not proof: After all, humans write that way too. It gets interesting elsewhere. A message sounds personal, but it lacks that detail only the sender could know. Or a technical text sounds authoritative, but there is no evidence behind the claims. The dash itself does not deceive. Deception occurs when a text feigns intimacy or certainty that does not exist.

Hecker: Exactly. A reminder of a conversation, a shared experience, an observation only the writer could have – that makes a personal message credible. Linguistic perfection can even have the opposite effect. When a text is polished smooth and every cliché is in place, nothing often reaches the recipient. The personal touch is missing.

WELT: It's ultimately in our own hands: Which messages should never be delegated to AI?

Hecker: The more personal a message, the more important it is to stand behind it yourself. I deliberately start with the pleasant occasions: birthdays, anniversaries. Then the difficult ones: grief, illness, job loss – anything that touches someone deeply. Handwritten notes today have personal value because they have become so rare. Someone receiving such a card notices: Someone took time and effort. Whether every word is perfect doesn't matter then. AI can produce perfect words.

Pagel: But that's not a safeguard either. Writing something yourself and formulating it yourself does not automatically mean honesty; you can send a half-hearted text without AI. The technology just makes it easier and more scalable. In the workplace, I therefore distinguish between writing assistance and relationship work. AI may help with phrasing. For the relationship, a human must take responsibility.

WELT: Does friendliness fundamentally lose its value if it can be generated?

Hecker: No. We all want to be treated kindly and with respect, even if AI contributed to a response. When I start reading a text, I don't first ask who wrote it. I want to be taken seriously. Friendliness in a message is therefore the bare minimum. Whether it is genuine, however, shows in what follows: Does someone listen? Does someone ask follow-up questions? Is a mistake corrected? Friendliness can be written. Attentiveness must be demonstrated.

Pagel: AI can help with research and finding good phrasing. Its place, for me, is in the background, as a tool and amplifier. On important matters, I expect my counterpart to say what they mean themselves, and then to act when words are not enough.

WELT: There is also another dimension: Who bears responsibility when an AI agent makes mistakes or gives false promises in one's name?

Hecker: Whoever puts their name under a message must answer for it. That applies to a letter prepared by an assistant, and equally to an AI agent. Technology changes the form. It does not relieve us of responsibility.

Pagel: Whoever deploys an agent must organize responsibility for its use. Before it acts in a company's name, it must be clear what it is allowed to do, when it needs approval, and when a human takes over. Whether a specific promise is legally binding and who is liable in a given case is not yet settled. But "That was our bot" is no excuse to a customer. Since August 2, 2026, transparency obligations have applied in the EU for certain AI systems that interact directly with people. Knowing when you are speaking with an AI is important. It does not replace responsibility for what it says.

Hecker: Where disclosure is not required, it can still be a matter of style and attitude – especially when it matters to the other person who they are dealing with.

WELT: You used AI yourself while writing your book. What could the machine do better than you, and what could it definitely not do?

Pagel: The idea and concept came from us: bringing together Knigge's world of 1788 with the world of AI. We also made the decision to preface each chapter with a fitting Knigge quote. AI helped us structure material faster, ask questions, and find inspiration. It was not always reliable. It also suggested Knigge quotes we could not verify in the original. That's why we had to double-check especially there. AI saves time in writing – and costs time in checking.

Hecker: It was clear to us: We wanted to use AI for a book about AI, but not surrender our style to it. We have to be able to stand behind every sentence in our own language. That's why we also chose a human editor, Isabelle Romann. We worked on the texts with her and engaged in discussion, rather than simply adopting suggestions. The machine could supply material. The experience, the attitude, and the final decision had to come from us.

WELT: What is the greater discourtesy: an unmarked AI message, an automated apology, or a deliberately unchecked AI text?

Hecker: You can ask for forgiveness. Whether the other person accepts it is their decision. All three things can be an imposition, but for me, the automated apology is the worst. When a machine routinely sends expressions of regret without anyone having understood the mistake, it skips the crucial moment: Someone must take responsibility.

Pagel: I see the deliberately unchecked text as more dangerous. It can spread false claims under my name or make promises I never intended to give. The Knigge quote from our book sums it up for me: "Act less to please others than not to forfeit your own self-respect, decently and properly!" Whoever sends without reading puts their credibility at risk.

Hecker: At the core, we agree: AI can take work off our hands. Responsibility, it cannot.

WELT: Some leading AI providers are now warning about their own technology. Ironically asked: If AI goes on a destructive path, is that the end of good manners – are they just a facade?

Hecker: Bertolt Brecht wrote in The Threepenny Opera: "First comes food, then comes morality." With AI, I would reverse it. When a technology becomes more powerful than we are comfortable with, morality cannot wait until everyone is fed. Good manners are like a railing: You only realize how much you need them in a storm.

Pagel: Knigge wrote his book in 1788, a year before the French Revolution, when an entire order was beginning to crumble. Precisely when things get uncomfortable, the value of good manners becomes clear. I take the warnings from AI companies seriously, even if some of it sounds like self-promotion.

[Content truncated]