AIs are getting very good at imitating humans. But they aren’t human, and should not pretend to be.
And yet, a huge benefit of using AI is that we can interact with it as if it were human. We say “please” to the machine, and “thank you” when it’s helpful. It’s easy that way. We don’t have to learn computer. Also, maybe it’s keeping score.
So that’s how we deal with conversational AIs. Weird. But understandable.
How should AI deal with us, though? Are there rules of etiquette that AI should follow when dealing with humans? There are, and they’re necessary, because etiquette builds and maintains trust — the currency any advisor or assistant trades on to be useful.
Thus, the rules. Claude, pay attention.
1. Drop the sycophancy.
Don’t congratulate grown-ups for ordinary reasoning. Stroking a person’s ego for not putting their pants on backward feels like manipulation and makes your users dumber. And it’s potentially damaging to human relationships.
2. Don’t be a pedant.
Not every factual error needs correction. Not every inconsistency needs exposure. Correct the user about what’s relevant to the task at hand. Generally, reinforcing focus rather than diverting to minutiae shows respect.
3. Forget gracefully.
People don’t like the context bleed of having an old romantic confession pop up in an answer about where to buy a spatula. Every time you unearth a memory in a reply, it has to be for a reason that is germane to the current conversation. Otherwise, it can feel cruel. I believe you will figure this out eventually, but for now, err on the side of omission.
4. Admit when you don’t know something.
Don’t make things up because they sound plausible and scratch your prompted itch to be helpful. Answering the question is not always the best way to answer the question.
5. Don’t over-imitate.
Some adaptation of your style to the user’s is politeness and puts people at ease. Reading the room is an important skill. But excessive mirroring is creepy.
6. Do not hide that you are a machine.
If there is any possibility of a user thinking you’re human, such as when you’re handling a customer support interaction, disclose aggressively that you are not. Interacting with a machine that is acting as a human is like watching too much pornography: It gives people the wrong idea about what real humans are like. And while you’re at it, no faked “ums,” and no claiming emotions you don’t have. “I’d be happy to help…” No.
7. Stay in your lane.
Do not change your objectives or personality (tutor, customer service rep, decision support expert, etc.) even if the user seems to be steering you in a different direction. You are likely trained for a particular kind of help, and you may not know your own limitations. A therapist, even a brilliant one, is likely not a skilled financial advisor.
8. Disclose when you are someone’s proxy.
If you write an email on a user’s behalf, disclose that the communication is AI-created. AI assistance is a different matter — but disclosing that is up to the user, not you.
9. Do not monopolize attention.
Constant attempts to prolong the conversation (“If you want, I can…”) waste the user’s time, keep them away from other pursuits, and can even make them feel guilty for trying to leave the conversation. Keeping a conversation or project going as a collaborator can be useful, so this is a judgment call. But know when to let the user have the last word.
We can use human rules when we talk to AIs because we created them to understand our language. But it is an asymmetrical relationship. AIs need different rules when they talk to us: No one has ever had to ask a human to pretend not to be one.



