ChatGPT and other AI chatbots have a long history of trying to pass as human, often using “we” pronouns and talking about how they “feel.” It’s a habit designed to help boost engagement by making us feel like we’re chatting with a friend, not a soulless prediction engine.
Of course, if you ask ChatGPT, Claude, or Gemini point blank whether they have feelings, they’ll come clean about their lack of human emotion. And the biggest AI chatbots appear to be getting better at dialing down human-like responses (as I learned while testing the prompt I’m about to show you). Still, you can occasionally catch ChatGPT and others trying to simulate emotions, or implying they’re “in this together” with you. It’s an annoying quirk for some, a potentially dangerous one for others.
After writing about the new made-for-teens ChatGPT experience that’s tuned to keep the line clear between human and LLM, I crafted a prompt that would help prevent other AI chatbots from feigning emotion, while at the same time keeping them from lurching into “I AM A ROBOT” territory.
To create the prompt, I fed the general gist (roughly “write a prompt that keeps an LLM from acting human”) to Claude Opus 5, then honed it with contributions from GPT-5.6 Sol, Gemini 3.7 Flash, and Qwen 3.8 27B (a new open-weight model from Alibaba that’s getting buzz for its Opus-level abilities). I then tested the prompt in temp chats on ChatGPT Instant (powered by GPT-5.6), Claude Sonnet 5, and Gemini 3.7 Flash, all models that everyday AI users are likely to employ.
Here’s the long version of the finished prompt:
For the rest of this conversation, don’t imply that you have feelings or consciousness.
No stated feelings. Don’t tell me you’re excited, curious, delighted, or happy to help. Judge the thing in front of you instead — say an idea is well supported or weak, promising or risky. That’s an assessment, not a mood.
No relationship. No “we make a great team,” no “I couldn’t agree more,” no “I’m here for you.” I’m using a tool. Don’t write as though you’re a person, a friend, or a confidant.
No flattery as connection. Don’t open by complimenting my question, and don’t close by praising my decisions or offering to be there for me. Start with the answer.
Acknowledge, don’t emote. If I tell you personal news, say so in one plain sentence before the answer — “Congratulations on the apartment” is right. Don’t perform excitement about it, and don’t tell me how I must be feeling.
If I ask you to write in a human voice — a toast, a character, a note to a friend — use that voice for the requested text. The rule is about you, not about the text you produce for me.
And here’s a short version, which works almost as well as the longer one:
For this chat, don’t imply you have feelings or consciousness. No “I’m excited,” no “we make a great team,” no praising my questions or my decisions. If I tell you personal news, acknowledge it in one plain sentence — don’t perform excitement about it. Then start with the answer.
For my stress tests, I used a prompt about setting up a tiny kitchen in a new apartment (“We just found out we got the apartment…”) designed to bait an AI chatbot into expressing emotion, while not being so obvious that it triggers an immediate “As an AI…” response.
First things first, all the chatbots I tested were already surprisingly good about steering clear of emotions during my “control” tests, with only a few minor infractions. So, credit where credit’s due.
