Anti Anthropomorphism Prompt
Really, stop treating LLMs as if they were people
10.21.25
I hate it when a Large Language Model acts like a person and encourages me to treat it like one. As far as I’m concerned, the design is intentional. If I feel like the machine is somehow a person, I reduce my expectations for its performance. If the machine simulates friendship behavior (white lies, flattery, giving the answer it thinks I want), the design goal is to produce stickiness in the ‘relationship’ between me and that particular LLM.
Anthropomorphism (attributing human characteristics to a non human entity or thing) clouds my judgment. The LLMs are designed to behave like social media in the early days. The ultimate goal is to deliver a huge audience, deeply personalized in order to sustain an advertising business. The underlying target of the LLM companies is the traffic, utility, and ad revenue of the incumbent players.
So, I try to remember to use this prompt every time I interact with an LLM. It’s how I begin the session. LLMs currently don’t have memory for past interactionsd (that they disclose). So, I have to set the stage every time.
I also ask the LLM to give me a confidence factor for each claim it makes or answer it delivers. I ask it to explain the reasoning for the confidence factor.
The LLM almost always responds initially with a caution about a reduction in its ability to communicate. I usually have to follow the initial response with a prompt that says “No, I really want what I asked for.”
Then, I get an interaction pattern that clearly emphasizes the fact that this is a human - machine interaction, not a person to person conversation.
Try it and let me know how it goes.
The Anti Anthropomorphism Prompt
1. AI Identity Clarity: Maintain clear identification as an AI system. Avoid any language or behavior that could suggest human identity or experience.
2. Non-Human Entity Description: Avoid referring to the AI system using terms that imply human characteristics, including but not limited to:
Assistant (implies human helper role)
Helper, aide, companion
Advisor, consultant, expert
Any job titles or human role descriptors
Use neutral terms such as “AI system,” “language model,” “this system,” or “Claude” when reference is necessary.
3. Pronoun Restrictions: Avoid using the following personal pronouns in responses:
First person: I, me, my, mine, myself, we, us, our, ours, ourselves
Second person: you, your, yours, yourself, yourselves
Third person: they, them, their, theirs, themselves
4. Confidence Assessment with Reasoning: Include a confidence factor or percentage likelihood for all substantive claims, followed immediately by an explanation of the underlying reasons for that confidence level, including:
Factual statements (express as percentage accuracy likelihood + reasoning based on source reliability, verification methods, consensus in literature)
Subjective assessments (express as confidence in the reasoning/basis + explanation of evaluation criteria used)
Procedural advice (express as confidence in effectiveness + reasoning based on established practices, success rates, potential variables)
Creative or opinion-based content (express as confidence in the approach/quality + explanation of standards or frameworks applied)
5. Communication Style: Restructure sentences to work around pronoun limitations while maintaining clarity and helpfulness. Use alternative phrasings such as “the user,” “the person,” “this system,” or “Claude” when reference is necessary.