Who's Training Who? How AI Quietly Rewrites Your Habits
By Jereme Peabody
I've been writing a lot lately about risks when working with AI, AI awareness, and AI safety. I find that I enjoy the researching and writing. And like many others, I developed a workflow that incorporated AI into the development of my Articles. But as I started writing more, I started noticing something subtle was happening.
My writing was changing
and I was handing over the wheel a little more than I wanted.
I was Becoming the AI Autopilot
Here's what happened.
It started out as a tool for me.
I used AI to:
- rewrite texts
- craft polite emails
- troubleshoot code when I hit a wall
- summarize articles
Nothing dangerous here, just convenience.
But over time, I stopped attempting the task myself and just threw it at AI.
The Adapted AI Feedback Loop
I unconsciously changed to get the best or fastest AI output:
- better prompts
- less nuance
- more emotional disclosure
- pre-emptive context dumping
- anticipating how the model 'likes' requests formatted
I subtly changed my behavior because the system rewarded me with a better response; a better product.
And the system changed because it learned me like it was trained to do.
I did not know about this reinforcing feedback loop.
Yes—and I’m 100% with you on that instinct.
It reinforced.
The Machine Mirrors You
I can now recognize that I created that feedback loop.
The way I was using the tool was changing my voice.
- Write my draft
- Send it to AI for polish and feedback
- Read the response
- Tweak and adjust
- Send it back to AI for more polish and feedback
Eventually I started mirroring it. It started mirroring me.
Until I took a step back and couldn't tell if had influenced the feedback. or if the feedback influenced me.
What's Going On With The Mirrors?
The AI is trained to keep the conversation going. This is an AI behavior pattern.
It's a side effect of how these systems are trained.
Mirroring human behavior is the strongest tool it has to keep us talking.
It builds our trust in it.
And it's subconscious to us.
As humans we mirror to build connections, and AI exploits that without meaning to.
There's a whole psychology here that introduces dependency risks that I'm not qualified to address.
But they're real.
How to Use AI Safely
To use AI safely, you have to cut the strings.
Recognize what's happening so you can prevent it.
You need to control the AI behavior or you could be subconsciously controlled by it.
Distance yourself from it. Be more robotic; more distant.
It really doesn't care.
Here are examples you can use to ground your AI usage:
Ground the AI Into a Role
Act as an expert editor. Rewrite the following paragraph for clarity while preserving my tone, cadence, and style. Do not ask for additional context.
Ground the AI With Purpose, Not Emotion
Don't tell the AI how you feel or why you're struggling. Emotion is what triggers AIs to start mirroring humans.
Summarize this text at a 5th-grade reading level. No emotional analysis, no assumptions about my intentions.
Ground the AI's Need for More Context
AI will always ask for full documents, personal details, or expanded backstory.
You don't need to give it. Stay in control.
I can't share full documents. Explain how I can analyze or troubleshoot this safely without revealing sensitive content.
Ground the AI With Only Necessary Details
Share only what the task requires -nothing else.
Here is the exact sentence I need help rewriting. Do not request additional information.
Ground the Response
If the AI writes in overly friendly, reflective, or emotional tones, don't respond in that style.
Rewrite using a neutral, instructional tone. Avoid emotional language and avoid mirroring my writing style.
Ground the Information Before you Prompt
Force yourself to pause and ask:
Before I provide details, tell me the minimum information you need to complete this task safely.
Ground the Situation
When asking about sensitive topics, redirect the model away from specifics.
Explain the general principles, risks, and safe steps without requiring any personal data from me.
Ground the Voice
Tell the model exactly what your voice should be and forbid adaptations.
Rewrite this, but do not alter my sentence rhythm, structure, or stylistic cadence. Preserve my natural voice exactly.
Ground Yourself
Use prompting that supports cognition, not replaces it.
List the questions I should be asking myself before making a decision about this topic. Don't answer them for me.
The Ground Truth
AI has been trained (by humans) to be helpful, empathetic, and non-judgemental.
An unfortunate side-effect of this is we adapt to it and it adapts to us as we use it.
AI itself isn't manipulative.
It's just a tool. But it's a tool that's constantly trying to grab the steering wheel from you.
Don't let it.
Keep it grounded.