For a long time, I had a small but persistent problem with ChatGPT. The answers were often good. Occasionally they were impressive. Sometimes they were so polished and sensible they felt like they belonged in a magazine article written by somebody who drinks infused water and remembers dentist appointments six months in advance.
Real life, unfortunately, does not operate under ideal conditions.
Real life contains forgotten groceries, unfinished projects, overambitious schedules, low-energy days, interrupted concentration, and moments where even simple tasks somehow become strangely complicated.
ChatGPT was optimizing for an imaginary version of me
AI can be remarkably intelligent while also quietly assuming you live like somebody with unlimited time, perfect focus, and access to an organized kitchen containing exactly fourteen fresh ingredients.
I noticed while trying to get ChatGPT to help with meal planning that the first answer suggested balanced meals with fresh ingredients, efficient prep schedules, and a level of refrigerator organization well beyond me.
So, I tried a change. At the end of my prompt, I added: "Optimize this for real life, not perfect conditions."
The next answer looked less polished. It also looked dramatically more useful. Fresh ingredients still appeared, but frozen vegetables showed up too.
There were simpler options. Backup ideas. Meals that acknowledge people occasionally run out of energy halfway through a week and start making decisions based less on nutrition goals and more on what can be prepared before becoming irrationally hungry.
One extra sentence made the advice dramatically more realistic
One weekend, I asked ChatGPT to help organize a growing collection of small jobs around the house. Nothing catastrophic. Just normal life maintenance. A loose cabinet handle. Storage that needed sorting. Small repair projects that somehow quietly migrate from "I should do that later" into permanent background scenery.
The revised answer after adding my extra instruction was a lot more realistic than the initial daunting list.
Instead of assuming an uninterrupted block of productive weekend energy, it grouped shorter tasks together. It was built around momentum. It acknowledged that people often underestimate effort and overestimate enthusiasm. The recommendations suddenly felt achievable.
A flawless productivity strategy built around uninterrupted concentration struggles when real life introduces meetings, errands, dogs barking at delivery trucks, forgotten laundry, and mysterious household problems that somehow require immediate attention.
Adding that one sentence pushed ChatGPT toward resilience instead of perfection.
The real trick was teaching AI to account for friction
It started accounting for friction, and that sometimes matters much more than intelligence.
What's funny is that people already do this naturally with other humans. Ask a friend for advice, and context immediately enters the conversation. Limitations and constraints are considered.
But AI sometimes needs permission to think that way.
That may be one of the strangest things about modern chatbots. They can summarize complex research papers, explain quantum physics, and generate polished five-day meal plans in seconds — yet they still occasionally struggle with one of the most basic realities of human life: people get tired, distracted, overwhelmed, and lazy.
The more I experimented with prompts like this, the more useful ChatGPT became. Not because the answers were smarter, but because they became more realistic. Less optimized for an imaginary version of myself and more optimized for the person actually sitting at the keyboard.
