Starch forms almost 80% of my diet, even though my country has diverse cuisines and cooking styles. Rice, garri, fufu, semolina, and pounded yam are common options.
Soup swapping changes the taste and feeds the illusion that you’re eating something different every day.
Starch itself isn’t bad. Like many things, it’s only a problem when you consume it without moderation. The body breaks it down into glucose, which can cause repeated blood sugar spikes.
In addition to refined carbs and processed foods, I could be facing long-term health issues.
This early realization encourages me to eat more responsibly. Using Gemini in expected ways has helped me discover mature meal plans.
Gemini gobbles up what you feed it
Starve it of context and you’ll get thin results
Vague prompting produces vague answers. Artificially intelligent models don’t know things like humans do.
Hence, they aren’t pulling fresh insight from some hidden sources, but summarizing the most common existing patterns.
In simple words, you’re getting a polished remix of ideas from somewhere on the internet.
If I ask Gemini “How do I eat healthily?” I’ll most likely get tips on eating more vegetables or drinking fruit smoothies.
These results are too generic to help me reach my goal, which was creating an eating system away from starch-heavy foods.
If I was going to use Gemini’s help, I would have to invite it into my lived experience. It must understand the context of my prompts and provide the most accurate answer.
User-level prompting doesn’t need to follow a strict formula. But I’ve noticed that some components are non-negotiable.
Roleplaying with Gemini is refreshing
Put it in your preferred kitchen and watch it cook
I always define a role for Gemini before giving it a serious prompt. That way, the app knows whether to approach it as a doctor, a financial adviser, a technical assistant, and more.
In this case, I asked Gemini to be my nutrition coach.
A normal person tends to give advice from the outside and focus on obvious actions. Then they provide quick fixes.
A coach, on the other hand, works from within to understand patterns and limitations. Then they provide context-based solutions, which was what I needed.
Here’s my example prompt:
Pretend to be / Act as my nutrition coach. Your goal is to help me build a realistic, sustainable eating system I can actually maintain.
Gemini was going to respond immediately, which would’ve interrupted the process of providing context.
I prefer to give myself space to explain. So, I used a safe word that told the model exactly when context-gathering ended.
Here’s my example prompt:
For now, do not ask me any questions or design my meal plan. I’m going to describe my current situation, habits, constraints, and past plans first. Only after I say the word “READY” should you begin inquiries or offer recommendations.
Next, I described the situation. Models are far better at reasoning within preset constraints than inventing them for you.
So, the quality of the answer will increase when you frame the problem accurately. Plus, hallucinations are reduced.
Here’s my example prompt:
I have a fast metabolism / My appetite has dropped over time / Most of my meals are [dominant food type] because of how local foods in [your country] are structured / I want to eat more intentionally, but [constraints] are making it hard for me to.
After these foundations were set, I told Gemini to build a timetable based on food availability, budget, my cultural eating patterns, and other factors.
Plan what’s in your pantry
You can’t get a gourmet plan on a street-food budget
Meal compositions vary widely by region.
It doesn’t matter if the generated plan is nutritionally perfect if I can’t realistically find or afford it.
For example, a 2kg bag of Kirkland organic quinoa costs around $26 currently.
It’s usually recommended as a healthier alternative to white rice because it has more protein and fiber. But I can buy a 5kg bag of rice for roughly the same price.
Considering inflation and an unstable income, it’s unfeasible for me to consistently choose quinoa over rice.
No matter how healthy the alternative is, I’ll naturally gravitate toward what’s affordable.
I don’t hesitate to flag such issues with Gemini.
I asked it to work with starches I already eat and reduce portion size or specify my muscle training period and have starch-heavy meals shifted to those days.
The bottom line is that you’ll keep iterating and readjusting your plan until you’re satisfied with most of it.
The first version won’t fully cover your budget, appetite, schedule, and access to food items.
Gemini, is at best, an ongoing collaborator instead of a one-time solution.
Here’s my example prompt:
I noticed you included [ingredient/food item] in the plan. This item isn’t affordable or easy for me to find where I live. Can you revise the plan using foods I already eat regularly, while keeping the same nutritional goal? If replacing it isn’t ideal, adjust the portion size or pairing instead without starting over.
Gemini gave me a plan that moved most of my starch intake to lunchtime. It’s when energy demand in the body is typically higher. Breakfast and dinner are lighter and more protein-forward to reduce overeating.
Dinners were mostly reserved for soups, grilled proteins, or vegetable-heavy meals that are easier to digest.
The tool also gave me rules to follow. My starch intake should not be larger than a fist size.
I must compensate for the size reduction with double the portion of soup and protein.
Also, I must eat vegetables and protein before starch.
It should slow glucose absorption and prevent the blood sugar spikes I mentioned earlier.
Let Gemini serve you
Gemini can give you an entire week of meals if you define your parameters. It’ll save you mental energy from selecting them yourself, and money.
The real win is that the tool doesn’t have to work alone. Pair it with apps you already use, like Google Keep for grocery lists, or even your grocery store’s app for tracking prices.
The entire process becomes frictionless, and it’ll stop you from cheating with takeouts or impulse buys.




