I usually wake up at 4 AM, sip a warm cup of coffee, and begin my gold analysis. It's become a daily ritual I must perform. Once, I tried using AI to help, but the results were as unclear as rain falling on dry ground until one day I built my own prompt system that made AI respond like a professional gold analyst rather than just providing generic data. Today, I'm going to share how this system actually works.
Define Clear Roles
{#sec-role-definition}
The first prompt I use is to tell the AI to take on the role of "a skilled gold analyst with 10 years of experience," not just "an AI that analyzes gold." I've tried switching to other roles like "technical expert" or "short-term investor," but the results weren't what I wanted. A clear role helps the AI use terminology appropriate to the gold industry and view factors from an expert's perspective. I learned that prompts shouldn't be too long—simply stating the role is sufficient. This part now makes up about 5% of my entire prompt.
Provide Real Data as Foundation
{#sec-real-data-input}
The most essential thing for AI is real data. I input the previous day's closing gold price, today's opening price, and today's high/low prices. I don't provide too much historical data—just today's information is enough. I once tried having AI analysis without providing real data, letting it use its own database instead. The result was that it made guesses that didn't match reality. This section accounts for about 25% of the prompt, with daily updates. If you want AI to analyze anything, you must always provide real data as a foundation, not让它 come up with things on its own.
Enforce Sequential Thinking
{#sec-thinking-sequence}
I previously encountered the problem where AI would jump straight to answers without step-by-step thinking. Now, I specify a clear thinking sequence, such as: "Step one: Analyze price trends from the specified time period. Step two: Check support and resistance levels. Step three: Evaluate external factors." This way, the AI follows the sequence without missing steps. I experimented with different thinking orders and found this one works best for gold analysis. This section makes up about 15% of the prompt but is extremely important, making the results deeper than ordinary analysis.
Specify Response Format
{#sec-response-signals}
The most crucial thing is to tell the AI to respond in the format we want. I ask it to divide the response into 4 parts: 1) Price trend 2) Risk level 3) Trading approach 4) Key factors to watch. I've encountered AI that responded in abbreviations or long, hard-to-read text. Now I force it to use tables or organized formats. This section accounts for about 30% of the prompt but delivers the clearest results. I learned that specifying the response format must be explicit—not just "respond as an analyst" but clearly stating what format you want.
Fill-in Word Checkpoint
{#sec-prompt-check}
The biggest problem I found was that AI often added unnecessary information or filler words like "um" and "ah" that made the results disorganized. So I added a final checkpoint in the prompt: "Do not use words like 'um,' 'ah,' or unnecessary fillers." This section makes up about 5% of the prompt but is highly effective, making the results clean and easy to understand. I once forgot to add this part, and the AI included unnecessary information, teaching me how important it is to set clear guidelines.
Optimal Prompt Ratio
{#sec-optimal-prompt-ratio}
After many experiments, I found that the optimal prompt ratio is 25% constant and 75% daily data. The constant part includes role definition, thinking sequence, response format, and fill-in word checking. The part that changes daily is the real price data. This makes the system flexible while maintaining a clear structure. I tried other ratios like 30% constant and 70% daily data, but the results weren't as good. Therefore, 25:75 is the optimal ratio for my gold analysis system.
Lessons and Next Steps
{#sec-lessons-next-steps}
Building a practical prompt system isn't easy. I had to try and fail many times to arrive at this system. What I've learned is: defining clear roles, providing real data, enforcing sequential thinking, specifying response formats, and checking for unnecessary fillers—all these help AI work exceptionally well. Next, I'm working to expand the system to support analyzing other assets like cryptocurrencies or stocks using the same prompt structure but adapted to each market. Using AI for investment analysis doesn't mean we can abandon our own thinking, but rather it's a tool that enhances the effectiveness of our decision-making.
For those who want to try real gold trading, open an XM account at: open an XM account via our partner
