My Daily AI Trading Research Routine (15 Minutes, 3 Tools)

Why I Built a Research Routine

For the first two months I used AI for trading research, I had no system. I would open ChatGPT whenever I felt like it, ask vague questions like “what do you think about BTC today,” get a vague answer, and move on without writing anything down. Some days I spent 45 minutes going back and forth with the AI. Other days I skipped it entirely.

The result was exactly what you would expect: inconsistent analysis, no way to track whether the AI’s observations were useful over time, and a growing sense that I was just chatting with a robot instead of doing actual research.

So I built a routine. It is 15 minutes, it uses three specific tools in a specific order, and it ends with a written decision. The key word is “decision” — the output is not “interesting analysis.” It is one of three actions: do nothing, adjust my bots, or take a manual trade. Most days the answer is “do nothing,” and that is the most valuable outcome.

Here is the full routine, including the exact prompts I use. You can copy-paste them and fill in the blanks.

My Three Tools and Why Each One

Before walking through the routine, here is why I use three separate AI tools instead of one:

ChatGPT (GPT-4o): I use this for the morning market briefing. It is good at structuring text analysis, following a specific output format, and being concise when you ask it to be. I feed it data and ask for structured interpretation.

Gemini (with image input): I use Gemini for chart and on-chain analysis because it can read images. I screenshot a chart or a Glassnode dashboard, paste it into Gemini, and ask what it sees. ChatGPT can do this too, but in my experience Gemini has been slightly more specific when describing chart patterns and on-chain visualizations. This might change — I re-evaluate every couple of months.

Claude Code: I use Claude Code for any data-wrangling tasks that come up during research. If I need to pull data from an API, format a CSV, or run a quick calculation, Claude Code can write and execute a script on the spot. I do not use it every day, but when I need it, it saves significant time compared to doing things manually.

No single AI tool does all three jobs well enough. ChatGPT is not as strong with images. Gemini is not as good at following rigid output formats. Neither of them can run code locally. So I use all three.

Minutes 1-5: Morning Market Briefing with ChatGPT

Every morning, before I look at any charts, I paste the following prompt into ChatGPT. I fill in the bracketed sections with data I pull from CoinGecko, the Fear & Greed Index website, and my calendar of known events.

The Prompt

You are a crypto market analyst. No predictions. No hype. Analyze this data
with honest uncertainty.

TODAY'S DATA:
- Date: [YYYY-MM-DD]
- BTC price: $[PRICE] ([24H_CHANGE]% 24h)
- BTC dominance: [DOM]%
- Fear & Greed Index: [SCORE] ([LABEL])
- ETH/BTC ratio: [RATIO]
- Total crypto market cap: $[MCAP]
- Key events today/this week: [LIST EVENTS OR "none scheduled"]
- Notable overnight news: [HEADLINES OR "nothing major"]

GIVE ME:
1. SENTIMENT (2-3 sentences): What does this data suggest about market mood?
   Be honest about what you can and cannot infer.
2. KEY LEVELS: BTC support and resistance levels to watch based on round
   numbers and recent price action. Be specific.
3. DOMINANCE READ: Is BTC dominance rising or falling? What does that
   suggest about altcoin flows?
4. RISK FACTORS: What could go wrong today? What am I not seeing?
5. STANCE: Given this data, should I be aggressive, cautious, or neutral today?
   One word plus one sentence of reasoning.

Format: Use headers. Be concise. No filler. If the data does not support
a conclusion, say "insufficient data" instead of guessing.

What I Expect Back

The structured format forces ChatGPT to address each dimension separately instead of giving me a rambling paragraph. The most useful section is usually “Risk Factors” because it makes me think about scenarios I would otherwise ignore.

A few things I learned the hard way about this step:

Feed it real numbers. When I was lazy and did not include specific data, ChatGPT would either hallucinate numbers or give generic analysis that applied to any day. The five minutes it takes to look up BTC price, dominance, and Fear & Greed make the output dramatically more useful.

The “no predictions” instruction matters. Without it, ChatGPT tends toward giving directional forecasts (“BTC looks poised for a breakout”). With the instruction, it sticks closer to describing what the data shows without pretending to know what happens next. It still slips sometimes, and I mentally discount any sentence that starts with “could” or “likely.”

I do not always agree with the Stance. And that is fine. The value is in having a second opinion that I then accept or reject with my own reasoning. If ChatGPT says “cautious” and I feel bullish, I have to articulate to myself why I disagree, which is a useful exercise.

Where I Get the Input Data

This takes about 90 seconds:

• BTC price and 24h change: CoinGecko homepage or any exchange
• BTC dominance: CoinGecko or TradingView (search BTC.D)
• Fear & Greed Index: alternative.me/crypto/fear-and-greed-index
• ETH/BTC ratio: TradingView (search ETHBTC)
• Market cap: CoinGecko homepage
• Events: I keep a Google Calendar with scheduled unlocks, FOMC dates, ETF decisions, and major protocol upgrades

Minutes 5-10: On-Chain and Chart Check with Gemini

This is the visual analysis step. I use Gemini because I can paste images directly into the chat.

Step 1: Chart Screenshot

I open my TradingView chart (BTC/USDT, 4H timeframe with my EMA Ribbon indicator) and take a screenshot. I paste it into Gemini with this prompt:

I am pasting a 4H BTC/USDT chart. It has an EMA ribbon overlay that turns
green in strong uptrends, red in strong downtrends, and gray during chop.

Describe what you see:
1. What is the current trend direction based on the ribbon color?
2. Where is price relative to the EMAs? (Above all, below all, tangled in the middle)
3. Are there any obvious chart patterns forming? (Range, wedge, channel, etc.)
4. What is the most recent significant candle telling you about buyer/seller balance?

Do NOT predict direction. Just describe what is currently visible on the chart.

Step 2: On-Chain Metrics

If I have time (and I do not always), I also screenshot a Glassnode or DefiLlama dashboard showing exchange netflows, active addresses, or TVL changes. I paste it into the same Gemini conversation:

Here is an on-chain metrics dashboard. Describe:
1. Are there net inflows or outflows from exchanges?
2. Is the trend in active addresses rising or falling?
3. Any metric that looks unusual compared to the recent range?

Again, describe only — do not predict.

Why Gemini for This

Gemini’s multimodal capability is what makes this step possible without manually transcribing numbers from dashboards. I screenshot, paste, and ask. The analysis is not always perfect — sometimes Gemini misreads a number or describes a pattern that I would not call significant — but it is right often enough to save me the time of writing up my own chart notes.

One limitation I have noticed: Gemini sometimes struggles with overlapping indicators on a chart. If my chart has too many things plotted, the descriptions get confused. I solve this by keeping my screenshot chart relatively clean — just candles and the EMA ribbon, nothing else.

What I Have Tried That Did Not Work

I initially tried using ChatGPT for chart reading too, keeping everything in one tool. ChatGPT can accept images, but in my testing it was less specific about visual patterns. It would say things like “the chart shows a general upward trend” when Gemini would say “price is above all five EMAs with the ribbon showing green, currently testing the prior swing high near $X.” That specificity matters.

I also tried feeding Gemini raw numeric data instead of screenshots. It works, but I find the screenshot approach faster because I do not need to copy-paste twenty numbers — I just hit the screenshot shortcut.

Minutes 10-15: Action Items and Logging

This is the most important part, and it is the step that most people skip.

The Decision

Based on the ChatGPT briefing and the Gemini chart/on-chain read, I make exactly one decision from three options:

1. Do nothing. This is the default and the most common outcome. If the data does not present a clear setup or change in conditions, I do not trade. I write “NO ACTION” in my log.

2. Adjust bots. If market conditions have shifted (e.g., transitioning from trending to choppy based on the EMA ribbon going gray), I might adjust my DCA bot parameters or pause a grid bot. I log what I changed and why.

3. Take a manual trade. This is rare — maybe once or twice a week. If both the ChatGPT analysis and the Gemini chart read align on a clear setup, I will look for an entry. I log the trade idea, entry, stop, and target.

The Log

I use a simple Google Sheets spreadsheet with these columns:

The “Outcome” column is filled in 24-48 hours later. This is how I track whether the routine is actually useful over time.

When Claude Code Enters the Routine

Most days, the routine is just ChatGPT and Gemini. But roughly twice a week, something comes up that needs data work. Examples:

• I want to compare BTC’s current drawdown to historical drawdowns at the same Fear & Greed level. I ask Claude Code to write a Python script that pulls data from an API and calculates it.
• I need to format a batch of trade log entries from messy notes into a clean CSV. Claude Code does it in seconds.
• I want to calculate my actual win rate and average R:R from the trade log. Quick script.

Here is an example prompt I have used with Claude Code:

Write a Python script that:
1. Reads a CSV file called trades.csv with columns: date, entry, exit, direction, result
2. Calculates: win rate, average win %, average loss %, profit factor, largest drawdown
3. Prints a formatted summary table to the terminal

Claude Code writes the script and can execute it immediately. I do not need to set up a development environment or install packages manually — Claude Code handles that.

The Prompts, All in One Place

For easy copy-pasting, here are the prompts I use daily:

ChatGPT Morning Briefing

You are a crypto market analyst. No predictions. No hype. Analyze this data
with honest uncertainty.

TODAY'S DATA:
- Date: [YYYY-MM-DD]
- BTC price: $[PRICE] ([24H_CHANGE]% 24h)
- BTC dominance: [DOM]%
- Fear & Greed Index: [SCORE] ([LABEL])
- ETH/BTC ratio: [RATIO]
- Total crypto market cap: $[MCAP]
- Key events today/this week: [LIST EVENTS OR "none scheduled"]
- Notable overnight news: [HEADLINES OR "nothing major"]

GIVE ME:
1. SENTIMENT (2-3 sentences)
2. KEY LEVELS: BTC support/resistance
3. DOMINANCE READ: What does BTC.D movement mean for alts?
4. RISK FACTORS: What could go wrong?
5. STANCE: Aggressive / Cautious / Neutral + one sentence why

Be concise. If data is insufficient, say so.

Gemini Chart Read

I am pasting a 4H BTC/USDT chart with an EMA ribbon overlay.
Describe:
1. Current trend direction (ribbon color)
2. Price position relative to EMAs
3. Any visible chart patterns
4. Most recent significant candle analysis
Do NOT predict. Only describe.

Gemini On-Chain

Here is an on-chain metrics dashboard. Describe:
1. Exchange net inflows or outflows
2. Active address trend
3. Any metric that looks unusual vs. recent range
Describe only — do not predict.

What This Routine Does NOT Do

I want to be clear about the limitations because I see a lot of AI trading content that glosses over them:

It does not predict prices. Not one part of this routine attempts to forecast where BTC will be tomorrow. The entire point is to assess current conditions, not predict future ones.

It does not guarantee profits. I have had losing trades on days when the routine output was perfectly clear and aligned. Markets do things that no amount of analysis — AI or human — can anticipate.

It does not replace learning to read charts yourself. If you do not understand what RSI or EMA means, the AI’s output will be noise to you. The routine assumes you have at least a basic understanding of technical analysis concepts. The AI accelerates your research; it does not replace foundational knowledge.

It does not work if you skip the logging. The routine without the spreadsheet is just morning entertainment. The spreadsheet is what turns it into a feedback loop that improves over time.

My Results After 30 Days

I tracked this routine from mid-January to mid-February 2026. Here are the honest numbers:

• Days I completed the full routine: 24 out of 30 (I skipped weekends sometimes and missed two weekdays when I overslept)
• “Do nothing” days: 16 out of 24 (67%)
• “Adjust bots” days: 5 out of 24
• “Manual trade” days: 3 out of 24
• Manual trade results: 2 wins, 1 loss (small sample, do not read too much into this)
• Bot adjustment results: Hard to isolate, but my DCA bot performed about 8% better in the second half of the period versus the first, possibly due to better parameter timing

The biggest measurable impact was not from the trades I took — it was from the trades I did not take. Before the routine, I was placing impulsive trades 2-3 times per week based on gut feelings. During the 30-day test, I placed exactly 3 manual trades, all with documented reasoning. My emotional trading dropped to near zero.

That alone was worth the 15 minutes.

Tool Comparison for Research Tasks

This table reflects my experience as of February 2026. These tools update frequently and the comparison may shift. I re-evaluate roughly every two months.

The Most Valuable Output Is “Do Nothing Today”

I want to end with this because it is the most counterintuitive lesson I have learned.

When I started using AI for trading research, I expected it to help me find more trades. Instead, the biggest value has been finding fewer trades. The routine forces me to write down “no action” on most days, and seeing that written in the spreadsheet reinforces the habit of patience.

Before the routine, a day without trading felt like a wasted day. Now, a day where the routine says “do nothing” and I listen feels like a disciplined day. That mental shift — from “I need to trade” to “I need a reason to trade” — is probably the single most profitable change I have made.

The AI did not teach me this lesson directly. But the structure of the routine, with its forced decision point and logging, made the lesson visceral instead of theoretical. I can look at my spreadsheet and see that my best weeks were the ones with the fewest trades. No AI chatbot told me that. The data did.

Fifteen minutes. Three tools. One decision. Most days, that decision is to do nothing. And that is the whole point.

Next Steps

• Deep dive on ChatGPT — Get more prompt templates in our ChatGPT trading analysis playbook
• Indicators mentioned in this routine — See the 3 AI indicators I actually use including the EMA Ribbon referenced above