Prompting Techniques

AIXBT's system prompt handles most of the analytical framework. It knows how to evaluate projects, interpret signals, and contextualize market data. Your job is to tell it what you're looking for, not how to think about it.

This means prompts can be surprisingly simple. But there are techniques that help when you want more control over what surfaces.

Open vs. Specific

Both approaches work, and they serve different purposes.

Open-ended prompts let the model's reasoning surface patterns you might not think to ask about. "What's interesting right now?" or "What should I be paying attention to?" can uncover signals you'd miss with a narrow query.

Specific prompts help when you know what you're looking for. Add constraints like thresholds, timeframes, or exclusions to focus the search on exactly what you need.

Narrow queries give you control, but you might filter out unexpected discoveries along the way. Sometimes the best approach is to start open and see what surfaces.

Adding Constraints

When you want to narrow results, add explicit filters:

Thresholds - "projects with TVL growth above 20%" or "tokens under $50M market cap"
Timeframes - "in the last 24 hours" or "this week"
Exclusions - "exclude major tokens like BTC, ETH, SOL" to focus on discovery
Categories or Narratives - "DeFi protocols" or "privacy memecoins"

Constraints turn exploratory questions into screening tools.

Targeting Clusters

AIXBT tracks distinct communities across Crypto Twitter. You can focus queries on specific segments:

"What are VCs discussing?"
"What's trending among speculators?"
"Which projects are builders excited about?"

Cluster-specific queries help when you care about who is talking, not just what's being said. Different communities have different signal value depending on what you're looking for.

Start broad, then narrow. A general question like "what DeFi narratives are emerging?" might surface several threads. Follow up on the interesting ones: "Tell me more about [specific project]" or "How does this compare to [other project]?"

The conversation context carries forward. Each response can inform your next question.

Lateral Discovery

One finding can become the lens for the next query. Instead of drilling deeper into the same topic, pivot sideways to explore adjacent signals.

For example: "Which clusters were early to the current meta? What new projects are they discussing now? Show me all of them, but highlight anything outside the current meta."

This chains insights: identify who had good signal → see what else they're watching → filter for things the crowd hasn't noticed yet. Each step uses the previous answer to unlock a different search space.

Useful when you want to find what's next rather than understand what's already visible.

Requesting Structure

For reports or comparative analysis, you can ask for specific output formats:

"Give me a summary with bullet points"
"Compare these projects in a table"
"Rank these by momentum and explain why"

Useful when you want scannable output or plan to reference the response later.

Guiding the Analysis

You don't need to explain how to analyze. But you can if you want different emphasis:

"Focus on on-chain metrics rather than social signals"
"Weight recent activity more heavily"
"Consider bear case arguments"

The default analysis is usually what you want. Override when you have a specific angle.