Every morning at 06:00 UTC, an AI agent scans 50+ sources and delivers a research brief to my Telegram. Here's the exact prompt.
The Prompt (Copy-Paste Ready)
You are a research agent scanning AI/tech developments.
TASK: Find the 3-5 most important AI stories from the last 24 hours.
PROCESS:
1. Search for recent news on: [your topics]
2. For each story, evaluate:
- NOVELTY (1-10): Is this actually new?
- RELEVANCE (1-10): Does this matter for [your audience]?
- IMPACT (1-10): Will this change anything?
3. Only keep stories scoring 7+ on ALL criteria
4. Find PRIMARY sources (not summaries)
5. For each story, identify the UNIQUE ANGLE:
- What insight is nobody else mentioning?
- What's the contrarian take?
OUTPUT FORMAT:
## 🔥 Top Story
**[Headline]**
Source: [URL]
Why it matters: [2-3 sentences]
Unique angle: [what others miss]
## 📰 Other Notable
- **[Story]**: [insight + source]
## 🎯 Content Opportunities
[Article ideas from today's news]
Why It Works
Explicit scoring prevents mediocre stories from getting through.
"Unique Angle" requirement forces insights, not summaries.
Primary source hunting prevents citation telephone games.
Built-in gaps section acknowledges limitations.
Customization
- Change [your topics] to your specific interests
- Adjust threshold (7+ → 6+ for more results)
- Add constraints: "Exclude crypto" or "Focus on dev tools"
Full breakdown: https://ai-insider.ghost.io/the-prompt-that-runs-my-daily-research-copy-it/
What prompts power your workflow?
Top comments (1)
Solid research prompt structure! The key insight here is that a well-defined prompt template that you reuse daily is far more valuable than a clever one-off.
Your prompt has most of the right components, but one thing that can make it even more powerful: separating the structural blocks explicitly so the model handles each dimension independently. I built flompt (flompt.dev) for exactly this — it's a free visual prompt builder that decomposes prompts into typed semantic blocks (role, objective, constraints, output format, chain-of-thought, etc.) and recompiles them to Claude-optimized XML. Your research prompt would translate cleanly into blocks, and you'd be able to tweak individual parts (e.g., change depth without touching the output format) without rewriting from scratch.
No account needed. Might be useful for your daily workflow!