Getting Started
Learn how to use Watch Your Tone and interpret the results
Video Walkthrough
Video tutorial coming soon
A 2-3 minute walkthrough showing how to navigate the dashboard, interpret sentiment scores, and understand emotion classifications.
Quick Start Guide
View the Dashboard
Start on the main dashboard to see sentiment trends over time. The timeline shows how news tone changes throughout the day.
Go to View the Dashboard →Explore Headlines
Use the Explore page to search and filter headlines by outlet, emotion, or sentiment range. Click any headline to see the AI breakdown.
Go to Explore Headlines →Compare Outlets
Check the Outlets page to see how different news sources vary in their emotional tone and sentiment patterns.
Go to Compare Outlets →Understand Emotions
Visit the Emotions page to learn about the 11 emotions we track and see their current distribution across headlines.
Go to Understand Emotions →How to Interpret the Data
Here are three real-world scenarios showing how to read and understand sentiment patterns. Remember: we analyze emotional tone, not political bias or factual accuracy.
📊Scenario 1: Breaking News Event
What you might see:
- All outlets show negative sentiment (e.g., -0.6 to -0.8)
- High "fear" (0.65+) and "sadness" (0.50+) emotions across outlets
- Low subjectivity scores (0.20 or less)
- High AI confidence (0.80+)
How to interpret:
When all outlets show similar negative emotions, the event itself is likely objectively negative (e.g., natural disaster, tragedy). Low subjectivity and high confidence suggest factual, straightforward reporting rather than emotionally-charged framing.
💡 Key Takeaway
Consistency across outlets = the story itself, not editorial framing choices
🔄Scenario 2: Political Announcement
What you might see:
- Outlet A: Positive sentiment (+0.5), high "trust" (0.60) and "anticipation" (0.55)
- Outlet B: Negative sentiment (-0.4), high "skepticism" (0.50) and "anger" (0.45)
- Higher subjectivity scores (0.40-0.60)
- Moderate AI confidence (0.65-0.75)
How to interpret:
When outlets diverge significantly in emotional framing, they're choosing different angles on the same event. Higher subjectivity indicates opinion-driven language. This reveals how outlets frame stories, not their political lean—use this to understand diverse perspectives.
💡 Key Takeaway
Divergence across outlets = different emotional framing of the same story
📈Scenario 3: Economic Report
What you might see:
- Sentiment shifts from negative (-0.5) to positive (+0.3) over a week
- "Fear" decreases from 0.70 to 0.30
- "Relief" increases from 0.15 to 0.55
- Topic stays the same but sentiment timeline shows clear trend
How to interpret:
When sentiment trends shift over time on the same topic, the narrative is evolving. Initial concerns may be easing as more data becomes available, policies are announced, or conditions improve. Use the timeline chart to track these narrative shifts.
💡 Key Takeaway
Trends over time = narrative evolution and changing public discourse
🎯 Pro Tips for Analysis
- Compare, don't judge: Look at differences between outlets, not which is "right"
- Check confidence scores: Low AI confidence (below 0.60) means uncertain classification
- Watch subjectivity: High subjectivity (0.60+) indicates opinion-heavy language
- Track over time: Single headlines are less meaningful than patterns and trends
- Use framing notes: Read the AI's framing analysis for context on tone choices
Frequently Asked Questions
What is sentiment analysis?
Sentiment analysis measures the emotional tone of text on a scale from -1 (very negative) to +1 (very positive). It helps understand whether headlines use positive, neutral, or negative language.
Does negative sentiment mean the outlet is biased?
Not necessarily. Negative sentiment could reflect the nature of the story itself (e.g., disasters, conflicts). Compare sentiment across outlets covering the same story to identify potential framing differences.
How accurate is the AI?
Currently we use OpenAI's gpt-5-nano & gpt-5-mini (with more to come in the future). However, no AI is perfect - context and nuance can sometimes be missed. Use this as one tool among many for media literacy.
What are emotions in this context?
We detect several emotions, and do not denote within prompts any restrictions on the model for emotion selection. Such emotions noted include but are not limited to: joy, fear, anger, sadness, disgust, surprise, neutral, trust, anticipation, relief, and admiration. Headlines can express multiple emotions simultaneously.
Can I use this for fact-checking?
No. This tool analyzes how news is framed (tone and emotion), not whether it's factually accurate. Sentiment analysis is complementary to, not a replacement for, fact-checking.
Why do some headlines have no classification?
Newly collected headlines may not be classified yet. Classification happens periodically as new headlines are gathered.
Ready to Start Exploring?
Head to the dashboard to see the latest sentiment trends and emotion patterns.
Go to Dashboard