Browse The Report
- Nonprofit Email Deliverability & Engagement Benchmarks
- Nonprofit List Sizes
- Ask The Expert: Is List Segmentation Really That Important?
- Nonprofit Email Bounce Rates
- Nonprofit Email Open Rates
- Nonprofit Email Unsubscribe Rates
- Nonprofit Email Click-Through Rates
- Nonprofit Email Fundraising Performance
- Nonprofit Email Performance by Date & Time
- Email Sender Superlatives
- A Data-Driven Approach to Subject Lines & Preview Text
- Convey Positive Emotions in Subject Lines
- Ask the Expert: How Did You Use AI for Subject Line Sentiment Analysis?
- Words to Include (or Avoid) in Your Subject Lines
- Experiment with Emojis in Subject Lines
- Write Compelling Preview Text
- Words to Include (or Avoid) in Your Preview Text
- Put It All Together — Performance Benchmarks & Word Usage
- Creating Effective Emails
- Ask the Expert: What Should I Keep In Mind When Creating Compelling Emails?
- Tip #1 — Include Imagery in Your Emails
- Tip #2 — Pay Attention to Salutations
- Ask The Expert: Do Salutations Really Make a Difference?
- Tip #3 — Use the Word "You"
- Tip #4 — Make Your Message Scannable
- Ask The Expert: How Do I Create a Great Call to Action?
- Tip #5 — Include Great Calls to Action
- Put It All Together - Build Clear, Compelling Emails
- Lessons from the Most Engaging Email of 2022
- Data-Backed Insights for GivingTuesday and Year-End
- Methodology & Appendix
- About Neon One
Ask the Expert: How Did You Use AI for Subject Line Sentiment Analysis?
Previously, the process of understanding the emotions behind the text was time consuming and relied on human empathy. But with the advent of large language models, nonprofits can now quickly and cost-effectively classify emotions in the text through automation.
“For the purposes of this report, we classified each subject line using OpenAI’s Davinci-003 model as one of the following 27 emotions: Admiration, Approval, Annoyance, Gratitude, Disapproval, Amusement, Curiosity, Love, Optimism, Disappointment, Joy, Realization, Anger, Sadness, Confusion, Caring, Excitement, Surprise, Disgust, Desire, Fear, Remorse, Embarrassment, Nervousness, Pride, Relief, Grief, or Neutral (not counted as an emotion).
“GPT-J provides a better opportunity for emotional analysis, and an updated finding will be shared when these tools can more easily and quickly ingest the volume of data provided. Davinci-003 did create some noise in the classification that slightly impacted the findings. For example, if the completion output was “The sentiment is love” instead of “love,” it is possible our natural language query of the database excluded this result. However, this was a very small percentage of the overall output.