When I start writing, I stare at the blinking cursor.
The hardest part for me is to write the first sentence.
When building Resonate AI, I made a point to model the best communicators.
I begged them to be my mentors.
I hopped on calls with them.
I read their writings.
Josh Fechter was one of the writers who agreed to help me out and teach me how to write well.
“Focus on your first sentence and your last sentence. It’s the most important part of your writing.”
His advice inspired me to take a deeper dive.
Here’s my step-by-step process for predicting a viral post:
Step 1: Collect a large dataset
I explored if there was a pattern around how the best writers craft their first and last sentence.
We hand-picked hundreds of articles with 1000+ engagements.
My eyes were hurting from reading hundreds of articles. I got the computer screens with bluelight filter.
We spent several weekends scrolling through feeds of top writers to build our dataset.
A dataset of engaging writings from award-winning writers.
Step 2: Remove the bias
Then we spent several weekends cleaning up this dataset to remove any biases.
If an article was too promotional, we removed it from the dataset.
If it had seasonal, we removed it.
If it was political, we removed it.
It was a “wax on, wax off” phase of our data science product development.
Step 3: Explore dataset. Define the questions
Once we got our data to a reasonable point, we started performing data jiujitsu on it.
The goal was to uncover insights around how best writers create their first sentence.
I’m a big fan of Sentiment Analysis API from Google ML Team.
First order of business.
Do articles whose first sentence have a positive sentiment perform better or worse than articles whose first sentence have a negative sentiment?
Step 4: Extract Insights
Sentiment Analysis API produces two values:
- Sentiment Score: A normalized value, it gives a range of -1(negative sentiment) to +1 (positive sentiment).
- Sentiment Magnitude: Strength of emotion.
We plotted a relationship between magnitude and sentiment.
Next, we combined magnitude and score to classify each sentence into groups as recommended per the Google NLP documentation:
- Strongly positive (score of 0.3+ and magnitude of 0.5+)
- Strongly negative (score of less than 0.3 and magnitude of 0.5+)
- Mixed (score between -0.3 and 0.3 but a magnitude of 0.5+)
- Neutral (any with a magnitude of less than 0.5).
Here’s what we found:
Writings exhibiting a strong negative sentiment in the first sentence perform the best and by a rather notable amount.
Negative opening sentences stomp positive opening sentences by a whopping 65%!
Mixed first sentences (sentences that show both strong positive and strong negative sentiment) perform the second best.
Articles with neutral first sentences (no strong emotion one way or the other) perform the third best, not that much better than articles with strongly positive first sentences.
Correlation doesn’t equal causation, but it is worth noting that there is a strong TREND among higher performing articles that the first sentence often registers a negative or strongly negative sentiment.
How to apply this in your writing?
To optimize for engagement, start with your pain.
Here are a few examples.
Note that all three of these examples have a pattern:
Pain + Tangible Example = An emotional tone of Openness
Step 5: Perform the last sentence analysis
The last sentence is one of the most important sentences.
We took a deeper look at all the last sentences from our dataset to uncover any patterns around them.
- Avg Engagements for Strongly Positive Last Sentences: 6124.8
- Avg Engagements for Strongly Negative Last Sentences: 3818.6
We processed all the last sentences through Google’s NLP API, and here’s what we uncovered:
Articles whose last sentence register as strongly positive perform an average of 61% better than articles whose last sentence register as strongly negative.
Mixed sentences perform the second best (and interestingly, is very close to the average of articles with strongly positive last sentence). Neutral sentences perform the third best.
How to apply the last sentence analysis in your writing?
Articles with a strong positive or mixed last sentence perform the best.
Most engaging articles are similar to a story.
An article is satisfying to read if it starts with a negative tone (problem) and ends with a positive tone (solution/answer).
Much like a story, both a conflict and a resolution are often crucial ingredients to a great article.
The hard part of high-engagement writing is not finding the right words.
The hard part is being vulnerable.