Phrasee's language insights give you a glimpse into the language experiments that the Phrasee Brain designs for your brand. The automatic insights and data visualizations allow you to see what is being tested and resonating with your audience. Language insights are currently available for broadcast and trigger projects.
This article covers the language insights available at the project or account level. Check out this article for information about language insight reports at the individual experiment level.
Where can I find them?
Language insights are automatically generated once the results for your experiments are received (i.e. completed experiments).
To view insights across experiments for a given project, select Reports in the left navigation bar.
To run a language insights report, select your project from the drop-down in the top right of the screen, and input a date range. Click Run report.
Note: We recommend you select a time period with at least 15 completed experiments with results for a given project to ensure there is enough data. You are best served using the experiment-level insights for anything less than this.
What am I looking at?
Once you've run a report, the screen will populate with various widgets and data visualizations about the language you have tested across the experiments within the selected project and time period.
The first thing you will see is a bar summarizing details of your report. This includes the project name, date range, number of experiments, and total variants.
Word and emoji performance
The next set of widgets displayed are the graphs for both word and emoji performance.
These graphs display the most frequently tested words and emojis, plotted against their relative performance for the selected time period.
Click on the three vertical dots in the top-right corner of the table to export the graphs.
These graphs are a good way to start getting a picture of what’s resonating with your audience:
The higher the bar is above zero, the more frequently the word is found in high-performing language variants.
The lower the bar is above zero, the more frequently the word is found in lower-performing language variants.
Note: If a word or emoji is shown to frequently appear in lower-performing variants, this does not mean that the word or emoji in isolation is contributing to lower performance and should be removed. What works for one audience or a particular time of year might not be what works forever. The key to keeping your results high is to continue testing!
Winning variants table
This table summarizes the details of your winning variants and the sentiments behind them for the selected time period.
The sentiment scores display the weighting of Phrasee's seven sentiments for each variant. The score is out of 10. The higher the score for a particular sentiment, the more prevalent it is in the variant.
Click on the three vertical dots in the top right of the table to export the table or show/hide columns. You can also sort the columns by clicking on each column header.
Average sentiment and sentiments over time
The average sentiment radar graph shows the average sentiments over the whole project and selected time period.
Check the tick box to view the average sentiments for only the winning lines for the selected time period and project.
The sentiments over time graph shows how the sentiments within your language have changed over time. If there are changes in how you employ different sentiments over time, you will be able to see those changes in this graph.
You can toggle the different sentiments on and off by using the checkboxes under the graph.
Click on the three vertical dots in the top-right corner of the table to export or save the graph.
Open rate by length
Testing a range of variant lengths increases experiment diversity, and diversity influences the uplift achieved. Many linguistic factors combine to affect performance; character count alone doesn’t tend to correlate with performance. High performers can be long as well as short - so keep testing! The Open rate by length plots your variants to show how length correlates to open rate across your experiments.
