Optimizing text length with ParagraphAI's Length slider | insights and benefits

A look into our dynamic and helpful Length edit slider, and how we ran tests to improve its overall quality.

ParagraphAI: The AI writing tool

At ParagraphAI, we are dedicated to empowering individuals to say what they mean. Our AI writing tool enables you to communicate instantly, anywhere, and in any language. Whether you're an ESL speaker, dyslexic, or a professional, our AI writing tool supports your communication journey, helping you say what you mean at the tap of a few buttons.

ParagraphAI's Length slider

Have you noticed any changes to our AI writing tool’s edit sliders recently? Today’s spotlight is on the ParagraphAI Length edit slider which helps users with text length adjustment. Need to trim down that novel you were about to text someone? What about a discussion post for school that could use some beefing up? ParagraphAI's Length slider helps you change the length of your text while maintaining the tone, meaning, or overall content with the help of our dynamic AI writing tool.

Why we chose it

We chose to experiment with the ParagraphAI Length slider because of some curious trends we observed in its usage data. Even though Length is designed to both shorten and lengthen text, an overwhelming majority of ParagraphAI users seemed to exclusively make their text longer. Not only this, but many users were choosing the option that made their text as long as possible. This begged the question of what people were using the slider for, and whether or not modifying it would cause infrequently-used positions to see more adoption.

Theories for testing changes

After reviewing existing data, we came up with a few ideas for modifying Length. There was a lot of emphasis on improving the shortening side of the slider since it was seeing far less use than its longer counterpart. Since the shortest option, which condensed a text into a bullet point summary, was used the least often, we decided to reevaluate its purpose: how else might someone want to reduce their text to the smallest amount of information possible? On the other end of the scale, what if the longest option was even longer? Would ParagraphAI users still see value in this part of the slider? We crafted a set of new instructions to prompt the model to make more dramatic changes in length between positions. This approach leverages the capabilities of our AI writing tool to offer enhanced flexibility and customization for our users.

How we measured success and chose the winner

In setting up the experiment, we chose to split test (or A/B test) the two versions of the Length slider for two weeks. These two versions were only shown to our iOS users. Both versions of Length were also made available to Free users for the duration of the test, so that we could get more representative user data. We chose our winning candidate based on the common usage metrics, such as how often people were using the slider, copying the outputs, and subscribing after usage. These metrics help to paint a picture of how much users want to keep using each version of Length, and how much overall value it is providing to them.

What we saw from our test

The new ParagraphAI Length slider was the clear winner between the two versions: users were more likely to both use and copy outputs. New Length also saw overall better conversion and marginally better retention compared to its older counterpart. Curiously, 16-17% of users shortened the text in both the old and new Length sliders, aligning with the trend seen prior to the experiment. We are always working to improve our edit sliders, so that users can edit their text in new and exciting ways. This improvement in our AI writing tool ensures users can efficiently adjust their text length as required with a simple, user-friendly slider. Testing helps us optimize the effectiveness of our AI writing tool and continue to set ParagraphAI apart in the AI market. That being said, what do you think? Give our new Length slider a try, and let us know your thoughts!

Check out another post from PAI: