Natural Language Processing

Using an AI-generated content model to give reader’s the key takeaways of published stories for quick consumption and an enticement for further reading.

Year
2024

Role
Senior Product Designer

Objective

Next Gen research has shown the value that our audience places in brevity. They demand, crave, and expect news organizations to provide quicker entry points and access to consume journalism. Our site experience was not conducive to this desire. This results in the majority of users not reading the full article and high bounce and exit rates.

To better serve our readers, we have created a Natural Language Processing model that pulls key takeaways from published articles. Regardless of the surface or device, users are able to, with one click, see an article’s summary to help them digest the main information conveyed. venience and increase engagement with our products.

Goals

Engage

Drive growth and engagement by meeting users where they are and evaluating changes in consumption patterns

Implement

Using the Natural Language Processing model for brevity and enhancing to accessibility across the site

Facilitate

Facilitate easier consumption of main information from published article

Results

As of May 28, 2024

App engagement

26K+

Web engagement

Analytics coming soon..

Recirculation engagement

Analytics coming soon..


Design Process

  • As a part of our Next Gen research operation, we’ve collected insights previously that alerts that our users value brevity. They demand and expect news organizations to provide quicker entry points and access to consume journalism. Insights include:

    • Attention span trends

      Studies show that attention spans have decreased over recent years, especially among younger generations. This suggests a preference for concise information delivery.

    • Social media behavior

      Analysis of social media usage patterns reveals a preference for quick, bite-sized content like tweets and Instagram stories, indicating a desire for brevity.

    • Mobile usage patterns

      With the increasing dominance of mobile devices for news consumption, there's a growing need for content that can be quickly consumed on-the-go, suggesting a demand for shorter, more accessible formats.

    • Engagement metrics

      Research on user engagement with long-form articles shows a significant drop-off in reading time after the initial entry point, indicating a need for quicker access to key information.

    • Consumption platform trends

      Popular content platforms like podcasts and video streaming services often offer shorter, more condensed versions of longer content, reflecting a broader trend towards brevity in content consumption habits.

    • Emerging News Formats:

      The rise of platforms like TikTok and Snapchat Discover, which offer short-form news content, demonstrates a shift towards more concise and easily consumable news formats among younger audiences.

    • As a reader, I want to see an article’s key takeaways with one click so that I can better understand the topic at hand and to help me decide if I want to continue reading.

    • As a reader, I want to see some visual element that introduces and explains the feature to me in case this is my first visit.

    • As an editor, I want to ensure every key takeaway is first reviewed and approved before published and visible to readers.

    • As an editor, I want the design and UI to have a clear transparency disclaimer that the takeaways were initially created by AI, and edited by humans.

    • As a reader, I want to provide feedback on an article’s key takeaways and if they were helpful in my discovery journey.

  • We first deployed the feature as an A/B test. The control group are users that see the article pages and recirculation as is, sans AI-generated summaries. The control group see’s article summaries on both the article pages and recirculation module. The test ran for two weeks in order for us to gather critical data on the impact of the feature to help determine how we can scale it further.

    While we are still currently collecting analytics, so far we are seeing a significant difference in users engagement on articles with summaries. This is positively supporting our hypothesis.

    1. Page views/sessions of users in test group

    2. Page views/sessions of users who interact with key takeaways

    3. Average session time of users in test group

    4. Session time of users who interact with key takeaways

    5. Scroll “complete” rate between control and variant

    6. Total summaries consumed

    7. Average active days of users in test group

    8. Average active days during test of users who interact with takeaways

  • News consumption trends are constantly evolving, presenting us with ongoing opportunities to experiment with innovative placements and formats in future phases. For instance, I am particularly interested in exploring how different presentation styles, such as bullet points versus narrative summaries, impact consumption patterns. Additionally, implementing user-facing controls could empower readers to customize their experience, potentially enhancing engagement and satisfaction.

    In our recent efforts to integrate AI into our journalism workflow, I learned the critical importance of transparency when introducing AI to our audiences for the first time. While our system leverages AI to generate initial summaries, we ensure that human editors meticulously review and refine these outputs. It was paramount to communicate this clearly to our users, emphasizing that while AI plays a role, human oversight remains integral to maintaining the quality and reliability of our content. This approach not only builds trust but also educates our audience about the evolving tools we use to enhance their experience.

Concepts