Headlines are critical in news coverage. They are perhaps the most important element of an article driving whether it gets read or not.
Every time I say this I’m reminded of why Joel Abrams from The Conversation said it to me: journalists work hard to create their stories, but then sometimes fall short when attaching a compelling headline to them.
The app puts the headline at the center of the user experience in Slack. Whether it's pulling what is in your story and relating it to what's in your headline, or using Generative Pre-trained Transformer (GPT) to aid in the creation of better headlines, YESEO aims to be a useful assistant helping newsrooms gain a wider audience with their most important stories.
How to use YESEO
As part of my Reynolds Journalism Institute (RJI) Fellowship, I worked with many newsrooms to build YESEO as an app for Slack. It is the culmination of months of improvements that have been made to YESEO since it launched in March through RJI.
This makes the app much more accessible to users worldwide: when you look to add an app in your workspace, search “YESEO” and you will find it right away. Once you click “Add to Slack” you will be brought to the yeseo.app homepage to add it.
Adding YESEO to the Slack app directory also means that it comes equipped with Slack’s best practices, like messaging “help” to get a tutorial within any command, or for general help. Now you can ask and the app will provide useful resources.
What I value most when I create tools is lowering the barrier to entry for users to get what they are looking for. The core features of YESEO seek to help users understand what keywords are in their story and give them ideas for how their story can be searched online.
To receive analysis of a story, simply use the “/analyze” command and paste a URL. Any unpublished story can be placed into a text box within Slack – it’s as easy as copy and paste.
The app presents buttons as jumping off points to quickly retrieve data. Each button leads to a new adventure, whether it’s finding keyword data, sharing a story to another user or using GPT data. Among these is a “Suggest Headlines” button.
“Technology and newsrooms equals fear for a lot of people,” one Canadian outlet told me recently. I have seen this first hand, where newsrooms can feel challenged by working with new technologies and want to resist change.
Every part of YESEO is seeking to show how technology like GPT can help a newsroom do their job better, and that each of the ideas it generates is a suggestion to help a newsroom craft a better headline. Users decide whether they want to use the data they see, based on the efficacy of the suggestions.
Journalists can also use YESEO to check their intuition. They can generate five ideas at the click of a button to jumpstart their idea process, and share headline ideas in group chats on Slack. Newsrooms I’ve talked to have said they have been known to combine parts of two headline suggestions to make a unique headline for their story.
Improving outputs through experimentation
I have been testing YESEO since it was released in March, looking to optimize the headlines it suggests. I first tested to see whether or not GPT 4 or GPT 3.5 returned more useful data to newsrooms considering GPT 4 cost 15 times more than the 3.5 model. In that A/B test I found that they were roughly equal.
A few comments I received in survey feedback stood out. Separate newsrooms told me: “Our style doesn't use title case for headlines” and “They aren't very specific and are in caps.” I knew this meant there was room for further optimization.
I made one small tweak to the headline prompt in my calls: I asked it to “use AP style.” This improved users’ satisfaction with the suggested headlines by 10 percentage points in each model, based on the survey feedback.
Many people have told me that GPT is not as knowledgeable at applying AP style wholesale, but when creating headlines it appeared to improve the suggestions substantially.
I have already begun testing how to optimize headlines by the outlet that is using it, which was inspired by feedback about how the app could “learn from its users.” The idea being, if your organization writes more conversational headlines, could the model start to understand that and deliver suggestions accordingly while still being able to deliver other style headlines for other outlets.
I am continually thinking about ways that I can optimize what YESEO has to offer for newsrooms. All tests through YESEO run as A/B tests, so users don’t know what data they are getting at particular times. Users receive either the control group or the test group, and then provide feedback.
Learning more about our users
More than 170 workspaces have installed YESEO for Slack since it launched in March, entrusted by users to give them data to make better decisions.
It has been amazing to learn the success stories from users, like when talking virtually to the Illinois News Broadcasters Association spring convention, hearing from Jenna Dooley, news director at NPR station WNIJ say insights from YESEO helped them get their highest viewed story in over a year. I’ve had multiple newsrooms in Oklahoma tell me how easy it has been to integrate the app into their workflow and get value out of it.
Learning these stories, from installation to usage to feedback, has also helped me shape the future of the product. Today, YESEO supports eight language models (English, French, Spanish, Portuguese, German, Finnish, Norwegian and Swedish). It doesn’t require an email, login or credit card to sign up.
The reason I wanted to be an RJI Fellow was because I wanted to use my skills to make a distributable product that can benefit more than just one newsroom, while simplifying the pain points of SEO and helping newsrooms incorporate useful, practical data to make their stories better.
Today, any workspace can install YESEO directly from the Slack App Directory and use a tool that helps journalists craft better headlines and get more from their stories.
Photo via Pexels by Pixabay.