How YESEO analyzed 60,000 AI-generated headlines and decided to pivot to paid source tracking

Ryan Restivo built YESEO to solve a simple problem: help journalists with SEO best practices, which set sight on the headline to start. Two years and 60,000 AI-generated headlines later, the data told him something unexpected about what newsrooms actually needed.

The free Slack app started as a button that suggested five headline options for any story. Reporters could click, review the suggestions, and move on. Simple utility. But Restivo noticed a pattern in how people used the tool. More than 70 percent of stories entered YESEO before publication, not after. Journalists wanted help earlier in the reporting process, not just at the headline stage.

That behavioral data drove Restivo to announce three new paid features last week - focusing on source tracking, story pitching, and follow-up angles. He’s partnered with Kimbap Media founder Emma Carew Grovum to recruit 10 newsrooms for a testing cohort that will run through early 2026.

The shift represents a calculated bet on what AI tools can actually sell to struggling newsrooms. Free headline optimization attracted over 650 workspace installations and helped process 16,000 stories. 

“These new features build upon what I’ve learned not just as an RJI fellow [Reynolds Journalism Institute (RJI) at the University of Missouri], but everything I’ve learned working with newsrooms to help solve problems,” Restivo told me. “I’m motivated to empower newsrooms by building practical innovation that solves workflow inefficiencies. The technology is a tool to help do that.”  

Restivo ran nine different A/B tests on the headline suggestion feature, generating more than 60,000 headlines to optimize the output. The data showed patterns in how large language models construct news headlines. Words like "reveals," "exploring," and "navigating" appeared frequently in AI suggestions. The word "discover" showed up often but ranked outside the top 10 verbs that models actually used.

The most common verb in AI-generated headlines appeared in less than 1 percent of real headlines that journalists wrote and the tool processed. Restivo presented his analysis at an Online News Association event in June, turning it into an interactive session where attendees tried to distinguish AI headlines from human ones.

“Creating this programming this year has been a great way to share everything I’ve learned”, Restivo said. It was cool to see people so far guess with about 61% accuracy a headline on a story versus one generated by an LLM.” 

YESEO only activates AI models when users click specific buttons, never analyzing content in the background. That permission-based design became central to his pitch for the new features.

The strategic direction came from work with The Oglethorpe Echo, a nonprofit newsroom in Georgia run through the University of Georgia. YESEO partnered with the Echo as part of the JournalismAI Innovation Challenge, supported by the Google News Initiative.

The problem: student reporters cycle through the newsroom each semester without institutional memory of local sources. Who had they quoted before? What positions did sources take on recurring community issues? The knowledge evaporated every few months.

Restivo built a source tracking system that analyzed every quote the Echo published. Reporters could see how many times they quoted specific people, what those sources said across multiple stories, and which topics they discussed. The tool identified organizational affiliations and mapped relationships between sources based on their appearances in coverage. “If I had had a tool like this as a young journalist, I could have scrutinized those points better with context and have provable facts on what a source is all about before I talk to them”, Restivo said.

Amanda Bright, the lecturer who oversees the Echo, said the newsroom produced one additional reported piece and one extra video story per week because of efficiency gains from Restivo's tools. That output increase enabled a video advertising program and subscriber growth.

The key insight: journalists needed help understanding their own archives before they needed help writing headlines. The source tracking solved a different problem than SEO optimization, but it used the same underlying capability to analyze text at scale.

“Most of their sources don’t have Wikipedia pages, which means The Oglethorpe Echo is one of the main sources of truth for the county,” Restivo said. “It’s vital that in this digital era journalists are delivering and using the most reliable information and I think how we have solved this has many applications for this industry writ large.”

The YESEO app was created through the RJI which enabled Restivo to offer his tool for free. Now Restivo faces the transition to embrace a freemium model. The existing features will remain free, but the three new ones will need to be paid for. Source auditing requires processing entire newsroom archives. The story pitch analyzer and follow-up suggestion tool need more complex natural language processing than simple headline generation.

Carew Grovum is recruiting 10 local newsrooms for the testing cohort with explicit conversations about pricing built into the program. International applicants are welcome, Restivo told me. 

“We want willing partners who are open to sharing their stories with us, so we can generate insights for them where they otherwise might not have,” Restivo said. “Our application is asking questions so we can understand the type of newsroom they are. We want to try our best to test with as broad a sample of newsrooms that we can that represent all archetypes of organizations around the US and the world, if possible.”

Participants won't pay during the testing period, but Carew Grovum wants to talk with people who have budget authority and can explain their purchasing cycles. The goal: understand what makes a newsroom become a paying customer.

The application requires that newsrooms have a content management system where YESEO can pull stories. That technical requirement filters for organizations with sufficient infrastructure to integrate the tools. Selected newsrooms will join biweekly meetings, complete monthly surveys, and develop case studies. Carew Grovum will facilitate while Restivo codes.

YESEO's evolution from free headline tool to paid source tracking illustrates a broader challenge for AI products in journalism. Building something useful is different from building something newsrooms will pay for.

Headlines are universal. Every story needs one. SEO optimization feels like table stakes, something that should be free or built into existing tools. But institutional knowledge about sources? That solves a specific pain point that varies by newsroom. Student newsrooms lose knowledge every semester. Small nonprofits lack searchable archives. Legacy outlets have decades of coverage buried in unstructured formats.

“The more I’ve shown what I have planned for how newsrooms can understand their sources better, the more I’ve seen people willing to insert their own newsroom into the experience I share to them,” Restivo said. “It’s been exciting to see how something that could scale for a small nonprofit news-academic partnership can also scale for larger publications with many tens of thousands of stories.”

Restivo is betting that newsrooms will pay for tools that analyze their unique archives rather than just optimize their individual stories.

Whether that bet pays off depends on execution and timing. Newsrooms face intense financial pressure. They scrutinize software costs. But they also recognize that institutional knowledge walks out the door with experienced reporters. If Restivo can demonstrate measurable efficiency gains in legacy newsrooms like The Oglethorpe Echo achieved, he has a case for paid adoption.

Applications for the test cohort close October 24, with the cohort starting in early to mid-November. If you have any questions you can book an info call with Crew Grovum.