Salt Lake Tribune: Tracking open-ended survey answers over time

For two years, the Salt Lake Tribune asked a segment of its audience a quarterly question. It consistently got back a clear Net Promoter Score (NPS) that was positive but variable, alongside a pile of open-ended answers that nobody had the time to read at scale. They had no good way to track topic frequency hidden in those comments and see how it had evolved over time.

Using the Audience Data Commons (ADC) schema, the Tribune’s Director of News and Audience Analytics, Ian Swenson,  structured open-ended survey responses into a repeatable framework that could help the organization see whether future strategic bets are paying off. 

Data in

As part of a pilot ran by the NPAI Co-Lab, Swenson pointed the ADC at a single data source: open-ended responses to a quarterly NPS survey that had run on The Tribune’s website since 2024. NPS is a standard measure of customer loyalty, and while the Tribune’s score was always positive, there was no obvious way to explain the number. 

About a third of the respondents had also filled in the single free-text comment field. In the pilot, Swenson wanted to test whether the Tribune could better understand how its audience felt about the brand by categorizing and analyzing the topics those respondents raised.

How the schema worked

In the past, Swenson had used ChatGPT to assess the general sentiment of responses, but it wasn’t particularly helpful. ChatGPT could identify themes in what readers said, but it could not quantify their feedback or support a repeatable framework. 

The schema gave the responses a fixed shape. “The data commons is useful because it can quantify that sentiment without taking manual time and energy,” Swenson said.

The ADC has a standardized structure for analysis, and once the open-ended text was mapped into standardized fields, topic frequency became something the analysis could return consistently so Swenson and his team can track changes over time.

Samples of how the schema categorized open-ended comments:

• Comment: “Your newspaper is very left-leaning, trying too hard to be the anti Deseret News.”

  • Output: political bias, competitor positioning

• Comment: “Love the trib”

  • Output: overall satisfaction 

• Comment: “Every LDS moment isn't news. We don't need 3-4 LDS stories as the lead item everyday.”

  • Output: LDS overcoverage, story prominence, topic diversity

Insight out

More than a strategic call, this pilot gave Swenson a method: a repeatable way to convert free-text audience feedback into a number that can be compared across quarters.

The Tribune dropped its paywall in May 2026, and because the NPS survey was tied to the subscription system, the survey stopped with it. Once the survey is running again, Swenson said he wants to use the ADC to measure audience reactions to the change  over time.

“If we made some strategic change somewhere, are we seeing a difference in how people responded to it?” he said. That could tell the Tribune how readers feel about the strategic decision to tear down the paywall.

Where it got hard

Swenson acknowledges that the tool, still in pilot when he tested it, was not a perfect out-of-the-box solution. He had to hack a bit to get the nuanced, topic-specific results he wanted, including by editing the tool’s configuration files. Organizations without staff engineers or data experts would need technical support to use the tool.

“It was still a technical lift, and I’m command-line proficient,” he said. “Ideally, future versions of the tool would allow users to quantify sentiments without modifying the system itself.”

The tool also required significant time and AI tokens to return its analysis. The Tribune’s initial ADC pilot project idea, which contemplated matching on-site user behaviors with audience segments, was even more token-intensive, so they scrapped that plan.

“The tool is best for … that connection between structured and unstructured data for known users,” Swenson said. After the pilot, he wants to explore attaching site comments for logged-in users with information from a data warehouse, which could help the Tribune develop personalization features to encourage on-site engagement.

Swenson said with the ADC, an analysis like that “seemed like it was near impossible, now it is possible.” 

👉 Want to test the ADC with your own newsroom data? Join us at the NPA Summit 2026 for a 3-hour speed hackathon! Come with one question about your audience that no single system can answer, and use the ADC to structure your first-party data around it. Space is limited and registration is required — register by October 13.