News
How building the AI-powered local news audience assistant bot helped us help local journalists – and learn about AI
By Sarah Vassello ’17, Project Manager, Center for Innovation and Sustainability in Local Media
For local newsrooms, 2024 is the year of AI.
At the Center for Innovation and Sustainability in Local Media, which exists to help the local news industry, we built the Local News Audience Assistant to help journalists find exactly what they need, using AI.
For those interested in how journalists and those who support journalists can use AI, the assistant can also act as an introduction to the new technology. It certainly did for our team.
The Local News Audience Assistant is a meta-search of vetted growth resources by journalists for journalists.
The chatbot was created out of the need for discoverability. There are hundreds, if not thousands, of valuable resources about building and growing audiences in the form of guides, Q&As, case studies, videos, audio conversations and more. However, once published, they often are lost to the black hole of the internet, making it hard for time-strapped journalists to find the most accurate resource they’re looking for.
For years, CISLM Director Erica Perel ’98 and I have discussed how the Center could solve this problem. We dreamed of a tool like the Audience Assistant, but we couldn’t identify an efficient path forward.
Then ChatGPT announced the ability to create custom chatbots, and we started reading about local news applications. Our interests were piqued.
Was this the answer that we were looking for?
Also, what was AI capable of, outside of the Grammarly browser extension? Should we be using it? We were starting at square one with the new technology and an incomplete understanding of what was possible.
Collaboration Over Competition
At the Center, it is collaboration over competition, whenever possible. This project was no exception.
A key goal with this project was to give credit and traffic back to the original sources: The journalism industry has struggled with the idea of ChatGPT using information without a link back to the original work, essentially stealing information while cutting site traffic. The team created this bot to do the opposite.
This project needed to list both the original source and give a link back to the site.
Both those at the University of North Carolina at Chapel Hill and in other journalism academic circles have been essential in helping us get this project off the ground.
We were inspired to try this project through Joe Amditis’ excellent guide — Beginner’s prompt handbook: ChatGPT for local news publishers — published by the Center for Cooperative Media at Montclair State University. In it, he laid out essential terminology, examples of use cases, tips on crafting helpful prompts and more. He also kindly made himself available to answer our questions.
While working on this project, I spoke to a UNC Hussman graduate, Anthony DeHart ’23, director of operations at Blue Sky Innovations at Reese Lab. DeHart is an AI expert helping Blue Sky and UNC students further media innovation through emerging media, including virtual reality and robotics.
DeHart told me that the biggest asset in building this bot was old school — the Google Sheet created to gather and vet resources. As low-tech as it might be, a Google Sheet, albeit an expansive one, powers this chatbot.
There are certainly ways to build a chatbot that involve GitHub, manual code and a full understanding of the complexities behind AI. But it doesn’t have to be that complicated. The tech stack for this product is a Google Sheet and a Zapier bot, with an embed code for the cislm.org website to make it consumer-facing.
It was DeHart who suggested Zapier as a consumer-facing web option.
A win for journalism: Talking to sources, getting out into the community, learning from experts is still the best way to learn.
The Process
Building this bot was a multi-step process, and an ongoing one.
First, I did some searches using ChatGPT and Microsoft Copilot to get a sense of how AI-powered tools worked: what results were frequently appearing, what was missing, etc.
The biggest piece of the process puzzle fell into place when the team utilized the Local News Support Database. This database is a resource of projects and organizations that support local news.
The database categorizes organizations as “Industry Support,” “Trade Organization” and “Vendor, Consultants and Platforms” — which was also a smart way to feature resources from journalism support organizations.
The student employees and I then searched through the sites to gather relevant resources.
Limitations help keep a project in scope and relevant. For this project, these limitations were as follows:
- Resources must be published in 2018 or later.
- Resources must not be overtly sponsored materials, such as marketing material for a specific product.
- Sponsored resources that featured a variety of sources and were topic-based rather than product-based were included (e.g. A panel recording about Twitter/X that includes a representative from the company as well as journalists using specific strategies).
- Resources must not be purely research but rather actionable advice.
We published the chatbot to our website and announced it in early 2024.
The Journey is Never Over
But what does launching a product really mean? For the chatbot, it’s just the beginning.
Through Zapier, we’re collecting user journeys to identify where folks are dropping off and which resources they’re most actively seeking.
The team is continuing to add more resources, both newly published and from other organizations that have yet to be imported.
And we’re still learning! The UNC community has been so supportive. UNC Research invited me to speak as a practitioner alongside two professors who have studied AI for 10+ years, and I was able to learn the theory around AI and communication. The data science school has taken meetings with me. Our research director, Jessica Mahone, also moderated a panel on AI and the First Amendment, bringing to light the ways local newsrooms are using AI. It’s wonderful to join the UNC community and learn from each other to experiment with what’s working and potential use cases.
The larger journalism community has been a wonderful source of knowledge as well. Perel and I were invited to Poynter’s AI and Ethics Summit this summer, which brought together about 50 journalists from across the country to discuss standards of ethics on AI and brainstorm AI-based solutions to newsroom problems.
This not only connected us with others doing incredibly cool and powerful work, but it inspired us to create our own Technology Ethics Policy.
We’re always seeking user feedback — feel free to email cislm@unc.edu with any thoughts, suggestions or feedback you might have. And please do give the bot feedback directly when chatting — it really helps!
It’s a journey that I’m grateful to have support for, and I would love to pay it forward. If you’re working in a local newsroom, you could experiment with AI by creating a database of articles on a topic and creating a chatbot to interface with readers, for example. Other applications of this technology could be a bot that gives readers a recipe from a local restaurant after they submit the ingredients in their fridge or a gift guide with featured items from local businesses.
Please don’t hesitate to reach out with any questions about AI in local news, starting your own AI journey or how you can support local news ecosystems.
