Guide

AI in local news: A practical list of dos and don'ts

How local news organizations can adopt AI while maintaining trust, judgment, and their sense of purpose.

October 19, 2026

Members of the Collaborative at an August 2026 convening. Photo: Hannah Yoon

This article is part of the Lenfest AI Collaborative and Fellowship Program’s 2026 Impact Report.

Artificial intelligence is rapidly becoming part of every newsroom. The question for local news organizations is no longer simply whether to use AI, but how to use it well.

We asked the Lenfest AI Fellows and their managers to help us put together a playbook for local news organizations, based on their experiences and learnings over the past two years. They came up with practical suggestions for what to do – and also strong recommendations on what not to do.

One underlying theme: Local news organizations should approach AI as an organizational capability, not simply a collection of tools. The goal is to use AI where it helps the organization do better journalism, serve its community and build a sustainable business.

Do this

Set policies, guardrails and governance

Develop clear, flexible policies and standards on AI use.

Create an equal balance between guidance and guardrails.
Establish the right way to get data sources as soon as possible. Create clear governance processes – and share transparently with your teams.
Default to approving requests early on to learn about alternative platforms. Try a task with different models, since each has its own strengths and weaknesses.
Over-specify at the beginning. An extra hour spent planning and articulating pays huge dividends.

Set realistic goals and expectations for reader-facing AI projects. “It’s not magic,” one participant added.
“Keep a human in the lead, not just in the loop.” “Intentionally set human intervention points.”

Culture

Work on the culture first.
Establish clear communication with stakeholders.
Find champions internally. Identify people in each department who would be happy to share knowledge and support their peers.
Require top leaders to lean in.
Meet people where they are.
Try vibe-coding with anyone in your organization.
Liaise with skeptics – and being kind and open to them.

Encourage experimentation and sharing. Set aside time to explore different AI tools, and “be open to creative chaos” that lets staff explore. Get leaders involved in experimentation.
Start small, get quick wins early, and show wins. Any learning is a win. Experiment in small, iterative steps so that you can learn quickly from results.
Create a process to share use cases regularly, like a demo camp every six weeks. “Nudge and prod people to share what they have built, especially if they were working alone.”
“Moving from prototype to a scalable tool is the hardest part. Account for that boring part.”
“Get the newsroom involved early and enthusiastically rooting for your AI project.”
Collaborate with other organizations on similar projects.

Where to use AI

Start with the problem, and don’t do AI for the sake of AI. “Don’t just sprinkle magical AI dust on a problem to say that you used AI. I call that performative innovation. It drives me crazy,” one participant said.
“Drive and optimize for business outcomes and team priorities – not from a ‘how should I tackle this with AI’ perspective.” Identify a genuine business need first and then ask whether using AI is the best way forward.

“Let AI do the drudgery.” Let humans do the value-added work that requires sensitivity and judgment.
AI works best “if you point it at a tight corpus of your own reporting. When we just ask it to summarize strictly within the bounds of the article, our testing showed that it was roughly 97% accurate.” LLMs can be highly accurate when supplied with high-quality context, such as reporting, documents, or other materials, as opposed to only asking them to reason over what’s in their training data.

Don’t do this

Don’t build AI tools in isolation

Don’t silo product and tech from the newsroom on AI projects. Engineering in a vacuum is a problem. Let your AI person work their way through your organization, enable and empower staff, and build relationships to ensure that their tools actually create an impact. Co-creation is ideal to develop the best solution and to get user buy-in.
Don’t give staff access to new tools without some training or guidance. “We have a responsibility of making sure that nobody is left behind in terms of AI skill. We need to be focusing on training them and getting them to some level where they are comfortable.”
“You shouldn’t assume you know where and how adoption is happening.”
“Don’t make employees feel unsafe by tracking and logging every single activity.”

Don’t make technology choices without a strategy

Don’t treat AI companies as your friend – or your enemy. Evaluate them objectively and repeatedly.
“Don’t jump to vendor point solutions. The technology is evolving too quickly.”
“Don’t just use the most powerful models.”
“Don’t take overwritten code as gospel. Sometimes, saying ‘no, not like that,’ is the best move.”
Don’t overestimate the capabilities of AI or remove human judgment or let the AI think for you.

Don’t wait for certainty before experimenting

Don’t expect a clear, demonstrable ROI right away. Don’t build without cost and revenue metrics and impact analysis in mind, but do remember: Adoption takes time, and using the tools is often the best way to find out whether they are genuinely useful or not.
“Don’t treat every problem as one size fits all.” Don’t automate before you understand the problem.
“Don’t delay getting your AI experiments in front of real readers. The perfect is the enemy of the good.”

Don’t let governance become bureaucracy

Don’t create “iron laws,” such as ensuring a human must be in the loop 100 percent of the time. Acknowledge that the level of risk differs according to the AI use case – for example, between internal newsroom assistance and producing content for public consumption. Be prepared to revisit your AI policies and guidelines as the technology evolves.
Don’t over constrain a model or apply guardrails and constraints too soon. You risk putting a freeze on experimentation without learning what the machine can actually do.
“Do not make it harder for employees to learn, experiment, and explore AI. It is a much-needed skill for their future.” As one participant said: “I think it’s absolutely our responsibility as leaders to bring people along on these changes, even when it’s uncomfortable. That’s the job.”