AI Chatbots Assist With Central Oregon Government Work

Local governments in Central Oregon have rolled out artificial intelligence chatbots that they say have saved hours of staff time, help in combing through documents, synthesizing emails, generating content for press releases and even rewriting website language.

Governments across the country face the question of how — or whether — AI could interact with the public to help make services more efficient, even while they grapple with potential risks of the new technology. That’s spurred a region-wide organized discussion in Central Oregon that could lead to new state and local policies on AI oversight.

The City of Bend began the process of using AI about two years ago, along with the rise in popularity of Chat GPT, said Adam Young, information security manager with the City.

“Right when Chat GPT blew up, we realized, employees are using this, we need to get ahead and start figuring out how we’re going to do it safely,” Young told the Source.

Chat GPT is a type of generative AI chatbot, “trained” with information on the internet to create text, image or video responses to prompts.

But Bend’s new AI program restricts employee use to one chatbot in particular: Microsoft 365 Copilot, which can draft or summarize emails, analyze data and create presentations.

“It’s just been a real productivity boost for a lot of the employees,” Young said.

Meanwhile, Deschutes County has adopted two different chatbots for staff use, mostly for administrative research and data analysis, said spokesperson Kim Katchur. That came alongside an AI policy adopted in 2024 that guides how the tools are used.

Bend adopted a similar policy in 2025, alongside a new steering committee on internal AI policy and trainings to help recognize when AI was providing biased or inaccurate information.

According to Young, about 400 of Bend’s 800 employees have asked to have Copilot on their computers. For about 60 of those employees, the chatbot is integrated with the suite of Microsoft Office apps. That way, it can form answers with information not only from the internet, but from an employee’s documents, calendars, spreadsheets and more.

Part of the reason Bend chose Copilot was because of its security measures, Young said. It blocks sensitive information like social security numbers and criminal justice data.

When to disclose

Bend’s policy says that staff should always fact check and validate work created with generative AI, and disclose when “AI is used to create an image or a document from scratch, or is used to significantly modify one.”

That can be a fine line.

Rene Mitchell, communication & engagement director with the City, said her department has used Copilot for anything from formatting information into a spreadsheets to summarizing community input to generating icebreaker prompts for meetings and coming up with concepts for press releases.

Copilot also played a role in reshaping language on the City’s website. During a recent revamp of its website, the City told Copilot to help “adjust copy” — or rewrite — complicated or technical information at an eighth grade reading level, Mitchell said. Copilot’s copy was reviewed by subject-matter-expert staff for accuracy.

“In every case, staff did not accept verbatim what Copilot produced/recommended,” Mitchell said in an email.

Copilot’s role isn’t disclosed on the City’s website.

Mitchell’s department also uses Copilot for “support in generating concepts and content for social media and press releases.” Still, the City hasn’t made it clear when press releases included content generated by AI.

Mitchell said that’s justified based on the policy.

“The policy specifies that ‘not all uses of AI require citation,’ and in the case of its use in helping generate concepts and content for social media, press release, or the website, all material produced is edited and reshaped substantively by human hand,” Mitchell wrote in an email.

According to Young, Bend’s AI policy relies heavily on guidelines from cities that are part of the GovAI Coalition, a group of several hundred agencies formed by the City of San Jose to promote “responsible and purposeful AI in the public sector.”

How local governments should be using AI might be the focus of a region-wide civic assembly on AI hosted this fall, said Josh Burgess, director of the Central Oregon Civic Action Project. Governance and decision-making was a top issue in a recent poll conducted by the project, with the vast majority of respondents agreeing that governments should disclose when and how they’re using AI and justify the cost to taxpayers. In September, the project will randomly select people across Central Oregon to be delegates to an assembly that will come up with a set of recommendations for policymakers.

Reporting from Cascade PBS has showed government officials in Washington state have relied increasingly on AI chatbots to draft letters, write emails to constituents and assist with policy reports. That showed both the potential utility of AI in government work but also the technology’s tendency to “hallucinate” or create factually inaccurate information.

Bend Mayor Melanie Kebler told the Source she doesn’t use Copilot for City business.

“I think AI tools can be useful in certain contexts, and can help city employees work more efficiently, within proper policies about when AI tool use is appropriate or not,” Kebler said in an email.

What’s next

While AI chatbots are now a widespread tool for local government staff, agencies aren’t using AI tools that the public would interact with — yet.

Katchur, the Deschutes County spokesperson, said the County is working toward adding a chatbot to its website that could help people get answers faster and find the services they need.

Young, who leads Bend’s AI efforts, said the City plans to look at “low risk, high value” areas to deploy public-facing AI. For example, other cities have used AI to speed up permitting times by helping developers shape their plans in advance of submitting their applications.

Those types of tools will take vetting, Young said.

“We want to be very, very cautious of, if the public’s going to interact with our AI solution, it’s extremely tied down on making sure the information fed back to that person is accurate,” he said.

This article appears in the Source June 25, 2026.