AI can be valuable tool for cities to enhance outreach, communication

By KATE COIL

TT&C Assistant Editor

At a time when it is more important than ever for municipalities to clearly communicate with citizens, officials can utilize artificial intelligence (AI) to support and improve their outreach.  

Kathryn Good, president and founder of Gallatin-based Good Circle Marketing, highlighted ways municipalities can use AI to enhance their communications during TML’s 85th Annual Conference in Chattanooga. Rather than rely on AI to do all of the work, Good said officials should approach it as a communication partner and a tool to improve what already exists. 

FINDING YOUR VOICE 

To begin this process, Good recommends training AI to match the “voice” or style the city needs. The first step is to use real content examples written by a human 

“You can take at least five pieces of content that were not written by AI, like press releases, emails, social media content that has performed well, letters, or the transcript of a speech by the mayor and put it into the AI,” Good said. “You then give it a prompt like ‘help me determine my voice, style, and tone.’ You then take those results into the customized instructions. That prompt should be the standard you set.” 

Users can further lock in their city’s voice through preferred terms, words, phrases, and instructions that should be applied with every response. Good said that giving feedback to the AI is essential with editing prompts and responses providing a more accurate voice over time. 

“The best way to train AI is to actually give it examples of the way you want it to behave, 
she said. “We are treating AI as an assistant. If the assistant we are using is not performing the way we want it to perform, we tend to shut the computer. Don’t shut your computer; go back and edit the prompt. It is not behaving the way you want it to because you haven’t given it enough training material. When it gets something right, tell it. When it gets something wrong, tell it you why you don’t like it.” 

Good said it is also essential to set guardrails for what the AI shouldn’t say, such as outdated terms, sensitive topics, and local issues to avoid.  

PICKING PROMPTS 

Often, the response of AI is only as good as what humans put into it. Good outlined six steps for creating strong AI prompts: Context, Instructions, Role, Constraints, Language, and Evaluating and Editing. 

For context, Good said prompters should set the scene by providing background and establishing goals they want to achieve.  

“We want to bring the AI assistant up-to-date on what is going on,” she said. “Then we want to give the instructions or make a request, whether you want to create social media content or a press release. Be very specific about what you want the assistant to create.” 

For the role, Good said you want to define the lens the AI uses as its persona, such as a legal or communications expert. Making the role more specific can lead to more accurate rules. However, Good cautions that AI can still make mistakes – even when best practices are followed. 

“We then want to set the constraints or rules, and I can tell you right now ChatGPT cannot count,” she said. “If you ask for something in 50 words or less, it will estimate. If you want to fit something into a social media post, it doesn’t always work. I will often say follow social media best practices or follow Facebook best practices instead to get that format. You can also instruct it to give the information in a table format or bullet points.” 

For those who have already built a voice into the AI engine with previous examples of the work and use, Good recommends specifying “use my voice” to set the tone or voice. Otherwise, prompting is a valuable tool. 

“You can also say ‘I want it to resonate with my audience’ or ‘write this in a way a five-year-old would understand’ to make it simple language,” she said. “You want to be specific in language and tone. Once you put the prompt in, you want to go back and edit it, review it, make sure it sounds like you, and make sure it is accomplishing everything you want. If not, you may want to edit the prompt or give it feedback.” 

When a prompt is successful, Good recommends saving it to build a prompt library. Saving the prompts that give the most effective results can help generate high-quality results in the future. 

WAYS TO USE AI 

There are numerous ways cities can begin to explore AI messaging on their own, beginning with generating first drafts of emails to citizens, social media posts, press releases, and emergency alerts.  

“Let’s say you want to sit down with your mayor and get them on camera,” Good said. “You can start with AI by indicating ‘I want to interview our mayor on these particular subjects. Act as an expert interviewer and give me engaging questions.’” 

Good said AI can transform lengthy meeting notes or complex policies into concise bullet points and plain language or adjust content for clarity, tone, and audience appropriateness while maintaining key messages. Another suggestion is to use AI to convert one piece of content – such as minutes or recordings of a city council meeting – and adapt it for multiple uses like press releases, social media posts, and email summaries.  

AI can also be used to convert communications into multiple languages.  

“In all of these things, we still need to bring a human into it,” Good said. “If you ask AI to translate something into Spanish, you should have someone who speaks Spanish review it to ensure it is correct and that the cultural meanings are the same. We can’t just put something out there without the right person reviewing it.” 

STAYING ON MESSAGE 

AI is far from perfect, and Good said there are several things to consider before sending AI work out into the world. The first step is to always fact check elements like figures and dates, as well as legal information. Good also encouraged having in-house experts review AI- generated work before it is presented to the public, such as having an HR director review AI-generated job descriptions or a lawyer look over policy documents. 

“Hallucinations are when AI makes stuff up, and it happens,” Good said. “It will sound so right, but it’s not. I see it a lot when I ask it to analyze the content of a website. It will come back, and I will be able to tell it didn’t analyze the entire content of the site. You always want to fact check. My trick on this is that you want to read it for sure, but if I generate it on ChatGPT I will also take it to other AI like Gemini or Claude and check it against another engine.” 

Users should also ensure sensitive information is never entered into AI to protect that data. 

“What you input to AI can be used for training,” Good said. “We want to use AI as an assistant, not a replacement for human judgement.” 

The goal is to preserve the city’s unique voice with AI supplementing rather than overtaking the authentic tone, style, and brand. Good also advised to be cautious with any visual content. 

“As a brand, you want to represent your city well. The best way to do that is with pictures of your buildings, your residents, and your government officials. You can use AI to correct some of that or for brainstorming, but I don’t recommend using it for true image generation.” 

Before sending out AI-enhanced communications, Good recommends starting with low-risk content, such as internal messages or announcements to build confidence before tackling more sensitive content to be seen by the public. During this phase, establish clear workflow for reviewing AI-generated content, such as checklists for fact verification and quality control. The team using these communications should also be trained in both the capabilities and liabilities of using AI tools. This may include creating shared guidelines or policy of AI use by the city.  

"Always review before publishing, Good said. "You have to remain the human in the loop with all AI-generated content, reviewing for accuracy, appropriateness, and your city’s voice. Measure your results. Especially as you go forward and encourage your teams to use AI, you want to be able say ‘I’ve saved 10 hours here.’ This is a revolution, but it is going to take some time for adoption. We need to show time savings. I want you to play. Have fun and get creative. Have one AI tool that you use and use it every day."