What is AI, in plain English?

The AI most people mean today is generative AI. You type a question or a request in plain words, and it writes a response. At work, it’s mostly a fast first pass at reading and writing: you give it your own document, say what you need, and check what comes back.

What staff actually ask AI to do

Most explanations of AI start with how it works. A more useful place to start is what people at work ask it to do.

Local government staff have typed more than 2.8 million real prompts into GovAI. Two patterns stand out. Half of all messages start with someone pasting in their own document: a staff report, a policy, a long email thread. And four requests sit in the top eight for almost every team, from the clerk’s office to public works: rewriting for tone, restructuring, proofreading, and drafting emails.

So at work, AI acts like a colleague who reads fast and writes a decent first draft of whatever you hand it. The document is yours. The judgment is yours. The first pass is the part it takes off your desk.

Source: GovAI requests from US and Canadian local governments, to August 2026. The shares describe what staff ask for, not results.

A road closure notice, start to finish

A public works coordinator has a road closure notice to get out by noon. The engineer’s version is accurate and hard to read: station numbers, contractor names, a paragraph about the traffic control plan.

She pastes it in and types: “Rewrite this road closure notice for residents. Use plain language, keep it under 150 words, and keep the closure dates and detour exactly the same.”

A few seconds later there’s a version a resident could read on a phone. She checks the dates and the detour against the original, fixes one street name, and sends it to communications. The writing took a minute. The checking took five, and the checking was the part only she could do.

What it’s good at

Writing and rewording: emails, notices, first drafts
Reading for you: summarizing long reports, policies, and meeting notes
Explaining: turning a complex topic into plain language
Thinking out loud: brainstorming options and next steps

How to get a better answer

You don’t need a special formula, but what you get back depends a lot on what you give it. A vague request usually gets a vague answer. Think about four things when you ask:

The task: what you want done. “Rewrite this,” “summarize this,” “compare these options,” “draft an email.”
The context: what it needs to know. Say what you’re working on and include the material it should use.
The audience: who it’s for. A resident, council, your department, or another staff member.
The output: what the answer should look like. Plain language, a certain tone, bullet points, a word limit.

Instead of “Rewrite this,” the road closure request above names all four. That’s why the first draft was close.

You don’t have to get it right the first time. Read the response and keep going: make it shorter, change the tone, tell it what it missed. Good AI use is usually a short back-and-forth.

How to tell when an answer is wrong

AI writes in a confident voice whether it’s right or not. It can invent a bylaw number, mix up two dates, or fill a gap with something that sounds plausible. A few habits catch most of it:

Check anything that would embarrass you if it were wrong: names, dates, dollar amounts, section numbers, and quotes.
Compare it with the source. If you gave it a document, ask it to say where each fact came from, then look.
Watch for answers with no source. Ask about a law or policy without giving it the document, and it may answer from general knowledge that doesn’t match your jurisdiction.
Ask it to find the weak spots. “What’s missing or unclear in this draft?” often surfaces a problem faster than rereading.

Try it this week

Pick three small things from your own work, using the AI tool your organization approves:

Something you rewrote last week. Paste in the original and ask for the version you ended up writing: shorter, plainer, or friendlier. Compare the two.
Something long you haven’t read yet. Ask for a five-bullet summary for your manager, then skim the original to see what it left out.
Something you’re about to send. Ask, “What would a resident find confusing in this?” and decide which points you agree with.

None of these needs training or a project. Each one shows you where AI is quick and where your own checking still does the real work.

What it isn’t

It doesn’t replace your judgment, your approvals, or your records process. In local government, what you send can end up in a council packet, a records request, or the local paper. A person reads the result before it’s used, every time.

It’s also not one tool. A free chatbot on your phone, the AI built into your office software, and an AI work platform your organization approves can all write a paragraph. They differ on who can see what you typed, how long it’s kept, and whether it fits how your organization works.

That’s the next question worth asking: What should staff never put into a public AI tool?