AI Should Be Your Helper, Not Your Doer

More and more radio stations are asking me about AI tools. Some are asking about ours, including AI Content Helper Pro, while others are experimenting with ChatGPT and just about every other AI product that comes along. I understand the interest. Most stations are already asking a small staff to do too much, and if AI can help turn a transcript into an article, clean up a press release, write a better headline or get a newsletter started, that’s useful.

What worries me is when the “helper” quietly becomes the “doer.”

There is a big difference between asking AI to help you create content and trusting it to create, fact-check and publish that content for you. AI can take information you give it and produce a pretty impressive article in seconds. It can also get an event time wrong, misspell the name of a local business, misunderstand something said during an interview or add a detail that was never in the original material. Worse, it can do all of that while sounding completely confident.

That’s why I’ve always viewed AI-generated content as a first draft. Even with our own AI Content Helper Pro, the idea is right there in the name: helper. It is there to help someone at the station get from a blank screen to something useful much faster. It shouldn’t replace the person who knows the community, understands the promotion or recognizes that Highway 25 isn’t the same thing as Highway 35 or I-25.

The amount of checking you do should depend on what you’re publishing. If AI cleans up a personality’s show notes or turns an approved station announcement into a web post, reviewing it may take less than a minute. Read it, make sure it sounds right, check the names and details and publish it. AI has probably saved you ten or fifteen minutes of writing without creating another ten minutes of work.

Other content deserves a closer look. Events, contests, school closings, community information and promotions are full of details that matter. If the website says registration ends Friday at 5:00, somebody needs to know where that 5:00 came from. If the original rules say 6:00, it doesn’t matter how well-written the AI version is. It’s wrong.

News raises the stakes even higher. Crime stories, accidents, court cases, health information, politics and anything that could affect someone’s reputation shouldn’t be treated as content that AI can simply “handle.” Use it to organize information. Use it to help turn a transcript into something readable. Use it to suggest a headline or help you get through a mountain of information faster. But the facts still need to come from the source, and a human being at the station still needs to take responsibility for what goes online.

I’ve seen the same potential problem with advertiser content. Give AI some rough notes about a car dealer’s promotion and it can turn them into beautiful copy. In the process, it might also turn “financing available” into “everyone approved,” change the terms of an offer or casually add something that sounds great but that the advertiser never actually promised. The finished copy can sound so polished that nobody thinks to question it.

That’s actually one of the things that makes AI mistakes dangerous. Bad AI copy isn’t always obviously bad. Sometimes it’s excellent copy containing one very wrong sentence.

Transcripts are another good example. AI is incredibly useful for taking a 15-minute interview and helping turn it into an article. But anyone who has worked with transcripts knows what happens with local names, street names, schools and businesses. Throw in a bad phone connection, background noise or somebody talking over somebody else and things can get interesting quickly. If you’re turning an interview into a news story, somebody still needs to compare the important details and quotes with what was actually said.

I don’t think stations need a giant AI policy manual to deal with this. Most probably need something much simpler: AI can help create the draft. A person at the station approves what gets published.

Who that person is can depend on the content. Promotions should own the important details of a contest. News should verify news. Sales or the advertiser should verify offers and claims in sponsored content. The person putting something on the website shouldn’t have to wonder whether somebody else already checked it. “I assumed they did” is how the wrong information winds up online.

I’d also be especially careful about automating the last step. There’s a huge difference between AI creating a draft in WordPress for someone to review and AI creating an article and immediately publishing it. The first saves time. The second removes the one person in the process who may recognize that something doesn’t look right. That’s why I’ve resisted the idea that the ultimate goal of AI is to remove people from the content process.

We’re certainly not the only ones trying to figure out where that line belongs. The Media Copilot recently highlighted how some of the country’s largest news organizations are approaching it, and the differences are interesting. The Los Angeles Times reportedly allows AI to assist with research and background for submitted opinion pieces, but draws the line at AI producing the copy itself. Meanwhile, The New York Times is testing AI-generated summaries in its website search results using only Times journalism as source material. According to Semafor, that’s notable because those summaries can reach readers without first being reviewed by a journalist or editor, and the Times Guild has raised concerns about exactly that lack of human oversight.

I think those examples are useful for a local radio station because even some of the largest newsrooms in the country haven’t decided that AI means it’s time to remove people from the process. They’re trying to decide where AI can help, where a human still needs to check the work, and ultimately who is responsible when something carrying the organization’s name reaches the audience.

There’s another side to this that I think matters just as much. The Wall Street Journal defended publishing an opinion column after its author, Stanley Druckenmiller, acknowledged using AI to help create it. His response gets pretty close to where I land on this: his name was on the piece and the message was his. AI involvement didn’t remove his ownership of what was published.

That’s a pretty good standard for radio, too. If your station’s logo is at the top of the page, your station owns what’s underneath it. It doesn’t really matter whether a reporter wrote every word, AI cleaned up the transcript or AI Content Helper Pro helped turn the original material into an article. Before you hit Publish, somebody needs to be willing to put the station’s name on it.

I think the better goal is to remove some of the tedious work from the people already doing it. Give someone a useful first draft instead of a blank screen. Turn a transcript into something they can edit. Help rewrite a poorly written press release. Suggest headlines. Resize the workload, not the responsibility.

And when something does get through, correct it everywhere it went. A wrong date on the website may have also made it into a newsletter, mobile app push notification or social post. Unlike a mistake that disappears after it goes over the air, a website mistake can keep showing up in search results long after someone thinks it has been fixed.

AI is going to become a bigger part of how radio stations create content. I think that’s a good thing, and we use it daily. AI helps with every one of our insight articles and podcasts. Rest assured though, the version you’re reading is 100% our words. We check every word and often change things we would never say or agree with. You’d be surprised how much we correct. However, it’s still better and faster than starting from a blank screen (we’ve been there).

Stations don’t have an abundance of extra people sitting around looking for something to do, and there are plenty of places where these tools can make a small staff much more productive. Just don’t promote the helper to doer and then leave the building. Your station still needs somebody behind the wheel.