How I Use AI as a Professional Writer

T here’s an ongoing debate about artificial intelligence: what it can do, what it should do, what harm it can cause, if and how to regulate it, and even whether or not it should exist.

Inside newsrooms, there are discussions about appropriate uses of AI in journalism. The stances range from “it is never acceptable” to “let’s use AI as if it is another employee.” For examples of the latter, note that the Washington Times and Business Insider both have an AI news desk, meaning a section of articles written by and attributed to AI.

I admittedly do use AI, albeit rather sparingly and for particular purposes. Here’s how and why:

1. Major search engines, including Google, have become completely useless. I can never seem to find the thing I’m looking for, so I’ve used Claude as a replacement for Google. The prompts go something like this: “Find proof of [X] and provide several sources.” Other times, my request goes like this: “[Public figure] said something to the effect of [paraphrase a particular quote]. Find an original source for that comment.” AI can also be great for broad searches. If I’m writing about a particular person or organization and want to know what they’ve said on a given issue, I’ll ask Claude: “What has [person/institution] publicly said on the issue of [something]?”
2. AI is good for finding somewhat obscure documents; I’ve asked AI to find court filings before, along with other sorts of business filings that don’t normally populate the results pages on search engines. If I’m doing a deep dive on a particular organization or institution, I’ve used AI to find large documents (like annual reports) that are sometimes buried on organizations’ websites.
3. When people or organizations get caught or called out by reporters, they may update websites (and sometimes do so without including a disclosure). When I suspect a website or document has been changed, I’ve asked AI to find archived versions. This usually doesn’t work, but it is still worth a try.
4. When I’ve finished an article, I sometimes plug it into AI and add the following request: “Review for typos, spelling mistakes, and any grammatical errors. Present any findings as a list.” It usually responds with a short list noting that I’ve missed a definite article somewhere, inserted an extra space in one spot, and/or accidentally used the British/Canadian spelling for a word. (The latter happens pretty often: My parents were both born and raised in Canada, and I lived in England for two years.) This prompt is particularly helpful when I’m speedily writing a breaking news story, since I hit the wrong keys as I’m trying to get the article out as quickly as possible.
5. Some journalists have a natural gift for coming up with snazzy headlines; I am not one of those people. On a handful of occasions, I’ve plugged an article into AI and said, “Generate ten different headlines for this article.” The suggestions usually aren’t exactly what I’m looking for, but I use the results as guidance. I tend to smush a few of them together to create what I think is best. In the end, though, my editor replaces it with something completely different and entirely better.
6. Reporters have to reach out to people for comment, but sometimes, the contact information for those people is not easily discoverable (possibly by design). In a few instances, I’ve asked AI to find the email or phone number for a particular public figure or expert, which works about half of the time. I have also prompted AI to find the press contact email addresses for some organizations, government units, businesses, and the like.
7. Every now and then, I have to work with a large data set or huge document, and in most cases, I’m just looking for one particular thing (a specific figure, a particular claim, a certain name, etc.). I have plugged those files into AI and asked it something like, “do a thorough scan and point out any references to [certain thing].” If nothing turns up, then I usually just forget about the document. If results do show up, then I tend to read the whole thing. (I don’t want to waste time reading hundreds of pages for something that isn’t there. Of course, there’s always the small chance AI missed something.)
8. If I’m working on a lengthy comprehensive article over the span of weeks or writing about a topic outside of my general area, I have plugged drafts into AI with the following prompt: “Use every source available and conduct a thorough fact-checking review.” This hasn’t been super helpful, but it might prove to be one day. (I can think of exactly one instance it was valuable: I had accidentally listed a previous job position for a person named in the article.) I like to think of myself as a detailed and thorough reporter, but when you’re working on a story for weeks, sometimes a detail you wrote ten days ago in the draft has been disproven by new stories, so it was seemingly accurate when you wrote it, but you missed the update.
9. Everyone who knows me in a professional capacity will tell you that I’m paranoid about plagiarism. Obviously, I would never directly cut-and-paste from another source, nor would I sneakily paraphrase and pass something off as my own. But I always worry that maybe a sentence I wrote is subconsciously drawn from something I heard years ago, or from something I scrolled past online. (As an example: I was writing an article about academic freedom, and I kept wondering if one sentence was inspired by something I heard at a free speech conference. Why? It seemed like too good an idea to have come from me. I couldn’t actually find any record of someone saying something similar, but I ended up cutting the line anyway, just to be safe.) On a few occasions, particularly when I’m writing about something that loads of people are also writing about, I’ve plugged a draft into AI and asked it to conduct a review for plagiarism. Even though I know I didn’t deliberately steal from someone else, I’m always worried that I accidentally wrote something that is suspiciously similar to what someone else has already said. It brings me psychological comfort that this type of prompt has never yielded any noteworthy results, yet I figure this is a good protective measure nonetheless.

I justify these uses of AI to myself, I think reasonably, on the grounds that I’m not engaging in epistemological outsourcing, since I am not recruiting AI to generate the intellectual content of my articles. I think I’m using it in the same way I would be helped by an editor, and how a more senior reporter might use an intern. But maybe I’m wrong. So, I’d be curious to hear from you all: Is it appropriate for journalists to engage with AI when producing their work, and if so, what might be some good applications that I’ve missed?