There’s a lot of chatter right now about an AI bubble, fueled in part by a perception that productivity gains from AI are largely illusory. I can’t speak to the market, and whether AI is broadly overvalued or undervalued, but I can tell you that in the past year alone, AI has completely transformed how I work.

Looking at the tools today that didn’t exist a year ago—deep research, browser agents, the big leaps in performance for all the latest models—there’s a host of ways AI can speed up or enhance many tasks of knowledge workers, especially journalists. As an independent journalist, I’m perhaps a little less constrained than most (my AI policy is whatever I want it to be), and AI has made me rethink how every part of the job gets done.

So bubble or not, that shift is already here. To better articulate this and highlight tools and techniques that are broadly useful, I’ve broken down several ways I use AI in my writing, researching, and reporting. 

The most intuitive way to organize this is by walking through the story process—from developing ideas to hitting publish to sharing on social media. Here are 10 ways I use AI as a journalist and content creator:

1. Beat monitoring: The set of stories that any individual reporter needs to keep up with is different, and AI’s ability to personalize and summarize is extremely helpful here. I find ChatGPT Pulse—which doesn’t just compile stories, but explains why they’re relevant to your work and includes other personal info (like your schedule)—to be extremely helpful, but it’s also constrained to the Pro plan ($200/month). A stripped-down, but still useful tool is scheduled tasks in ChatGPT or Perplexity, where you craft a prompt that searches for the latest stories you’re interested in and sends it to your inbox.

2. Accelerated skimming: Once I land on an article that I’m interested in, I often read the summary before deciding if I want to spend the time reading. There’s a dedicated button for this in Perplexity’s Comet browser, and there are a host of browser extensions that do this for Chrome. On X, I find the ability to simply reply to any post with “@grok what is this post about?” for a short, instant explanation to be a huge time-saver, especially for memes and trends I’m seeing for the first time.

3. Going deeper: When Google’s NotebookLM made its big splash last fall with its ability to create instant conversational podcasts, many dismissed it as a gimmick. But I find the feature to be an incredibly useful tool for getting primed on a topic or news story. You can either drop a single URL in the folder or use the new Fast Research feature to find a story, generate your audio overview, and boom—you have a short podcast all about the thing you’re researching. Listen at 1.5x speed to blaze through it even faster.

4. Story ideas: Finding the connections and missing angles in between stories will always be a mostly human-driven process, but NotebookLM is a good partner here as well. Prompting it to suggest story ideas based on the questions implied or not answered in the set of articles in the notebook is often a great starting point for a story idea.

5. Getting in the weeds: Once I know the idea I’m running with, it’s time to research. AI is obviously great at this, specifically the deep research tools that have emerged in the past year. I use them all, but differently: Perplexity is excellent for a first pass—it’s fast, but typically not as thorough as the others. I generally find Gemini to be the best at finding sources on the web, but less great at prioritizing them. My go-to for serious research is ChatGPT, which I find to be the most thorough. It’s also excellent for turning its deep research abilities inward—directing them at a large set of files in a Google Drive or Dropbox folder (via Connectors).

6. Targeting highly specific information: When the key to your story is a singular document or piece of data, it’s helpful to outsource the task of finding it to a browser agent. For example, court documents are typically kept in hard-to-navigate services like PACER (Public Access to Court Electronic Records). Just finding the right case often involves performing several searches, especially if you’re unsure of the exact names involved or the specific court. When given clear tasks like this, a browser agent like Comet or ChatGPT Atlas is the perfect research intern, usually finding exactly what you’re looking for in just a few minutes.

7. Writing coach: Now that I’ve gathered and processed all the background material, it’s time to write. For my columns and original reporting, I don’t let AI write for me, but I do often use it as a coach. I’ve crafted a Custom GPT to act as an inquisitive interviewer: probing me with several questions—verbally—recording my long responses, and then assembling them all into an outline once I’m done. From there, I’m off to the races, sometimes returning to the coach when I go in an unexpected direction. The process helps me write in about half the time as before.

8. Writing intern: I write a news digest for my Thursday newsletter, with each item coauthored by AI. For these short blurbs, I’ve built a Claude Project that’s trained on my style, the digest format, and the target audience. Once given a story or news trend, it writes a one-paragraph summary in my style, which I then edit and add to. It turns a 15-minute process into something that takes about five minutes. (And, yes, I’m fully transparent to my audience when AI acts as a coauthor.)

9. Copy desk: Everything I write goes through my AI copy desk: First, a Custom GPT looks at it critically but constructively, suggesting edits (and sometimes further research) to make the piece stronger. Then another GPT does a proofreading pass, aimed at making it either more newsy or conversational, depending on the context. Importantly, neither GPT does automatic rewrites—they’re all suggestions that I can take or leave.

10. Social media manager: Finally, once the piece is live, I have specific prompting that writes social copy for LinkedIn—both for my personal profile and The Media Copilot company page. In this case, I don’t manually paste anything into an assistant on the web, but instead use Zapier, an automation tool that automatically generates the social post and queues it up for my review. That saves a lot of back-and-forth between browser windows and ensures I never miss a post.

I hope you see a pattern in what I’ve laid out: Each process involves AI bringing things to my attention, but it prioritizes my attention. In my view, AI is an extraordinary accelerant for specific tasks in story creation, but the process needs human review and judgment throughout. And while AI can be a helpful writing “intern” for very specific formats, I keep the text in anything substantive (such as this column) human written.

That human centricity will remain, no matter how intelligent and automated parts of the process get. While I don’t doubt that AI will continue to improve, the point of using it is to sharpen, accelerate, and amplify how I communicate with audiences—human audiences. Machines can be great partners, but the moment they become the focus is when we stop communicating and are just creating “content.” The media just experienced an era dominated by SEO, algorithms, and lowest-common denominator thinking. Let’s not do that again.