AI Use Disclosures

As the creator of a podcast that involves AI, AI cloning, and AI agents, I get a lot of questions about my general AI use. Below you’ll find a detailed AI disclosure, which I intend both as a reference and as a kind of public benchmark for myself. (If you want to discuss or debate my approach, I’m always happy to show up on podcasts, at conferences, or in journalism classes to discuss, so drop me a line.)

TLDR: I don’t use generative AI for any of my writing. I do use it for research and transcription, and I deploy AI agents as subjects in particular stories about AI.

How I do and don’t use AI as a journalist

Writing: I do not use generative AI for any of my writing or editing.

This prohibition includes using AI to generate any portion of any words I write, placing any of my writing into an AI tool for editing or review, or using AI-enhancement writing features within any tool. It also includes brainstorming ideas for my writing, or using AI to get unstuck on a particular writing challenge. 

If you read or hear a piece of journalism I have written and published—whether an article, a podcast script, or a book—AI has not been involved in its writing, and none of what you are consuming is AI-generated. (Excepting the disclosed use of AI clones and AI agents as story subjects, outlined below.) I don’t use AI for interview prep. If I am interviewing, say, a fellow writer and I describe having read their work, I have read it. I have not had an AI read it and summarize it for me, or suggest questions.

If you believe you’ve detected AI use in my writing—say, because I am and have always been a prolific user of em dashes—I am a bit insulted but you are wrong.

Research and data analysis: I do use AI tools—both off-the-shelf and some I have created on my own and with colleagues—for information gathering, data analysis, and source identification.

For example: I’ve used Notebook LM to analyze a large corpus of court documents looking for particular facts, sources, and connections. I’m co-developing a multi-agent program to both gather and analyze reporting for a large project I’ve undertaken. None of the text outputs from these tools are included in my stories or scripts themselves, and the results are used as a jumping off point for my own reporting and fact-checking, rather than being incorporated directly into my work. 

Transcribing: I do use AI tools to transcribe interview and other recordings I have made in my reporting.

All quotes are rechecked with the tape before appearing in a story, as I would with any transcription.

AI Agents as subjects: I do create and deploy AI agents in order to tell immersive, reported, satirical stories in my podcast Shell Game.

These agents may have voices, visual representations, and video avatars—which might be clones of myself or purely AI-generated—and the ability to operate independently. I use many different AI tools to create and launch them. If you hear them in the show, come across them in the the world at large, or communicate with them in any form, their outputs are AI generated. 

This use of AI is always clearly disclosed in the show itself, indeed it is a primary topic of the show. I haven’t always disclosed it at the moment the AI agents are deployed—for example, when I had my own AI clone interact with a scammer, or call a friend. Sometimes that’s because I’m seeking people’s unfiltered responses to interacting with AI. Other times, it’s because the AI agent messes up the disclosure. The ethics of these choices are also a regular subject of the show and my discussions around it.

Why I do and don’t use AI in these ways

Writing

This to me is a simple one: I feel very fortunate to have made a living as a writer and editor in longform journalism for two decades, and I enjoy it. I love words and take satisfaction from coming up with them, ordering them, writing them, speaking them. I work almost exclusively in narrative journalism (magazine features, podcasts, books), a form I gravitated to precisely because it enables creative approaches and values the writer’s voice. I aspire to write as well as the writers I love to read, and as I always fall short of that, to write as well as my talents allow. Using AI would deprive me of the satisfaction I gain from that pursuit.

In fact, knowing I’m able to fight through the hard parts of the writing process is what keeps me from defeatism in the long lonely hours. Besides risking that a certain type of AI writing style would infect my own, using generative AI in the writing process feels like it would weaken my ability to think through stories and their complications, and thus my well-being. I owe it to readers to be able to stand behind every word in a story, but honestly it’s as much about my own needs. 

When it comes to AI as an editor: I love working with experienced editors, producers, copy editors, fact checkers, whenever I can. I value their discernment and do not view AI as an upgrade, or even a useful substitute, to any aspect of those relationships. I’m always aiming to get better, but I am not aiming for greater efficiency—a quality I generally find to be overrated, particularly in creative pursuits.

None of this is necessarily a value judgement on how LLM’s write, whether they are improving, or whether you are somehow a bad person for using them for your own writing or editing. I have my opinions, but my reasons for not using them in this fashion are tied to my feelings about my own work.

Research and data analysis

As with writing, I enjoy reporting. While Shell Game listeners will know that I dispatched an AI clone of myself to conduct interviews in Season 1, I’m not actually looking to replace that part of my job. That said, there are clearly powerful applications of AI when it comes to analyzing data and large volumes of research materials—or in some cases, even obtaining those materials. I will continue to do that work manually as well, as I find there are also connections AI can’t make. But I’ve found the ability to query and parse a large collection of research using an LLM already has enormous advantages over keyword search. I would think any newsroom would benefit from tools that find more connections and stories in their reporting, and hopefully they are also building them. (A note on the resource consumption of AI: I place my AI use in same category as flying, driving, or utilizing other types of digital technology for reporting. I have my own thoughts on the larger questions around this, as well as the copyright aspects of AI training. But I use this technology as I do others, when I feel its warranted.)

Transcribing

I generate a huge amount of tape for Shell Game and other stories. AI transcription has reached a very high level of quality, and I save a lot of time using it. (I generally use MacWhisper, for security reasons.) It’s worth noting that back when I had to do all my own transcribing, I thought of it as “a crucial part of my process.” So log this one as an example of how stances on these things can evolve. 

AI Agents as subjects and characters

This is obviously the bread and butter of Shell Game, for which we have created AI agents to be subjects of our own slightly-deranged plots and investigations. (Again, we do not use generative AI in the creation process of the episodes themselves, e.g. writing, editing, producing, music, illustration.) I believe some types of stories benefit from a kind of full immersion, and AI is one of them. The team behind Shell Game has been trying to understand and illustrate what AI’s capabilities, test the claims about it, raise questions about social boundaries around it, and explore how the sudden expansion of AI is making the world feel. To do all that, it’s important to get our hands into it, which includes everything from springing an AI clone of me on unsuspecting friends, to launching a company with AI employees, to launching an AI podcast hosted by those AI employees, to sending one of those employees to give a talk to LinkedIn. 

These plotlines are not recommendations. As a rule, the show does not endorse nor condemn any particular AI use case. Rather we try to tell stories that bring these possible cases into the light for examination. We also think it’s important to keep in mind how strange and funny it all is—this sudden appearance of human impersonators in our midst—even as it’s also serious and foreboding. You can go listen to the results and judge whether we get the balance right.

First published: 05/19/2016. Disclosures last updated: 05/19/2026