AI Co-Host Workshop

Maya: Before AI gets a microphone, what is this show for? Diego: If it’s, uh, everyone, your concept is soup. Maya: Welcome to AI Co-Host Workshop. Today, tight concepts, AI roles, ethics. Diego: Podcast Index finds 39% of new podcasts are likely AI-generated. Deliberate beats shiny. Maya: Choose one listener, problem, and promise. Give AI one job, host leads, co-host challenges, narrator bridges. NotebookLM can synthesize audio from documents, but verify them. Maya: Let’s write a sixty-second intro. Live: Welcome to Midnight Meals, for night-shift workers needing dinner on a weird Tuesday. Diego: Generated AI co-host: I’m Pantry. I turn your schedule, budget, and fridge scraps into one realistic meal plan. Maya: Live: Each week, one recipe, one shortcut, and a shift-worker question. We link sources, say what we tried, skip miracle claims. Diego: Generated: If I’m unsure, I’ll tell you. I’m a co-host, not a mysterious casserole oracle. Maya: Live: Tonight, a fifteen-minute lentil bowl and a rescue mission for lonely spinach. Diego: Generated: Grab a pan, keep expectations humane, and let’s make the next shift easier. Maya: Live: This is Midnight Meals. Let’s eat. Diego: Pantry has one lane, practical planning. It doesn’t run the whole kitchen. Maya: Your six checks: consent for voices and materials, source transparency, privacy, AI disclosure, inclusive language, and no harmful or discriminatory language. Diego: Review them. AI can echo blind spots, and overreliance makes a bubble. Maya: Assignment: name your show, choose its AI role, draft those six checks. Diego: Next time, scripts and workflow. Bring your robot responsibly. Maya: See you then.

Maya: Last time, in “Concept, Role & Ethics,” we gave your AI co-host a real job, not a borrowed human identity, which, honestly, saves trouble later. Diego: Right, and today the trouble is, uh, twelve tabs open, one synthetic “hello,” and no episode. Very modern. Maya: So, quick chain. Start with a source brief, then a prompt that names the audience, the segment goal, and what the host must not invent. I’ll use ChatGPT, or another drafting option, to shape a first script, but I keep the source beside it. Diego: Because “make it engaging” is not a prompt. That’s a wish you throw into a fountain. Maya: Exactly. Then audition a voice model for clarity, pace, and fit, not because it sounds like somebody famous. ElevenLabs is one option creators may test. Keep two takes, even if, you know, take one feels charming. Diego: Charming is how a five-minute intro becomes a hostage situation. Maya: Here’s a tiny source document: “This show helps freelance designers price projects with confidence. This episode explains a three-question estimate.” Diego: Okay, cue the generated thirty-second AI-hosted clip... “If pricing feels awkward, start here: what outcome does the client need, what work creates it, and what changes if the scope grows? Those questions make your estimate clearer.” Maya: That’s the move. With NotebookLM, you can use documents as the workflow starting point for audio, then check the result against those documents. Don’t publish the first pass. Diego: Then basic edit, cut repeats, fix weird phrasing, check names and numbers, tighten pauses. PodcastorAI or PodPal can be options, not magic wands. Maya: For deeper practice, pick one path, like relevant Udemy tracks, Calimax, or Academia IA, instead of collecting courses like browser bookmarks. Diego: Next time, we launch it without hurling it into the void. Maya: Ha, yes. Bring your draft, and we’ll make the release plan feel doable.

Maya: Last time, in Toolchain & Quick Production, we built a minimal AI podcast pipeline and made a short segment you could ship. Diego: Yeah, and now comes the part where nobody gets to hide behind a very polished robot voice. Maya: Exactly. A voice can sound warm and still leave listeners asking, wait, who said that, and can I trust it? Diego: Especially with health, money, grief, or, uh, somebody’s real life. I want a human there, not Smooth Uncle Bot. Maya: Ha, fair. Human hosts matter most when nuance and accountability are the product. AI can help, but no cloning without consent, no hidden synthetic speaker, and no fuzzy sourcing. Maya: Okay, tiny on-air demo. Here’s a twenty-second disclaimer, plain human copy: “This episode includes an AI-generated co-host voice. We created the script, checked the claims, and approved the final audio. That voice does not represent a real person, and if we make a mistake, we will explain and correct it publicly.” Diego: And here’s the AI-rendered version, same exact copy: “This episode includes an AI-generated co-host voice. We created the script, checked the claims, and approved the final audio. That voice does not represent a real person, and if we make a mistake, we will explain and correct it publicly.” Maya: Same facts, different feel, right? The human read can feel like a personal promise. The AI read can work, but only if disclosure is easy to find. Diego: So, uh, don’t bury it in episode notes nobody opens. Maya: Before launch, write a public transparency statement, check every claim, then give ten listeners both versions in random order. Ask what felt clear, trustworthy, or confusing. Diego: All right, go test it, okay? See you next time.