AI podcast generators vs. an AI co-host: what's the difference?
by Danny Abdeljabbar, founder of manekuThe short answer
An AI podcast generator makes a whole episode for you from a prompt, a document or a script. Nobody records anything, and the hosts are synthetic. An AI co-host joins a recording you are actually making, as one participant alongside real people, and speaks when you ask it to. A generator replaces the recording session. A co-host adds a voice to it.
I have co-hosted a podcast with my friend Henry for about a decade. When I tell people I now build an AI co-host, the first thing most of them picture is one of those tools where you paste in an article and a finished podcast comes out the other end.
That is a real category, and it has grown fast. It is also not what I built. The phrase "AI podcast" now covers two tools that point in opposite directions, so this post separates them.
What an AI podcast generator is
An AI podcast generator turns source material into a finished episode. You give it a topic, a prompt, a document or a script. It writes a conversation, assigns the lines to synthetic hosts, and renders audio or video. The whole thing happens without anyone sitting down to record.
Here are the best-known ones right now, described from their own sites and help pages:
- NotebookLM, which Google renamed Gemini Notebook in July 2026. Its Audio Overviews turn the documents you upload into a conversation between two AI hosts. You can choose a format (Deep Dive, Brief, Critique or Debate) and a length. I wrote a longer piece on NotebookLM's podcasts.
- HeyGen Video Podcast, launched in July 2026. It produces a video episode with two avatar hosts in a studio scene. According to HeyGen's help center, an episode can be one, two or three minutes long. I looked at it more closely in a separate post.
- Jellypod, which generates podcasts with AI hosts from a topic, a document or a script. You can edit any line, choose or design the hosts' voices, and publish to Spotify and Apple Podcasts through an RSS feed. More in my Jellypod post.
- ElevenLabs GenFM, a feature in the ElevenReader app from the voice company ElevenLabs. It turns documents and articles into a conversation between two AI hosts, and launched in November 2024. More in my GenFM post.
- Wondercraft, which now leads as an AI video studio but still offers Convo Mode, a way to generate a scripted conversation between two AI voices. More in my Wondercraft post.
There are plenty of others, and more arrive every month. They differ in polish, control and price, but the shape is the same: material goes in, an episode comes out, and no conversation ever actually happened.
These tools are genuinely good at some things. If you have a forty page report and a commute, a generated overview is a great way to absorb it. If you run marketing for a company and need a steady stream of short explainer videos, a generator gives you volume that no human team could match. I don't think either use is fake or bad. They're just not podcasting in the sense I care about.
What an AI co-host is
An AI co-host works the other way around. You are recording a show with real people, on a platform you already use, and the AI joins that recording as one more participant.
With maneku, which is the one I build, the co-host shows up in your Descript, Riverside or Restream room with its own tile, its own name and its own voice. It listens to the whole conversation in real time. It stays silent until you summon it with a button or a hotkey. When you do, it can give a quick take, research a question and cite its sources, or go deep and carry a segment. About a second and a half later it is speaking in the room, and your audience hears it the same way they hear any remote guest.
Nothing is generated ahead of time. The show is still you, your guests and whatever happens between you. The co-host is one more voice in that conversation, and you decide when it speaks.
Side by side
| AI podcast generator | AI co-host | |
|---|---|---|
| Where the content comes from | A prompt, document or script you supply | A live conversation between real people |
| Is anything recorded? | No, the episode is rendered | Yes, it's your real recording |
| Who the hosts are | Synthetic voices or avatars | You and your guests, plus one AI voice |
| Your role | Editor and publisher | Host |
| Length | Set by the tool (HeyGen caps an episode at 3 minutes) | As long as your recording |
| Best for | Studying material, content at volume, internal comms | Podcasts, video podcasts and livestreams with a real host |
The question that sorts it out
The simplest way I've found to tell which one you want is to ask yourself whether you want a podcast or whether you want to podcast.
If what you want is the finished artifact, an episode that exists without you having to record it, a generator is the right tool. That's a legitimate goal, and the tools for it are getting better quickly.
If what you want is the act itself, sitting down with a microphone and having a conversation that your audience gets to listen in on, a generator does nothing for you. It removes the part you came for. An AI co-host keeps that part and adds something to it: a second voice when you are recording alone, or a fact checker and moderator when there are already two or three of you.
Where the line blurs
Both of the big generators have added features that sound like they cross over, so it's worth being precise about them.
NotebookLM's interactive mode lets you press Join while an Audio Overview is playing and ask the AI hosts a question out loud. They answer from your sources and then carry on. It's clever and useful. But you are joining a show the AI made for you, privately. Google's help page says people you share the overview with can't interact with it, and it only works in English on newly generated overviews.
HeyGen's Video Podcast lets you invite a co-host by email. In HeyGen's written announcement, that person joins "with their own avatar." So the human appears as a rendered character reading lines from the script. There is still no live conversation.
Neither feature puts an AI into a show that real people are recording for a real audience. That is the specific gap an AI co-host fills, and I haven't found another tool that does it.
What maneku is today
Since I build one, here's exactly what maneku is right now, so you can tell quickly whether it fits.
- It is a macOS app, currently in open beta.
- It joins Descript Rooms, Riverside and Restream Studio.
- The beta runs on your own AI provider keys, which take about ten minutes to set up with the key setup guide. You pay those providers directly for what you use.
- The co-host appears as a named tile with a voice visualizer.
- It's built for a show you record yourself. If you want an episode without recording one, a generator is the better tool.
Questions people ask
Is an AI-generated podcast a real podcast?
By any reasonable definition, yes. It has a topic, a conversational shape and audio you can subscribe to. Whether people want to listen to it is a different question, and that one is still being answered.
Can NotebookLM or Gemini Notebook join my live recording?
No. Its interactive mode lets you interrupt an Audio Overview it generated and ask the AI hosts a question, but that happens inside its own app, for one listener. It does not join a Riverside, Descript or Restream session.
Can HeyGen's Video Podcast include a real person?
A person you invite by email can take part, but they join as an avatar reading a script. HeyGen's own announcement describes it that way. Nobody is recorded live.
What does an AI co-host actually do during a show?
It listens to the whole conversation and stays quiet until you summon it. Then it can give a quick take, look something up and cite it, or carry a longer segment. On Descript and Riverside it records to its own track, so you can edit it like any other guest.
Do I need to tell my audience I use an AI co-host?
I think you should say so clearly at least once and keep it in your show notes, and never let the AI pass as a human. I wrote up how I handle disclosure on my own show.
Danny Abdeljabbar
I built maneku. I also make all the artwork and marketing for Dry Quest, an app I developed with my wife, and I'm usually building a smaller app or two on the side. I co-host the Bro History podcast, and by day I work in revenue operations. Before all of this, I was an artist across a lot of mediums, including music and audio.
Want a co-host in your next recording?
maneku is in open beta on macOS. It joins Descript, Riverside and Restream as a real participant.
Get early access