Podcast Agent

How the Podcast Agent turns a prompt into an episode.

The Podcast Agent is the Podcasts chat that researches a topic, writes a script for your hosts and generates the audio.

Start an episode

1. Choose Podcast Episode on the Studio start screen.
2. Describe the episode. See write a strong prompt. Click Enhance to have Jellypod rewrite your draft into a clearer, more specific brief that keeps your topic and requirements.
3. Optionally attach sources, such as URLs, files or YouTube links. Pasting 2,000 or more characters into the input turns the text into a source. With sources attached, the prompt is optional and the sources set the focus.

The agent researches, reads your sources, writes the full script and generates the audio in one pass, with no approval step. Credits pay for the narration audio, not the research.

One prompt makes one episode, even when the topic spans several subtopics. To get several, see generate a series.

Write a strong prompt

A strong prompt names:

• Topic: what the episode covers.
• Length: pick a duration with the stopwatch, or state one in the prompt, such as “a 15 minute episode”. A duration in the prompt overrides the stopwatch. With neither, an episode defaults to 7 minutes.
• Tone: casual, authoritative and data-driven, or humorous.
• Audience: beginners, experts or a general audience.

Weak prompt:

Make an episode about AI.

Strong prompt:

Create a 10-minute episode about the future of remote work, focusing on recent trends in hybrid models and what they mean for small businesses. Keep the tone conversational and accessible.

Tell the agent how to use the sources you attach, or it may not emphasize the parts you care about, for example “Use the attached report as the primary basis for this episode” or “Pull the key statistics from the linked article and build the discussion around them.”

Each host’s personality and backstory, set in the character, shape how they speak. Direct the agent to use them, for example “Have Host A challenge Host B’s perspective on this point” or “Make the conversation feel like a debate, not a lecture.”

For multi-part content, number the topics and say how to open and close:

Create an episode covering three topics:

1. The rise of AI coding assistants
2. How they’re changing developer workflows
3. Predictions for the next two years

Open with a hook about how much code is now AI-generated. Close with practical advice for developers.

Web research

The agent searches the web and reads pages whenever fresh information would improve the episode. Web research is always on and reads public pages only, so for paywalled material, copy the text and attach it as a source. To keep an episode to your own material, say so in the prompt, for example “use only the attached sources”. When the request is about the attached material itself, such as discussing your own manuscript, the agent skips the web unless it needs outside facts.

Revise in chat

After the episode generates, ask for specific changes in the chat. The agent edits the existing script instead of starting over. For example:

• “Make the opening more attention-grabbing. Start with a surprising statistic.”
• “Add a section about hybrid work models after the second chapter.”
• “Add more back-and-forth between the hosts in section three.”
• “Have Maya voice every line Sam speaks.”

The agent applies a requested change in the same turn, without asking you to confirm, and you can undo each edit in one step. A recast can swap in any active Character with a Voice, not only the podcast’s hosts. Lines you recorded yourself keep their voice.

When a change needs fresh information, the chat shows Web Search and Read URL activity and cites the URLs it used. To fix a single line, regenerate that segment instead of re-prompting.

The chat is the AI Assistant pane on the left of the episode editor, and each episode keeps its own chat history. To get back to it:

1. Open Podcasts and choose your podcast.
2. Open the episode’s menu (three dots) and choose Edit Episode.
3. If the pane is hidden, click Show AI assistant above the preview. In a narrow window, use the Chat tab.

Generate a series

A series is two or more episodes generated in one batch, each script building on the one before it. You can review an outline first or ask for several episodes in one prompt.

Review a series outline first

1. Open the podcast and click Generate a Multi-Episode Series below the Create New Episode chat input.
2. Describe the series and choose 2 to 8 episodes. The hosts, attached sources, Auto-publish and Include show notes apply to every episode.
3. Set the shared duration, which defaults to 5 minutes per episode. Changing it updates every episode, so set it before adjusting individual rows.
4. Click Create Outline. Jellypod reads your sources and, for factual topics, can run one web search and read up to two pages. To keep the plan to your material, say to use only the attached sources.
5. Review the plan. Each row shows a title and a one-sentence summary. Click Edit to change an episode’s detailed prompt, then Save to refresh its title and summary. Delete episodes or change one episode’s duration as needed.
6. Click Create Episodes. The button shows the estimated credit total.

Edits stay in the dialog until you create the episodes. You can’t delete the last episode.

Generate from a prompt

Jellypod skips the outline and generates several episodes when your prompt clearly asks for more than one.

A broad or ambiguous request, such as “an episode covering X, Y, and Z”, stays one episode. A duration in the prompt (“15 minute episodes”) applies to each episode.

How episodes build on each other

Research starts for every episode at once, but scripts are written in order. Each writer gets the full previous script plus short summaries of older episodes, so names, facts and running threads stay consistent.

An episode waiting on an earlier one shows Waiting for Episode N. If an earlier script fails, or doesn’t arrive within 90 minutes per earlier episode, later episodes stop instead of starting from a gap. Generation continues if you close the tab.

Limits, cost and failures

The outline dialog plans 2 to 8 episodes, and a direct prompt makes up to 10; ask for more and Jellypod asks you to request 10 or fewer. An organization can run 10 generations at once and a batch needs one free slot per episode, so a batch is rejected if you already have generations running.

Each episode is charged for its narration at the credit rates. If an episode fails, its held credits are released and it is not charged. Episodes that stopped because of it fail too. Retry failed episodes from the podcast page, and see if episode generation fails for causes and fixes.