Research to Podcast: An AI Workflow You Can Review

“Turn research into a podcast” hides several separate jobs: choosing sources, writing for listening, creating speech, storing the file, and distributing it. A useful first automation makes those boundaries visible. You should be able to see which material produced an episode, approve its wording when needed, and recover from an interrupted request without accidentally starting another paid generation.

What real content-to-audio workflows already do

Two published, creator-authored n8n templates illustrate the pattern. A daily RSS-to-podcast example separates article collection, script creation, speech chunks, and audio assembly. A blog-to-podcast example adds file storage and feed metadata.

These are implementation examples described by their authors, not measured evidence of quality or a tested AudioAvenue integration. They inform a practical design choice: keep collection and distribution in your existing workflow tool, and treat AudioAvenue as the short-episode production step.

Start with one question and a source brief

For the first run, choose one listener and one question. A two-minute product-team briefing might answer, “What did the exploratory interviews suggest we should test next?” That is a more checkable goal than summarizing everything in a research folder.

Before generating, assemble a brief with the source name or URL, publication or observation date, key findings, limitations, and the intended takeaway. Keep observations separate from your interpretation. This is our recommended preparation format, not a requirement that the tool reads links automatically.

The downloadable sample is an invented product study with six volunteers. It deliberately includes a small sample and an untested recommendation. Use it to check that the script does not turn three expressions of interest into three paying customers.

An editorial brief for the sample

Create a two-minute briefing for a product team from this fictional demonstration data. State the research question, summarize the three observations, and end with the proposed next experiment. Preserve the six-person sample and the limitations. Separate observations from recommendations. Do not invent quotes, statistics, adoption, or revenue. Return only the spoken script.

Try this once in the studio with automatic audio generation turned off. Review the script, then generate narration. Keep that approved example as a reference when building the external workflow.

Choose where a person reviews the work

There are two useful production paths:

• Combined generation: submit reviewed source material and an approved credit estimate, then receive a script and audio through one task. This suits a format whose editorial behavior you have already tested.
• Script-first generation: create a draft, generate a script, inspect and edit it, then request audio with a fresh audio estimate. Use this when the actual wording needs approval.

n8n's human-review documentation describes pausing an agent before selected tool calls. In an AudioAvenue integration, you can use such a review step before a paid request. A tool-call approval is distinct from reading the generated script: that editorial check requires the script-first path and a separate pause after the script exists.

Map the production steps to the API

Sign in, open Manage workflow access, and create a podcast key with read and generation scopes. Store it in your automation tool's credential settings. The base URL is your backend origin plus /api; requests use bearer authentication.

The combined-workflow body contains title, material, language, minutes, instructions, serverIds, and expectedCredits. An optional voiceId selects an existing verified personal voice. Omit it for the studio narrator. Use an empty serverIds array when you supply the finished research without MCP tools.

Calculate the live estimate as the configured script price plus audio price per target minute multiplied by target minutes. A missing price disables the corresponding operation. Submit the reviewed numeric amount as expectedCredits; a price change or insufficient balance stops the request.

The script-first API path

Create a free draft with POST /podcasts/episodes. Read its quote at GET /podcasts/episodes/:id/quote/script, then submit POST /podcasts/tasks with that episode ID, operation script, and the reviewed estimate. After completion, retrieve the episode and review its script. Save edits through PATCH /podcasts/episodes/:id/script. Read the current audio quote and submit operation audio only when the narration is approved.

How this maps to n8n

The official HTTP Request documentation covers authentication, headers, JSON bodies, and file responses. Based on that capability and AudioAvenue's API, the following is a proposed node sequence, not an importable or end-to-end-tested template:

Manual trigger or reviewed source brief
  → Read configuration and balance
  → Review source, selected voice, and credit budget
  → Persist request body and idempotency key
  → Submit production task
  → Wait 2–3 seconds → Read task status
      queued / running → wait and poll again
      failed → retain results and stop
      succeeded → authenticated MP3 download
  → Store file and review before distribution

Use JSON responses for task calls and the File response format for the audio download. The HTTP 202 response is an acceptance receipt, so do not treat it as an MP3. Keep the saved task ID and episode ID available to later nodes; storing only the latest poll response can lose context.

Make an interruption recoverable

Persist one Idempotency-Key with the exact generation body before submitting. If the response is lost, replay that body and key. A new key represents a new generation attempt. If the task is already known, resume polling its ID instead of resubmitting.

When a task fails, inspect its stages and charges. A successful script remains saved and charged if later audio fails. Retain it; if you approve another attempt, request audio alone with the current quote and a new key. Bound your polling time and report a still-running task so it can be resumed later.

What remains outside the production step?

Fetching an RSS feed, extracting PDF text, scheduling the run, and sending or publishing the final file belong to your external workflow. AudioAvenue's optional MCP connections are administrator-configured research services; they are separate from the workflow API and do not create personal document-account connections.

The studio currently targets one-to-three-minute, single-host episodes from 20–40,000 characters of source text. Keep your first automation within that scope. For a long conversation, a multi-speaker production, or feed hosting, evaluate those requirements separately. The podcast creation guide covers the editorial checks to carry into your workflow.

Sources and further reading

• Creator-authored RSS-to-podcast template — collection, generation, and audio assembly example.
• Creator-authored blog-to-podcast template — storage and feed-output example.
• n8n HTTP Request and human review for tools — official configuration references.