Build a Audio & Video Transcription + AI Summary — Whisper + GPT-4o to Notion & Email Workflow with n8n
Your team generates hours of audio and video every week — sales calls, user interviews, standups, webinars, podcast recordings, voice memos. Every one of those files holds decisions, action items, and
Your team generates hours of audio and video every week — sales calls, user interviews, standups, webinars, podcast recordings, voice memos. Every one of those files holds decisions, action items, and quotes worth acting on. But nobody has time to re-watch a 45-minute call to find the one moment a customer described their real pain point. So the value evaporates. The recording sits in Google Drive, unindexed and unread, until it's deleted to free up space.
This article shows you how to build an n8n workflow that ingests any audio or video file — from Google Drive, Dropbox, or a direct URL — transcribes it with OpenAI Whisper, generates a structured executive summary with GPT-4o, writes the result to a Notion database, and emails it to your inbox. End to end, hands-off, in under two minutes per file.
The Problem: Recordings Are Write-Only Storage
Manual transcription is dead on arrival for busy teams. Paying a service $1.50/minute doesn't scale past a handful of files, and the turnaround kills any chance of same-day follow-up. Scrubbing through a recording yourself costs you the full runtime plus the note-taking. And even when you do transcribe, a raw 8,000-word wall of text isn't useful — nobody reads it. You need the signal: what was decided, who owns what, the three quotes that matter, and the timestamp where the important moment happens.
The gap isn't transcription technology — Whisper solved that. The gap is the pipeline: something that watches for new files, transcribes them, distills them into a usable brief, and drops that brief where your team already works. That's an orchestration problem, and orchestration is exactly what n8n is for.
The Solution: A Whisper + GPT-4o Pipeline in n8n
The workflow is a linear chain of nodes, each doing one job:
- Trigger — a new file appears (Drive/Dropbox watch) or arrives via webhook with a URL.
- Fetch the binary — download the media into the n8n execution as binary data.
- Transcribe — send the audio to Whisper, get back text with timestamps.
- Summarize — feed the transcript to GPT-4o with a structured prompt.
- Store — create a Notion page with the summary and full transcript.
- Notify — email the executive brief to your inbox.
Because n8n passes binary data and JSON between nodes natively, you never touch a server or write glue code. Each step is a configured node. The whole thing runs on your existing n8n instance (self-hosted or cloud) and costs only the OpenAI API usage — roughly $0.006/minute for Whisper plus a few cents of GPT-4o per file.
Step-by-Step Setup in n8n
1. The trigger. Pick your source. For Google Drive, use the Google Drive Trigger node set to File Created on a specific folder, polling every minute. For Dropbox, use the Dropbox Trigger. For arbitrary URLs, drop in a Webhook node (POST) that accepts a JSON body like { "url": "https://..." } — handy for wiring the workflow into a form or another app.
2. Fetch the binary. If you triggered from Drive, add a Google Drive node with operation Download, passing the file ID from the trigger. For a URL, use an HTTP Request node set to GET, and under Options set Response → Response Format to File so n8n stores the download as binary rather than parsing it as text. Whisper accepts mp3, mp4, m4a, wav, webm, and mpeg — video files work because Whisper reads the audio track directly.
3. Transcribe with Whisper. Add the OpenAI node, resource Audio, operation Transcribe. Point its Input Data Field Name at the binary property from the previous node (usually data). Under Options, set Response Format to verbose_json and enable Timestamp Granularities → segment — this is what gives you the timestamped segments GPT-4o uses to build "key moments." Whisper handles files up to 25 MB; for longer recordings, add a step to split the audio (see pitfalls below).
4. Summarize with GPT-4o. Add a second OpenAI node, resource Text, operation Message, model gpt-4o. In the system message, instruct it to act as an executive assistant and return structured JSON. Turn on JSON Output so the response parses cleanly downstream. A prompt skeleton:
Return JSON with keys: executive_summary (3-4 sentences), key_points (array), important_quotes (array of {quote, speaker_guess}), key_moments (array of {timestamp, description}), action_items (array). Base key_moments on the segment timestamps provided. Transcript: {{ $json.text }}
Pass the verbose transcript, including segment timestamps, into the user message so GPT-4o can anchor each key moment to a real position in the recording.
5. Save to Notion. Add a Notion node, resource Database Page, operation Create. Map the summary to a Title property, and set additional properties (Date, Source, Duration, Status). Put the long content — full transcript and quotes — into the page body using Blocks, so the page is readable rather than crammed into a property. Use an expression like {{ $json.executive_summary }} to map fields. Now every recording becomes a searchable Notion entry your whole team can find later.
6. Email the brief. Finish with a Gmail or Send Email (SMTP) node. Build the body from the GPT-4o output — subject line as the summary's first sentence, body with key points as a bulleted list and a link back to the Notion page. Set the content type to HTML for clean formatting. The moment a call ends and the file lands, the person who needs it gets the brief without lifting a finger.
Why This Beats Every Alternative
Speed to insight. A 45-minute call is summarized before you've finished your coffee. Follow-up happens same-day instead of "when I get to it."
Cost. Roughly $0.30–$0.50 to fully process an hour of audio in OpenAI credits, versus $90+ for human transcription — and no per-seat SaaS subscription.
Data ownership. Transcripts and summaries live in your Notion and your inbox, structured the way you want, not locked inside a meeting-notes vendor you'll have to migrate off later.
Composability. Because it's n8n, you can extend it in an afternoon: route summaries to Slack, create tasks in Linear from the action items, tag calls by sentiment, or fan out to different Notion databases by source folder. The pipeline is yours to grow.
Common Pitfalls (and How to Avoid Them)
The 25 MB Whisper limit. This is the one that bites everyone. A long video easily exceeds it. Before the transcribe node, add a step to extract and compress the audio track — an Execute Command node running ffmpeg to convert to 16 kHz mono mp3 shrinks most files dramatically, and for very long recordings, split into chunks and loop the transcription, then concatenate the text before summarizing.
Passing the wrong binary field. If Whisper returns "no file provided," your Input Data Field Name doesn't match the binary property produced upstream. Open the previous node's binary tab and copy the exact property name (commonly data).
Unstructured GPT-4o output. Without JSON Output enabled and an explicit key schema, the model returns prose that breaks your Notion field mapping. Always pin the output shape in the prompt and validate with an If node before the Notion write.
Notion property type mismatches. Sending a string to a Date property, or text to a Select that doesn't have that option, fails the node silently in batch runs. Match each mapped value to its Notion property type, and pre-create Select options.
Trigger loops and rate limits. A folder watch can re-fire on files n8n itself writes back. Scope the trigger to an inbox folder only, and add a small Wait or batch limit so a bulk upload of ten files doesn't slam the OpenAI rate limit all at once.
Build it once, and every recording your team makes from now on transcribes, summarizes, files, and delivers itself. The write-only archive becomes a searchable knowledge base — and the value you were throwing away starts compounding instead.
Ja construimos isso pra voce
Nao comece do zero. O Audio & Video Transcription + AI Summary — Whisper + GPT-4o to Notion & Email e um workflow n8n pronto para instalar — conecta suas ferramentas em minutos, sem codigo.
Instalar por $29.0 →
