Call us

How we built an automated AI industry newsletter

Once a day, a newsletter summarises new videos from selected AI channels on YouTube, as text and as a voice recording. We built it without writing code or running our own servers, on Make.com, RapidAPI, Runpod, ChatGPT and ElevenLabs.

Adrian Baczyński

Example issue of the newsletter as an email (opens the full-size image)

New tools, strategies and breakthroughs in artificial intelligence appear so quickly that keeping up with them is hard. To advise our clients well, we needed a way to monitor, evaluate and apply the most important AI updates on a continuous basis.

A no-code approach built on APIs

Instead of writing our own code or maintaining our own servers, we chose a no-code approach built on APIs. It lets us collect, process and distribute information quickly while keeping operating overhead low. The goal was a newsletter that gathers insights from YouTube channels which explain complex AI topics in plain language.

The tools we used

We chose tools that are reliable, inexpensive and easy to connect:

  • Make.com: the core of the system, where the automation logic is built.
  • RapidAPI: extracts the audio from YouTube videos as MP3 files.
  • Runpod: converts the audio to text with a dedicated transcription model.
  • ElevenLabs: creates the voice version of the newsletter.

A spreadsheet as the database

The database is a simple spreadsheet that holds everything the newsletter needs. The main sheet has eight columns:

  • STATUS: reserved for future use.
  • ACTION: reserved for future use.
  • IsEnabled: whether the channel should be monitored.
  • Category: context about the channel, used to tailor the prompts.
  • Channel URL: a direct link to the YouTube channel.
  • RSS ID: the identifier used by the RSS Feed module in Make.com.
  • Group Name: the newsletter group the channel belongs to (there can be several groups).
  • Filter Prompt: filters videos by title, so that only relevant ones go into the newsletter.
The newsletter's main sheet with YouTube channels and their settings (opens the full-size image)

How the automation works, step by step

The newsletter is built by several scenarios in Make.com.

1. Collecting the YouTube feeds

Once a day, at a set time, the first phase of the automation starts. The spreadsheet module finds all rows marked as enabled. After irrelevant entries are filtered out, the remaining records are copied to a working sheet for that day's newsletter. The process then aggregates all entries and waits 10 minutes, so that every operation has finished before the next step.

Make.com scenario, phase 1: collecting new videos from YouTube channels via RSS and saving them to the spreadsheet (opens the full-size image)

2. Processing the videos with RapidAPI

In the next phase, started with the right parameters, the system sends the video URL to RapidAPI. The API fetches the audio as MP3, and the status in the spreadsheet is updated.

Make.com scenario, phase 2: sending the video URL to RapidAPI and updating the status in the spreadsheet (opens the full-size image)

3. Transcription

RapidAPI returns the task information through a callback URL. If the task succeeded, it goes to Runpod for transcription. If it failed, the system retries the request automatically. If the errors continue, an alternative RapidAPI call fetches the MP3, and once that task succeeds, Runpod starts the transcription.

Make.com scenario: the RapidAPI callback and a retry after an error (opens the full-size image)
Make.com scenario, phase 3: transcription in Runpod and saving the result to the spreadsheet (opens the full-size image)

4. Compiling the newsletter

After a 12-minute wait, the main scenario starts this phase, which gathers all the data from the newsletter spreadsheet and then:

  • Verification: checks that the videos are not YouTube Shorts.
  • Summary: sends the transcript to ChatGPT, which writes a short summary of the video.
  • Voice script: prepares the text for the voice version and sends it to ChatGPT.
  • Audio: sends the text to ElevenLabs, which creates a voice file saved to Google Drive.
  • Final document: ChatGPT compiles the thumbnails, video links and summaries into a final HTML document.
  • Distribution: the HTML goes out through Gmail to the team members who use the newsletter.
The newsletter data sheet the automation uses to put each issue together (opens the full-size image)
Make.com scenario, phase 4: putting the newsletter together with OpenAI and sending it by email (opens the full-size image)

What the team gets every morning

Every morning, the team can catch up on anything they missed the day before and on updates published overnight. The voice version can be played over a morning coffee.

Plans

We are planning to monitor more than YouTube: blogs and manufacturer websites will be added as sources, so the newsletter no longer depends on YouTube content alone.

Costs

A newsletter based on 10 videos usually takes about 400 operations in Make.com. The other costs are:

  • GPT API: about 10 cents per newsletter.
  • Runpod transcription: about 1 cent per video.
  • ElevenLabs: the largest cost – a $22 monthly subscription covers about 30 voice newsletters.

Read also: How we automated lead qualification with Retell AI, Make.com and ClickUp.

If you would like a newsletter like this, or another AI automation for the information your team works with, write to us at kontakt@proop.pl.

Contact

Let's start with a call to get to know you and your needs

On the first call, you tell us about your company's situation and we assess whether we can help.

Call uskontakt@proop.pl

Technologies we work with