I have spent years writing about SEO and AI search. Long-form articles, LinkedIn posts, newsletter editions — all text, all the time. Then one afternoon I opened Google's NotebookLM, hit a button I had never pressed before, and thirty minutes later I was listening to two AI hosts discuss my own content as if they were recording a real podcast episode.
That moment changed how I think about content distribution. And within a few days, I had a published podcast show on Spotify — built entirely with AI, zero audio production skills required.
This is the story of exactly how I did it.
What Is NotebookLM — And Why It's Different
NotebookLM is Google's AI research assistant. You upload sources — PDFs, articles, Google Docs, YouTube links, web URLs — and it synthesises everything into a personal knowledge base you can query like a chat interface.
But the feature that changed everything for me is called Audio Overview.
Audio Overview takes your uploaded sources and generates a fully produced podcast-style conversation between two AI hosts. Not a robotic text-to-speech reading. An actual conversation — with natural back-and-forth, analogies, emphasis, moments of curiosity — that covers the key ideas from your content.
Audio Overview is not a text-to-speech converter. It genuinely understands your content, restructures it into a conversational format, and generates dialogue from scratch. The output sounds like two informed people genuinely discussing the topic.
Why I Decided to Start a Podcast
My reasoning was straightforward. A growing segment of my audience prefers audio — commutes, workouts, passive learning. My written content was already solid. The only barrier was production.
I cannot record consistently. I do not have a studio setup. I have never edited audio in my life. Traditional podcasting was off the table for me as a solo operator running a consulting practice.
NotebookLM removed every single one of those barriers. The question became: can I go from AI-generated audio to a properly published Spotify show? The answer, it turns out, is yes — and faster than I expected.
Step-by-Step: How I Built the Podcast
Feed NotebookLM Your Content
I created a new Notebook in NotebookLM and uploaded sources — my blog posts, key LinkedIn articles, and a few industry PDFs on AI SEO topics. You can use URLs, Google Docs, PDFs, or paste text directly. The more structured and content-rich your sources, the better the audio output.
Generate the Audio Overview
In the top-right area of the Notebook interface, click "Audio Overview" then hit "Generate". NotebookLM processes your sources and creates a 10–20 minute conversation. My first episode took about 4 minutes to generate. The output is an MP3 file you can download directly.
Customise the Focus (Optional)
Before generating, you can give NotebookLM specific instructions — "focus on practical tips", "keep it beginner-friendly", "explore the debate around AI search replacing traditional SEO". This shapes the conversation significantly and gives you editorial control over what gets discussed.
Download the MP3
Click the download icon on the generated audio player. You now have a production-ready MP3 file. No editing, no noise removal, no levels adjustment needed — the audio quality is clean and consistent straight out of NotebookLM.
Create a Podcast Host Account
I used Spotify for Podcasters (formerly Anchor) — it is free, it distributes to Spotify automatically, and the setup takes under ten minutes. Create an account, name your show, write a description, upload a cover image (1400×1400px minimum), and set your category.
Upload Your Episode
In Spotify for Podcasters, hit "New Episode", upload the MP3, add a title, write episode notes (I repurpose the written version of the content here), set a publish date, and hit publish. Distribution to Spotify typically happens within a few hours.
Repeat and Build a Show
Each new blog post or content piece becomes a new episode. I create a dedicated Notebook per episode, feed it the relevant sources, generate the audio, and publish. The whole pipeline takes 30–45 minutes per episode, most of which is waiting for generation.
Listen to the Show on Spotify
The podcast is live and you can listen to it right now. Every episode is a deep-dive conversation on AI SEO, search visibility, and the future of content — all built from my written work, reimagined as audio.
What the Audio Quality Is Actually Like
This is the question I get most often. People expect it to sound robotic or synthetic. It does not.
The two AI hosts use natural pacing, interrupt each other occasionally, emphasise key points, use relatable analogies, and ask each other questions as if they are genuinely curious. There are moments where a host says something like "that's a really interesting way to put it" — and it sounds unrehearsed.
Is it perfect? It is not identical to human-hosted podcasts. Occasional phrasings feel slightly formal. The hosts cannot draw on personal experience or go off-script with genuine anecdotes. But for educational and thought-leadership content, the quality is more than sufficient — and most listeners I have shared it with did not immediately identify it as AI-generated without being told.
The real benchmark is not "does it sound human?" — it is "does it communicate the ideas clearly and hold attention?" By that measure, NotebookLM's Audio Overview passes without question.
How NotebookLM Compares to Other Options
| Tool | Ease of Use | Audio Quality | Free to Use | True Conversation Format |
|---|---|---|---|---|
| NotebookLM | ✓ Very easy | ✓ Natural dialogue | ✓ Yes | ✓ Yes (2 hosts) |
| ElevenLabs | Moderate | ✓ Excellent | ✗ Paid | ✗ Single voice TTS |
| Murf AI | Moderate | Good | ✗ Paid | ✗ Narration only |
| Podcastle | Moderate | Good | Limited free | ✗ Narration only |
| Record yourself | ✗ High effort | Varies | ✓ Yes | Depends |
The SEO and Visibility Angle
Beyond the convenience of production, there is a strategic reason I care about this as someone focused on AI search and content visibility.
Audio content is underindexed in AI search. Most brands competing for visibility in AI-generated answers are focused entirely on text. Podcast transcripts, episode descriptions, and show notes are relatively uncontested territory for entity reinforcement and topical authority signals.
Publishing on Spotify also creates a separate brand presence — the show, its metadata, and eventually its mentions across the web all feed back into how models like Claude, Perplexity, and ChatGPT understand and associate my name with specific topics.
It is not just a podcast. It is a distribution layer and an entity signal.
Publish detailed episode show notes on your podcast host and cross-post them as blog summaries. This creates a text anchor for each episode that search engines and AI models can crawl, index, and cite — amplifying the SEO value of every audio piece you produce.
What I Would Do Differently
A few things I have refined since the first episode:
- Give NotebookLM a clear focus instruction. Open-ended generation produces good episodes, but targeted instructions produce great ones. Specify the angle, the audience, and the key takeaway you want the hosts to land on.
- Use higher-quality source documents. The better your input content, the more specific and insightful the conversation. Thin or generic sources produce thin conversations.
- Write proper episode descriptions. Do not just upload the audio and move on. Spend 10 minutes writing keyword-rich show notes — this is where the SEO value lives.
- Maintain consistency. Sporadic publishing hurts Spotify's algorithmic visibility. Batch-generate episodes in advance so you can publish on a reliable schedule.
- Use a consistent cover image and show identity. The show brand matters for click-through on Spotify. Invest 20 minutes in a clean, professional cover.
The Bigger Picture
We are at a point where the production barrier for audio content has effectively been removed for individual creators. What used to require a studio, a co-host, a recording schedule, and hours of editing can now be done in under an hour with a free tool and content you have already written.
That does not mean the thinking disappears. The strategy, the positioning, the quality of the underlying ideas — those still determine whether a podcast is worth listening to. But the execution layer is no longer the bottleneck.
If you are already producing written content and have been putting off starting a podcast, the reason to wait no longer exists. Open NotebookLM, upload your best articles, and hit generate. You will have a listenable episode in under 30 minutes.
The tools have caught up with the ambition. The only thing left is to use them.
