AI and Authenticity: Fighting the Dead Internet with Voice
- 6 minsLet’s start with this premise: we’re living in an internet dominated by bots—the Dead Internet Theory.
And the reason behind it’s simple: if you’re a content creator, you easily realize how much more convenient it is to generate a blog post, user-generated content for an Instagram reel, or now they’re even coming up with YouTube and Twitch simulations, completely generated by AI. This is clearly made possible by models like LLMs, classic text-based ones, GPT and similar, or multimodal LLMs, specifically services like GPT-4o Image Generation, Sora from OpenAI, but also Veo3 from Google, which are capable of generating realistic content that’s indistinguishable from reality, let’s say from a viewer’s perspective, but which are nothing more than AI-generated.
The Downside of AI-Generated Content
This clearly presents enormous problems. When you generate a blog post with artificial intelligence, you don’t even realize that the model, as we think, can hallucinate.
Maybe it does hallucinate from wrong information, maybe from information that doesn’t resonate. And so you generate content that might be cheap. Search engines at the SEO level notice it, there’s a deranking of this content and so fundamentally you get almost the opposite effect—you’re not publishing content, but you’re publishing really cheap content that won’t attract users’ attention.
The Alternative: Authenticity
However, there is an alternative. I strongly think that the people who create content on the internet have the desire to be authentic. They have the desire to expose themselves, to show themselves, to show what they have inside. This applies both in terms of blog posts, personal blog posts like this one you’re seeing, but also in terms of blog posts about their products, which are still an extension of themselves and therefore their artistic expression, but also regarding various content.
Why AI Blog Posts Feel Inauthentic
Regarding blog posts, a measure I’ve found, which I’m currently using for the blog post you’re reading right now, is to think more about the medium.
In my opinion right now, from my direct experience, but also from that of others I’ve had the pleasure of talking to, the generation of blog posts with AI is not fulfilling, but it’s convenient. The convenience, the fact that from a prompt, from just a few pieces of information, I can generate a blog post really quickly compared to taking an hour to write it, far outweighs the feeling of writing something inauthentic. And the reason is clear: we all compete in an era where we need to seek others’ attention. If we produce more content, we have more chanches to attract the user attention.
If we give content that attracts attention, we get more clicks, more impressions, and so this clearly has advantages. If you do it on a personal blog, where maybe you talk about other products, clearly the influence you gather also goes to your products. If it’s a product blog, clearly you attract clicks to the product and so on.
My Solution: Voice-to-Text Authenticity
But in my opinion, there are more efficient ways to be authentic. One of these is using voice.
A Bit About Me
A bit of background about me: I’m a somewhat atypical developer, I’m not the classic developer who likes to stay in corners, let’s say.
I was born with an introverted mindset that then, through various life experiences, became very extroverted. I understood that expressing myself makes other people understand me, but it also helps me first and foremost to be sure of what I have inside, right? And this over the years has made me become more and more extroverted. And many colleagues would also say that I’m quite talkative, I like to talk, I like to express myself, I like to elaborate concepts and I definitely have a small problem with the art of summarizing, right? In fact, when I talk, the various concepts I have in my head all seem much more important than they are, and so I tend to say more things rather than fewer things.
But I like doing it and it often happens that I send long audio messages on WhatsApp, I send long audio messages on Telegram, both about personal things and work, and so I experimented with this thing.
My Voice-to-Text Workflow
Initially, I started writing blog posts with voice for EasyPapiro. Specifically, I would go out for a walk, go on Telegram to send myself an audio message, and this audio was the blog post.
I would talk, talk, talk, talk, and then I would give this audio to a speech-to-text model that would then be reformatted a bit to make it into a blog post. I did this with VoiceLog, a product I created a year ago that allows you to do exactly this—it creates a preview of a blog post from a Telegram audio. Then I took a step further in the sense that speech-to-text models have improved a lot, and now my workflow for creating blog posts is a bit different.
I use Super Whisper, which has a great ability to understand what I’m saying and synthesize it well. Then I use a minimum of AI to correct those small spelling things or, I don’t know, when I talk about a product, to put the right links, I do a bit of post-processing, and then with my blog, with this blog you’re reading, I go to publish and I publish the post.
me “writing” this blog post
Why This Works
This, in my opinion, is a very beautiful way to use AI to be authentic. It’s in the sense that through speech you’re as authentic as possible—you can’t read in the blog post when maybe I go from one topic to another, or maybe when while I’m talking I focus on something that particularly interests me, which is an expression of me, in fact, an expression of the content I want to give. With post-processing I don’t distort the content, I only correct spelling or format it, and so it comes out as a really very human discourse.
The Result: Fighting AI with AI
The beauty of this approach is that it leverages my natural communication style. As someone who enjoys talking and can articulate complex ideas in a single flow, I can now create content that feels genuinely human while still being efficient. What used to take an hour of careful writing now takes just 20 minutes from thought to publication.
Here’s the timeline: I started recording this audio at 3:50 PM on Sunday, June 22nd. By 4:10 PM, it was live on my blog. Twenty minutes total—from voice to published post.
The irony is perfect: we’re using AI to combat the very problem AI created. While AI-generated content threatens to make the internet feel artificial and soulless, we can harness AI’s speech-to-text capabilities to preserve human authenticity. It’s fighting fire with fire, but in the best possible way.
Luigi Donadel
Senior Software Engineer and Entrepreneur. Building stuff.