rawblog.ai

Most blog platforms quietly punish you for sounding like yourself. You write something honest, then spend an hour editing the soul out of it so it looks “professional”. By the end it reads like a slide deck. Half your drafts never ship at all.

I have had this problem my whole life, and not only on blogs. I think in many directions at once. I can hold an idea and its opposite at the same time and be convinced by both. I speak fast, my thoughts double back, and I have watched people decide I am scattered simply because my thinking does not move in a straight line.

Language is a compression algorithm. It takes something multidimensional and flattens it into a sequence of words. The frameworks survive. The texture, the contradictions, the raw signal, gets cut. rawblog.ai started from a refusal to accept that loss as the price of being understood.

The idea

What if you did not have to choose between thinking freely and being understood?

On rawblog you publish the thought exactly as it left your head: a voice-memo transcript, a late-night ramble, a half-formed argument. An AI then writes a clean, readable companion version alongside it. Not instead of it. Alongside it.

Every post carries a toggle: Raw or Polished. The reader picks. Want the texture of how someone actually thinks? Read raw. Want the signal in a straight line? Read polished. Same idea, two versions.

I had the idea long before AI could build it. The moment language models got good enough to find structure inside chaos without sanding the voice off, it became possible.

How it works

1. Dump it raw. Paste anything from 50 to 15,000 characters, or dictate it. No formatting, no headings, no cleanup.
2. AI polishes alongside. A structured version is generated in seconds. Your phrasing and intent stay intact; the AI cleans, it does not replace.
3. Publish both. Readers get the toggle on every post. Search engines get a readable version. Your readers get to see how you actually think.

Three principles hold the whole thing together:

• Raw is canonical. The original is never overwritten. The polished version is a companion, not a replacement.
• Voice-preserving polish. Your contradictions and odd turns of phrase are kept. No generic LinkedIn tone.
• Custom polish prompts. Each author can write their own instructions for how the AI should clean their work: tone, length, formatting.

It is free to start, with no credit card, and authors can publish up to 20 posts a day.

Not another AI writer. The opposite.

Most AI writing tools generate text for you. You end up sounding like every other post on the internet, because you are reading the model’s voice, not yours.

rawblog flips the direction. Your raw thinking is the input that matters. The AI’s job is to make it legible, not to replace it. The About page of the site is itself the proof: it began as a fast, messy voice memo and you can toggle to read the original, word for word, next to the clean version.

Humans helping AI

There is a second dimension I think about constantly. AI has consumed the surface of human knowledge at a scale no person ever could. What it does not have is how we actually think: new signals, live contradictions, ideas forming in real time inside real human minds. That raw material has never been captured.

In a world drowning in recycled content and optimized noise, a genuinely new human signal is the rarest thing on the internet. rawblog is built so the exchange goes both ways: AI helps humans communicate more clearly, and unfiltered human thought gives AI something it cannot generate on its own.

How it is built

rawblog is an Astro site with server-rendered pages, deployed on Vercel. Authentication, profiles and posts live in Supabase, with row-level security and a schema that has evolved through a series of migrations, from per-author polish prompts to drafts and visibility controls. Polishing runs on Anthropic’s Claude, server-side only. Dictation uses Deepgram speech-to-text so you can talk a post instead of typing it. Sign-up and login are protected by Cloudflare Turnstile.

Every author gets a public profile, posts are grouped by tag in a shared feed, and there is an RSS feed for readers who prefer their own tools. The site is also built to be read by machines: a public “HeyAi” page documents exactly how AI agents can crawl and retrieve its content, so the arrangement is transparent to humans and machines alike.

Where it is going

The public roadmap describes rawblog growing from a publishing tool into a thought platform, in four phases:

• Create and distribute. One raw thought, published everywhere: blog, X, newsletter, with channel-specific styles and canonical links back to the original.
• Thought OS. Voice inbox, atomic thought units, an idea graph that maps themes and contradictions, and a builder that turns related thoughts into a series.
• Network intelligence. Follow graph, reactions and passage comments, and matching between people who are thinking about the same things.
• Teams. An “Ask your thoughts” assistant with citations, shared workspaces, and links into project tools.

The principle behind it: ship in layers, and let real usage decide how deep each feature goes.

Where it stands

rawblog.ai is live and open for sign-ups at rawblog.ai. If you have a thought sitting in your drafts waiting to be polished, that is the exact problem it was built for.