You already made the content. You said it out loud, to a camera, with your actual opinions in it. This is how to stop letting it expire in 48 hours. The whole thing in one line: your reel is already a blog post — you spoke it, you just never typed it. Transcribe it, rewrite it in your voice, filter hard, and let it open a pull request instead of publishing itself.
Why this is worth doing
A reel dies in 48 hours. A post gets found at 2am, eighteen months later. A friend of mine posts on Instagram every single day. His website is dead — zero Google traffic. He has 500 videos. He could have 500 blog posts. It was easy to give him that advice, so I went and built it for myself that night, on a laptop in the car. Reels are a treadmill. The only way to get more reach is to make another one, forever. Posts are an asset. They sit on a URL you own and get found by someone typing your exact problem into Google. The hard part is already done. Knowing what to say is the expensive bit. You did that on camera. Typing is the only thing missing — and typing is what a model is genuinely good at.
Where people get this wrong trap 1
No filter
Auto-publish
Every reel becomes a post. Two-thirds of them cannot rank and sell nothing.
Nothing waits for a human. Some posts should never go live and you only learn that by looking.
The big dump
The model's voice
22 posts become 400 overnight. It looks exactly like what it is.
No voice file, so everything comes out grey and reads like everybody else.
The 6 steps
1 Pull your own reels
Fetch
Instagram's Graph API gives you every post you have ever made, with the video URL, the caption and the original date. Use the me/media endpoint and paginate it. Nothing exotic. Write each reel id into a small key-value store the moment you touch it. That is your "already done" list so tomorrow's run never repeats itself. Keep the original post date. You will need it in step 6 — it is what makes a 400-post backlog look like history instead of a dump.
2 Transcribe the audio
Listen
Whisper does this well and cheap. Point it at the video URL, get text back. Set a floor: under roughly 120 characters of transcript, throw the reel away. A six-second "wait for it" clip has no article in it. The transcript is the whole point. Everything downstream is just editing.
3 Score it BEFORE you write it
Gate
This is the step everyone skips and the one that saves you. Ask the model to rate 1–10 whether this reel deserves a standalone post at all. Score on three things: does anyone search for this, does it tie to something you sell, is there one insight that is actually yours. Below a 7, mark it processed and move on — so it is never re-scored tomorrow. You pay for one cheap scoring call instead of a full article you will never publish.
4 Write it in YOUR voice, not the model's
Write
Keep a voice file: how you actually talk, words you never use, the products you mention, how you open and close. Feed it the transcript plus that file. Without it you get grey mush that reads like everybody else's blog. Ask for a real title, a slug, a meta description and a date — structured so it drops straight into your site with no hand-editing.
5 Ship it as a pull request, never straight to live
Gate #2
The pipeline opens one pull request a day against your site repo. It does not publish. You skim it on your phone, merge the good ones, close the rest. Thirty seconds of taste per day. This is the difference between an asset and an embarrassment.
6 Trickle, do not dump
Ramp
What i got wrong
THE MISTAKE I MADE, SO YOU CAN SKIP IT I had this running for a month before I understood what it was doing to me. WHAT I BUILT FIRST
The only filter was "is the transcript long enough." Every single reel became an article. WHAT THAT PRODUCED
117 posts across 26 unmerged pull requests — on a site that had 22 live articles. WHY IT WAS GARBAGE
Two-thirds were generic dev explainers. What is a key-value pair. Basic CLI flags. Things that cannot rank and sell nothing. WHAT MERGING IT WOULD HAVE DONE
Thin content at scale, pointed straight at my main SEO asset. The exact thing Google is built to catch. THE FIX
A relevance score that runs before the expensive write step. I kept 33 and rejected 84. The pipeline got cheaper AND better on the same day.
The lesson is not "don't automate." It is that an automation with no taste in it just produces your bad ideas faster. Put the filter in front of the generator, not behind it. THE FILTER THAT FIXED IT This runs on the transcript before a single article gets written. A cheap call that saves an expensive one: Score this reel transcript 1-10 on whether it deserves a standalone blog post. Score HIGH only if it has: - real search demand (people actually google this) - a tie to something I sell or do - one unique insight that isn't in 40 other posts Score LOW for generic explainers: "what is a key-value pair", "how to install X", basic CLI commands, anything a doc page already answers better. Return: { "score": <1-10>, "reason": "<one line>" } Transcript: $TRANSCRIPT
Ship this week
YOU BUILT IT RIGHT IF… Every unchecked box is either money left on the table or a penalty waiting to happen. Every reel id I touch gets written to a store, so nothing is processed twice. Reels with almost no transcript are dropped before they cost me anything. A relevance score runs before the article is written, not after. My voice file is a real file — not "write in a casual tone" shoved in the prompt. Nothing publishes itself. It opens a pull request and waits for me. Posts are capped per day and backdated to the original reel date. My sitemap picks up new posts automatically — I never edit it by hand.
Keep going see the output
sanskartiwari.io/articles
buildcheap.app
The posts this pipeline produced are live. Read them before you build it.
Describe the app, it gets built. Shows the real $ cost of every action before it runs — no credits, no guessing.
Get found
Then sell
sanskartiwari.io/dofollowback
sanskartiwari.io/convert-visitors-to-buyers
166 dofollow backlinks. New posts need links pointing at them.
Traffic is step one. Turning it into money is step two.
If you do only one thing: take your ten best-performing reels, run the transcripts through a model with your voice notes attached, and publish them as posts this week. Ten posts is enough to find out whether Google likes you. Build the automation after that answer comes back.
Found this useful? Come build with me.
More no-fluff guides, product breakdowns & behind-the-scenes — every week.
instagram.com/sanskarr.tiwari
