TL;DR: LLMs can't watch videos. To use Reels and TikToks in ChatGPT, Claude or a RAG pipeline, you need them as text. The reliable way: use platform captions when they exist, transcribe the audio with Whisper when they don't, and return structured JSON. Below you'll find a no-code option, JavaScript and Python examples, and a pipeline for transcribing a whole niche.
Short-form video is where a lot of today's knowledge, opinions and marketing lives: product demos on TikTok, expert tips in Instagram Reels, announcements on X. But none of it is searchable or usable by AI until it's text.
This guide shows how to get clean, timestamped transcripts from Instagram, TikTok, X, Facebook and YouTube videos, one at a time or thousands at once, without running any servers yourself.
Why transcribing social videos is harder than it looks
If you've tried it, you've probably hit some of these:
- Every platform is different. YouTube often has captions, TikTok sometimes does, Instagram and X almost never do.
- Downloading is the hard part. Social networks block scripts, links expire, and formats change every few weeks.
- Speech-to-text has to be good. Reels are full of music, accents and fast talking.
- It needs to scale. Researching a niche means hundreds of videos, not one.
The approach: captions first, Whisper second
The most reliable pipeline looks like this:
- Resolve the video URL and read its metadata (author, caption, views, publish date).
- Use the platform's own captions if they exist in the spoken language. They're free and instant.
- Otherwise download only the audio track (a few hundred kilobytes for a Reel instead of the whole video) and transcribe it with Whisper large-v3-turbo, which handles music, accents and 90+ languages well.
- Return structured JSON with the full text, timestamped segments and optional SRT/VTT subtitles.
- Optionally, read the text shown on screen (many Reels carry their message in captions over music) and let an LLM break the video down: summary, hook, key points and call to action.
You can build this yourself with yt-dlp, ffmpeg and a Whisper API. If you'd rather skip the maintenance (the platforms change constantly), the Video to Text Actor on Apify does exactly this, and I'll use it in the examples below:
Option 1: No code (2 minutes)
- Open Video to Text: Instagram, TikTok & X Transcripts and click Try for free.
- Paste video links, one per line. You can mix platforms. These two are real, so you can try them right away:
https://www.instagram.com/reel/DMflZnXOKUj https://www.tiktok.com/@apifylife/video/7338085038258457889 - Optional: turn on AI analysis to also get a summary, the hook, key points and the call to action of every video, or pick a language in Translate transcript to.
- Click Start. A few seconds later you get one row per video with the transcript, author, views, likes and publish date.
- Export it as CSV, JSON or Excel, or send it straight to Google Sheets.
Only need one platform? The single-platform editions run the same engine for $5 per 1,000 videos instead of $6: Instagram Reels Transcript and TikTok Transcript, which also takes whole TikTok profiles.
For the NASA Hubble Reel above, the transcript starts like this:
"Just a few years after its launch in 1990, NASA's Hubble Space Telescope made a groundbreaking discovery..."
Option 2: JavaScript
import { ApifyClient } from 'apify-client'; const client = new ApifyClient({ token: process.env.APIFY_TOKEN }); const run = await client.actor('aemon/video-to-text-for-ai').call({ urls: [ 'https://www.instagram.com/reel/DMflZnXOKUj', 'https://www.tiktok.com/@apifylife/video/7338085038258457889', ], subtitleFormat: 'srt', aiAnalysis: true, // summary, hook, key points and call to action }); const { items } = await client.dataset(run.defaultDatasetId).listItems(); for (const video of items) { if (video.status === 'ok') console.log(video.platform, video.author, video.aiAnalysis?.hook, video.text); else console.log('Failed:', video.url, video.errorCode); } Option 3: Python
from apify_client import ApifyClient client = ApifyClient("YOUR_APIFY_TOKEN") run = client.actor("aemon/video-to-text-for-ai").call( run_input={"urls": ["https://www.tiktok.com/@apifylife/video/7338085038258457889"]} ) for video in client.dataset(run["defaultDatasetId"]).iterate_items(): print(video["author"], video["wordCount"], video["text"][:200]) What the output looks like
{ "status": "ok", "platform": "instagram", "author": "Hubble Space Telescope", "publishedAt": "2025-07-24T14:16:39.000Z", "durationSec": 152, "likes": 11256, "language": "en", "transcriptSource": "whisper", "text": "Just a few years after its launch in 1990, NASA's Hubble Space Telescope made a groundbreaking discovery...", "wordCount": 307, "segments": [ { "start": 0, "end": 7.76, "text": "Just a few years after its launch in 1990, NASA's Hubble Space Telescope made a groundbreaking discovery." }, { "start": 8.52, "end": 14.34, "text": "Nearly every galaxy in the universe appears to have a supermassive black hole at its center." } ], "aiAnalysis": { "summary": "The video explains why intermediate-mass black holes are hard to find and how Hubble and Chandra caught one tearing apart a star in galaxy NGC 6099.", "hook": "Just a few years after its launch in 1990, NASA's Hubble Space Telescope made a groundbreaking discovery.", "keyPoints": ["Nearly every galaxy has a supermassive black hole", "Intermediate-mass black holes only shine when they consume a star"], "callToAction": null, "topics": ["black holes", "hubble", "astronomy"], "contentType": "educational", "sentiment": "positive" } } Videos that are private, removed or have no audio come back as rows with "status": "error" and an error code, so a batch never silently loses links.
Doing it at scale: from a creator or hashtag to transcripts
For a single creator you can paste the TikTok profile or YouTube channel link directly, set Max videos per URL (for example 50) and optionally Only videos published after a date.
Transcribing a whole niche is a two-step pipeline:
- Collect video URLs with a scraper such as TikTok Scraper (hashtags, searches) or Instagram Reel Scraper (Instagram only lists a profile's posts to logged-in users, so profiles have to go through a scraper).
- Pass the URLs into Video to Text with AI analysis turned on.
Every row then already has its hook, key points and call to action. Feed them to an LLM with a prompt like:
"Here are 200 transcripts of top-performing TikToks about home workouts. List the 10 most common hooks in the first 3 seconds, with an example of each."
You get a research report on what actually works in a niche in minutes instead of days of watching videos.
Using it from an AI agent
The Actor is available through the Apify MCP server, so an agent in Claude, ChatGPT or Cursor can transcribe videos on its own: give it a link, and it gets back the transcript as JSON.
How much does it cost?
You pay only for successful transcripts: $0.006 per video up to 2 minutes (about $6 per 1,000 Reels or TikToks), plus $0.003 for each extra minute of AI transcription. The add-ons are optional: on-screen text costs $0.002, AI analysis $0.004 per video and translation $0.003 per 1,000 words. Failed videos are free, and the Apify free plan's monthly credit covers roughly 800 short videos.
Working with podcasts instead? The companion Podcast to Text Actor takes Apple Podcasts links and RSS feeds and charges per minute of audio.
Wrapping up
Turning short-form video into text unlocks it for search, analysis and AI. Use platform captions when they exist, Whisper when they don't, and keep the output structured so it plugs straight into your pipeline.
What would you build with transcripts of thousands of short videos? I'd love to hear your use cases in the comments. And if a link doesn't work, open an issue on the Actor page.
Disclosure: I'm the author of the Video to Text Actor used in the examples, published on Apify under my agency Aemon.

