Instagram Transcript Scraper

Instagram Transcript Scraper: the spoken words out of any public reel or video

Paste reel or video links and get what was actually said, as plain text plus timed segments, SRT and WebVTT subtitles. Each row also carries the creator's caption, the hashtags, the author and the engagement counts, so the transcript arrives with its context attached.

The part to sort out before your first run: Instagram publishes no transcript of its own, so the audio has to be listened to by a speech-to-text service. You bring your own key, in transcriptionApiKey, and you pay your provider directly at their rate. Without a key the run returns one labelled sample row and nothing else.

🔍 What Instagram Transcript Scraper does

It resolves each link, pulls the audio, sends it to the endpoint you named, and writes the result as a row. Works with OpenAI, Groq, Deepgram's OpenAI-compatible endpoint, any audio-capable model behind an OpenAI-compatible gateway, or your own self-hosted Whisper server.

Two shapes of endpoint exist and the run tries the better one first. /audio/transcriptions gives you per-cue timings, which is what makes segments, srt and vtt possible. A gateway that only implements /chat/completions still gives you the text, with srt and vtt coming back null. transcriptSource on every row tells you which you got.

No Instagram account needed. The cookie field is optional and separate from your speech-to-text key.

Your key is used for the requests and nothing else. It is never written to the dataset, never logged, and error text from your provider is scrubbed of anything key-shaped before it reaches a diagnostic row.

📥 What you give it

{
  "postUrls": ["https://www.instagram.com/reel/DcBsBkPOtmB/"],
  "transcriptionApiKey": "<your key>",
  "transcriptionBaseUrl": "https://api.openai.com/v1",
  "transcriptionModel": "whisper-1",
  "maxItems": 10
}

Start a run with no links and no key and you get one labelled sample row showing the output shape. That row is not charged.

📤 What you get back

Every run on this account so far has been a shape check rather than a paid transcription, so what follows is the actor's own labelled sample row from a recent run, trimmed, with one emoji cut out of the caption. A real transcript row has exactly this shape with _sample absent.

{
  "ok": true,
  "_sample": true,
  "recordType": "transcript",
  "shortcode": "DcBsBkPOtmB",
  "url": "https://www.instagram.com/reel/DcBsBkPOtmB/",
  "authorUsername": "example_creator",
  "authorIsVerified": true,
  "caption": "The one habit that changed my mornings ... #habits #morningroutine",
  "hashtags": ["habits", "morningroutine"],
  "durationSeconds": 41.6,
  "createdAt": "2026-08-02T15:11:04.000Z",
  "productType": "clips",
  "language": "en",
  "text": "Every one of them. You can't skip. It has to be about your long-term legacy.",
  "wordCount": 38,
  "characterCount": 196,
  "segmentCount": 3,
  "segments": [
    {"start": 0, "end": 2.8, "startTime": "00:00:00.000", "endTime": "00:00:02.800", "text": "Every one of them. You can't skip."}
  ],
  "srt": "1\n00:00:00,000 --> 00:00:02,800\nEvery one of them. You can't skip.",
  "vtt": "WEBVTT\n\n00:00:00.000 --> 00:00:02.800\nEvery one of them. You can't skip.",
  "transcriptSource": "speech-to-text (timestamped)",
  "audioSource": "audio track",
  "audioBytes": 472064,
  "viewCount": 1840000,
  "likeCount": 92100,
  "commentCount": 431,
  "musicTitle": "original sound",
  "scrapedAt": "2026-09-16T14:07:27.343Z"
}

🧾 Reading the output

Three kinds of row land in your dataset, and only the first kind is billed.

A row is only billed when the transcript text is not empty. Rows also carry a charged flag, which is how the actor labels real rows against samples and diagnostics.

No diagnostic row is billed.

▶️ How to run it

1. Open Instagram Transcript Scraper and click Try for free. 2. Paste your links into Instagram reels or videos. 3. Put your key in Your speech-to-text API key, and set the endpoint and model if you are not using OpenAI. 4. Set Maximum transcripts to 1 for the first run, so a wrong key costs you one attempt. 5. Click Start, then download the dataset as JSON, CSV or Excel.

💰 How much does it cost?

$2.50 per 1,000 transcripts. Flat on every Apify plan, no volume tiers.

On top of that you pay your own speech-to-text provider, at their rate, on your own account. That is the whole reason the key field exists.

A row is billed only when there is transcript text in it. A photo, a silent reel, a video skipped for length, a failed call to your endpoint: none of those are billed here, though a failed call may still count against your provider.

💡 What people use it for

• Turning a saved folder of reels into searchable text, so you can grep six months of a niche.
• Pulling the hook from the first three seconds of a set of reels to see what actually opens them.
• Building subtitle files for reels you have the rights to repost.
• Feeding spoken claims into a review process, where the caption says one thing and the voiceover

says another.

🚧 What it does not do

• It will not work without your own speech-to-text key. No key, no transcript.
• Accuracy is your provider's, not ours. Accents, background music and crosstalk hurt, and a

different model gives a different result on the same audio.

• No timings from a chat-shaped gateway. You get the text, and srt and vtt come back empty.
• No photos or carousels. They have no audio.
• Audio over 24 MB is refused rather than truncated.
• maxDurationSeconds is a guard, not a guarantee. Instagram does not always publish a duration

to a signed-out reader, and when it does not, the length check has nothing to go on.

• No private accounts.
• No speaker labels and no translation. One language in, the same language out.

🧭 Which Instagram scraper do you need?

❓ Questions people ask

Which key do I need? Any OpenAI-compatible speech-to-text key. OpenAI with whisper-1 is the default path. Groq, Deepgram's OpenAI-compatible endpoint and a self-hosted Whisper server all work.

Is my key safe? It is a secret field on Apify, used for the requests and nothing else. It is never written to the dataset or the log, and provider error text is scrubbed before it is shown.

Why did every post fail? Almost always a wrong key, endpoint or model name. Run one link first and read the TRANSCRIPTION_FAILED row, which carries the reason.

Do I need an Instagram account? No. The cookie field is optional and only for posts Instagram will not show a signed-out reader.

Can I get subtitles I can upload? Yes, when your endpoint returns timings. Save the srt or vtt field straight to a file.

Is scraping Instagram legal? This reads public posts only, never private accounts or messages. Results can still contain personal data, which GDPR and similar laws cover, so have a reason for collecting it. Apify's write-up on the legality of web scraping is a good starting point, and we are not lawyers.

🆘 If something breaks

Open the Issues tab on the actor page. Send the link you used and the run ID, and never your key. The errorCode on the diagnostic row usually explains it on its own.