Video Ad Analyzer: Facebook & TikTok Ad Transcripts & Hooks

Video ad analyzer for competitor research: turn the video ads you scraped from the Facebook Ads Library, Meta Ad Library, TikTok Ads Library or any ad-library scraper into structured data — the full transcript with word timing, a timed timeline of what is on screen, the 3-second hook, spoken calls to action and brand mentions, and the ad's pacing. One row per ad creative, ready for a spreadsheet, a dashboard, or ChatGPT and Claude.

deAPI AI tools

• Transcribing any audio or video file? → Audio & Video Transcription: MP3, MP4 & Podcast to Text
• Analysing TikTok, YouTube or X videos by link? → AI Video Analyzer

What does Video Ad Analyzer do?

Ad-library scrapers give you thousands of ads with a link to each video — but not what the video says or shows. This Actor takes those video files, listens to them and watches them, and returns what a creative strategist would note down for every ad:

• What is said — the full transcript, with the time of every sentence and every word, and the language.
• What is shown — a timeline of on-screen moments, one to a few seconds each: people, products, actions, settings, and the text and brand names it can read.
• The hook — what is said and shown in the first 3 seconds, and how long until the first word.
• Calls to action — when "shop now", "use code", "link in bio", "download" and the ad's own CTA button text are spoken.
• Brand mentions — how often and how early the advertiser's name is said and shown.
• Pacing — words per minute, the share of the ad with someone talking, on-screen moments per 10 seconds, length and aspect ratio.

Each row also keeps the ad's own data from the scraper — ad ID, advertiser, CTA button, landing page, ad text, start date and platforms — so you can sort, filter and compare creatives in one table.

Why use it?

• Works on the ads you already scraped — give it the dataset of Facebook Ads Library Scraper or another ad-library scraper; it finds every video and its ad data.
• One analysis per creative — ad libraries list the same creative in 9:16 and 1:1. Versions of the same length but another shape are analysed once (the vertical cut), and the copies are free.
• Timed, not just transcribed — every word and every on-screen moment has a timestamp, so "what happens in the first 3 seconds" is a field, not a guess.
• Reads the screen — the HD video is analysed, so product names, prices and title cards are picked up. Detailed reading catches small text and logos.
• Silent ads too — music-only and text-on-screen ads get the full on-screen timeline, and videos without a soundtrack skip the transcript fee.
• Priced per ad — about $0.02 per ad: $0.021 for a 30-second ad. Failed and skipped videos are free.

What data can you get?

Data	Fields
🏷️ Ad	adId, advertiser, ctaText, linkUrl, adText, startDate, platforms (from the scraper), videoUrl, videoNumber, videosInAd
🎞️ Video	durationSeconds, width, height, aspectRatio (9:16, 1:1, 4:5, 16:9…), hasAudio
🗣️ Transcript	transcript, language, segments (sentences with start and end), words (every word with start and end), hasSpeech
👁️ On screen	onScreenEvents (start, end, description), sceneSummaries (longer descriptions of each part)
🪝 Hook	hook.speech and hook.onScreen for the first 3 seconds, timeToFirstWordSeconds
📣 CTA & brand	ctaMentions (phrase and the seconds it is said), brandMentions (spoken and on-screen counts, first times)
⏱️ Pacing	wordsPerMinute, speechCoveragePct, onScreenEventCount, onScreenEventsPer10s
💵 Billing	billedVideos, billedTranscripts, billed10sUnits, billedDetailUnits, billedConversionUnits

How to analyse video ads

1. Scrape the ads. Run an ad-library scraper — for example Facebook Ads Library Scraper with an Ad Library search or page URL filtered to Media type: Videos.
2. Give this Actor the scraper's dataset. Pick it in Dataset of ads, or paste its ID. To do it automatically, connect this Actor as an integration of the scraper (see the tutorial below).
3. Click Start. To see what you get first, run it with the prefilled example — a public-domain U.S. Army TV ad — then swap in your scraper's dataset. Each ad video becomes a row; open the Ad videos, Transcripts or On-screen timeline views, or export everything as JSON, CSV or Excel.

Run it soon after the scraper. Ad-library video links are signed and expire after a few days (about four for Meta's). An expired link is a free failed row.

You can also paste direct video links in Video URLs — any public MP4, MOV, WebM, MKV, AVI or MPEG file.

Which scrapers work

Any scraper whose results contain links to the video files. The Facebook Ads Library Scraper layout is tested end to end on real ads; the others are recognised by their field names:

Scraper	Videos found in	Ad data kept
Facebook Ads Library Scraper (apify)	snapshot.videos[], snapshot.cards[] (videoHdUrl, videoSdUrl)	ad ID, page name, CTA, link, body text, dates, platforms
Meta Ad Library scrapers (snake_case)	video_hd_url, video_sd_url, videos[]	ad_archive_id, page_name, cta_text, link_url…
TikTok Ads Library scrapers	videoUrl, video_url, videos[]	adId, advertiserName…

The HD copy is used when there is one (it lets the analysis read the screen), and the SD copy if the HD link fails. Image and text ads get a free skipped row, so every ad in the dataset is accounted for.

Limits

Video length	up to 5 minutes (free plan: up to 60 seconds)
File size	up to 150 MB (free plan: 50 MB)
Large files	files over 48 MB are converted first, which takes several minutes; 1080p ads of up to about 4 minutes and 4K ads of up to about 1 minute are taken (with a conversion fee from November 11, 2026), heavier files are refused at once and not charged
Formats	MP4, MOV, WebM, MKV, AVI, MPEG
Ads per run	up to 1,000 ads with video (maxAds, default 50), up to 10 videos per ad
Languages	many languages, detected per video, or set language

Input

Field	Values	Notes
datasetId	dataset ID or link	The dataset of an ad-library scraper run.
videoUrls	list of links	Direct links to video files.
preset	balanced (Standard) · detail (Detailed)	On-screen reading. Detailed reads small text and logos, for a surcharge. Default balanced.
language	en, es, de…	Two- or three-letter code of the spoken language. Empty = detect per video.
analyzeAllVariants	true · false	Analyse every version of a creative instead of one. Default false.
maxAds	1–1,000	Video ads taken from the dataset. Default 50.
maxConcurrency	1–5	Ads at the same time. Default 3. No effect on price.

{
    "datasetId": "YOUR_SCRAPER_DATASET_ID",
    "preset": "balanced",
    "maxAds": 50
}

Output

A real row, for a public-domain U.S. Army TV ad from 2012 (lists shortened):

{
    "source": "link",
    "videoUrl": "https://upload.wikimedia.org/wikipedia/commons/4/4b/Army_Strong_TV_ad%2C_2012.webm",
    "durationSeconds": 30.15,
    "width": 640,
    "height": 360,
    "aspectRatio": "16:9",
    "status": "success",
    "hasSpeech": true,
    "language": "en",
    "transcript": "It's back up. This is a breakthrough. It's a magnet. It's a game changer. It's relief. It's a role model. It's progress. It's more than a uniform. It's a chance to change the future. There's strong, and then there's Army strong. Try it on at GoArmy.com.",
    "words": [{"word": "It's", "start": 1.771, "end": 2.111}, {"word": "back", "start": 2.272, "end": 2.492}, "…"],
    "onScreenEvents": [
        {"start": 0.0, "end": 2.0, "description": "A soldier in camouflage fatigues adjusts his gear."},
        {"start": 2.0, "end": 3.5, "description": "A soldier runs across a dusty outdoor field."},
        {"start": 3.5, "end": 5.5, "description": "A soldier raises his hand in a salute."},
        {"start": 5.5, "end": 7.5, "description": "A soldier smiles while looking down."},
        "…"
    ],
    "hook": {
        "seconds": 3.0,
        "speech": "It's back This is a breakthrough.",
        "onScreen": ["A soldier in camouflage fatigues adjusts his gear.", "A soldier runs across a dusty outdoor field."]
    },
    "timeToFirstWordSeconds": 1.66,
    "wordsPerMinute": 91.5,
    "speechCoveragePct": 35.4,
    "onScreenEventCount": 14,
    "onScreenEventsPer10s": 4.64,
    "ctaMentions": [{"phrase": "try it", "atSeconds": [27.06]}],
    "brandMentions": null,
    "billedVideos": 1,
    "billedTranscripts": 1,
    "billed10sUnits": 3,
    "billedDetailUnits": 0,
    "billedConversionUnits": 0,
    "error": null
}

The hook follows the word timings: words that start inside the first 3 seconds, in spoken order.

A row from an ad-library dataset also carries the ad's data — for example "adId": "2055369298426424", "advertiser": "Sephora", "ctaText": "Shop now", "linkUrl", "adText", "platforms": ["FACEBOOK", "INSTAGRAM"] — and brandMentions — for a 34-second Sephora ad from the Ad Library, {"name": "Sephora", "spokenCount": 4, "firstSpokenAtSeconds": 5.84, "onScreenCount": 3, "firstOnScreenAtSeconds": 3.5}.

Videos that could not be analysed have status: "failed" and an error. Copies of a creative, repeated videos, ads without a video, and videos left out by a run cost limit or the free plan have status: "skipped". Neither is charged.

How much does it cost?

$0.00684 per analysed video + $0.01032 for its transcript + $0.00134 per started 10 seconds of video. Videos without a soundtrack have no transcript and skip that fee. Detailed on-screen reading adds $0.00043 per 10 seconds.

Ad length	Standard	Detailed	No soundtrack
up to 15 s	$0.0198	$0.0207	$0.0095
up to 30 s	$0.0212	$0.0225	$0.0109
up to 60 s	$0.0252	$0.0278	$0.0149
up to 2 min	$0.0332	$0.0384	$0.0229
100 ads of 30 s	$2.12	$2.25	$1.09

Length is counted in started 10-second blocks, with half a second of slack: a 30.3-second ad bills as 30 s, a 31-second ad as 40 s.

Large files (from November 11, 2026): a file over 48 MB has to be converted before it can be analysed, which takes several minutes of processing. Such a file adds $0.0025 per started 10 seconds at 1080p and 30 fps — 4K counts four times, 720p less than half, 60 fps twice. A 3-minute 1080p file at 25 fps adds $0.0375, a 1-minute 4K file $0.06. Files up to 48 MB, which covers almost every ad from an ad library, never pay it; the units are in billedConversionUnits.

The price covers the analysis. Ads collected with an ad-library scraper also cost whatever that scraper charges for its own run. Video files you link directly — your own ads, files on your server or cloud storage — need no scraper, so the price above is all you pay.

Failed videos — expired links, files that are not a video, videos over 5 minutes — are never charged, and neither are skipped copies of a creative, ads without a video, or videos left out by a run cost limit or the free plan. A soundtrack with music only costs the same as one with speech.

Set maximum cost per run to cap your spend: the Actor checks each video's price before it starts and stops at the first one the remaining budget cannot cover; that video and the rest get a free skipped row.

Free plan: 5 ad videos per calendar month (resets at 00:00 UTC on the 1st), each up to 60 seconds and 50 MB, within a daily capacity shared by all free users. Any paid Apify plan removes the free-plan limits.

Use cases

• Competitor ad research — see what your competitors say in their first 3 seconds, how fast they talk, and when they ask for the click.
• Creative strategy — compare hooks, CTAs and pacing across hundreds of ads in one table; sort by timeToFirstWordSeconds or onScreenEventsPer10s.
• Swipe files — keep every competitor ad's script and shot timeline, searchable.
• AI briefs — give the rows to ChatGPT or Claude and ask for angles, offers and new hook ideas grounded in what is really in the ads.
• Brand and compliance checks — find where a brand name, a price or a claim is said or shown.

Tutorial: analyse a competitor's video ads

1. In the Console

Run Facebook Ads Library Scraper on a brand's Ad Library page URL filtered to Media type: Videos. Open this Actor, pick that run's dataset in Dataset of ads and click Start. Each video ad becomes a row in the Ad videos view.

2. Automatically after every scrape

In the scraper's Integrations tab, add Run Actor, choose this Actor, and set its input to {"datasetId": "{{resource.defaultDatasetId}}"}. Every scrape is then analysed while its video links are fresh.

3. From Python

from apify_client import ApifyClient

client = ApifyClient("<YOUR_APIFY_TOKEN>")
ads = client.actor("apify/facebook-ads-scraper").call(run_input={
    "startUrls": [{"url": "https://www.facebook.com/ads/library/?active_status=active&ad_type=all&country=US&media_type=video&q=skincare&search_type=keyword_unordered"}],
    "resultsLimit": 20,
})
run = client.actor("deapi/video-ad-analyzer").call(run_input={"datasetId": ads["defaultDatasetId"]})
for row in client.dataset(run["defaultDatasetId"]).iterate_items():
    if row["status"] == "success":
        print(row.get("advertiser"), row["durationSeconds"], row["hook"]["speech"])

4. From JavaScript or cURL

import { ApifyClient } from 'apify-client';

const client = new ApifyClient({ token: '<YOUR_APIFY_TOKEN>' });
const run = await client.actor('deapi/video-ad-analyzer').call({
    videoUrls: ['https://upload.wikimedia.org/wikipedia/commons/4/4b/Army_Strong_TV_ad%2C_2012.webm'],
});
const { items } = await client.dataset(run.defaultDatasetId).listItems();
console.log(items[0].hook, items[0].onScreenEvents.length);

curl -X POST "https://api.apify.com/v2/acts/deapi~video-ad-analyzer/run-sync-get-dataset-items?token=<YOUR_APIFY_TOKEN>" \
  -H "Content-Type: application/json" \
  -d '{"videoUrls": ["https://upload.wikimedia.org/wikipedia/commons/4/4b/Army_Strong_TV_ad%2C_2012.webm"]}'

FAQ

How do I analyse Facebook ads?

Scrape them with a Facebook Ads Library scraper, then give its dataset to this Actor. It downloads each ad's video, transcribes it, describes what is on screen and measures the hook, and keeps the ad's ID, page, CTA and text next to the results. It does not search the Ad Library itself.

How much does it cost to analyse a Facebook ad?

About $0.021 for a 30-second video ad and $0.025 for a 60-second one; $0.0185 for the shortest. See the price table above; failed and skipped videos are free.

Can I get the transcript of a Facebook or Instagram ad?

Yes — every analysed ad has its full transcript with word timing in transcript and words. Meta ads run on Facebook and Instagram with the same video, so one analysis covers both.

Can I transcribe Facebook ads in bulk?

Yes. One run takes up to 1,000 video ads from a scraper's dataset, several at a time, and the results export as one CSV, Excel or JSON file.

Do I need an ad-library scraper?

No. Paste direct links to video files in Video URLs — your own ads before a campaign, files on your server or cloud storage, or any public video file — and the Actor analyses them as they are. It is also the cheapest way: you pay only the analysis. A scraper is needed only to collect other advertisers' ads from an ad library, which shows no direct video links; it bills its own runs separately.

Does it work with TikTok ads?

Yes, with a TikTok Ads Library scraper whose results include the video file link. For organic TikTok, YouTube or X videos by their page link, use AI Video Analyzer.

Does it work with YouTube or Google Ads Transparency ads?

Only when the scraper returns direct video files. Ads given as YouTube links cannot be downloaded here — analyse those with AI Video Analyzer, which takes YouTube links.

What is the hook of an ad?

The first seconds that decide whether someone keeps watching. The Actor records what is said in the first 3 seconds (hook.speech), what is shown (hook.onScreen) and when the first word comes (timeToFirstWordSeconds).

Does it read text on screen?

It describes what is on screen and reads larger text — title cards, product names, prices — in the standard mode. Detailed reading looks at the picture at a higher resolution and catches small text, labels and logos. Text is reported inside the on-screen descriptions, not as a separate subtitle file.

Does it work on ads without speech?

Yes. Music-only and text-on-screen ads get the full on-screen timeline; hasSpeech is false and the transcript is empty. Videos with no soundtrack at all are not charged the transcript fee.

Why is brandMentions 0 for my ad?

Brand mentions are matched on the advertiser's page name as the ad library lists it, as whole words. A page called "Nike Running" does not count a spoken "Nike".

Why was a video skipped as a copy?

Ad libraries list one creative several times — for example the 9:16 and 1:1 cuts of the same video. Videos of the same length (within half a second) but a different shape count as one creative: the vertical one is analysed, and the others get a free row with copyOfVideoNumber. Videos of the same length and shape are treated as different cards and all analysed. Turn on Analyse every version of a creative to analyse every version.

Why did a video fail?

Most often the ad-library link has expired — run the Actor soon after the scraper. Other causes: the link opens a web page rather than the file, the file is damaged, or the video is longer than 5 minutes. Failed videos are never charged; the reason is in error.

Is it free to try?

Yes. On Apify's free plan you can analyse 5 ad videos a month, each up to 60 seconds.

Can I use it with ChatGPT or Claude?

Yes. Export the dataset as JSON or CSV and give it to the model with a question, or call the Actor through the Apify MCP server from your AI assistant. The timed transcript and on-screen timeline give it the facts of each ad to reason about.

Is this legal?

The Actor analyses video files you give it. Ads in public ad libraries are published for transparency; check the terms of the source and your local law before reusing creatives, and do not use the results to copy protected work.

Support / feedback

Found a bug or need another scraper's fields supported? Open an issue on the Actor's Issues tab.