Facebook Ads Library Scraper: Ads, Video Transcripts, Image Text & Google Ads

Scrape the public Facebook and Instagram Ad Library (Meta Ad Library) for any advertiser, page URL, page ID or keyword. You get every ad for $2.50 per 1,000 ads, plus the layers other scrapers leave to you: the landing page behind each ad, the same advertiser's Google ads, text printed inside image ads, video ad transcripts, and one advertiser verdict row (scaling, testing, steady, dormant) that says why.

No login, no cookies, no setup: paste an advertiser and run.

Free vs Paid

	Free plan	Paid plan
Advertisers or searches per run	1	up to 50
Ads per advertiser	20	up to 5,000 (default 200)
Landing page, Google ad, image text, transcript and AI creative analysis rows	1 sample row of each layer (the Google ads sample can be missing: the free plan asks Google without the proxy and may be refused), the rest counted as locked	every row
Date range and "only new since my last run"	locked (paid plans only)	yes
Price	list price x 1.5 ($3.75 per 1,000 ads)	list price ($2.50 per 1,000 ads, 33% cheaper per row)

A free run ends with free-plan-locked rows (for example 20 of 73 ads shown) and one free-plan-preview row that says what the same input returns on a paid plan, with an estimate in dollars: the total the same input costs on a paid plan (paidTotalUsd) next to what the free run cost (freeRunUsd). The free plan also reads one sample video ad (the sample video transcript) even when transcripts is off; it is charged as a normal video-transcribed event at the free price. Free runs finish as SUCCEEDED; nothing looks like an error.

What you pay (pay per event)

List price, what BRONZE and every higher plan pays. The free plan pays 1.5x.

Event	Charged for	Price
run-started	one run (also when the input has nothing to look up)	$0.005
ad-found	one ad row, all fields, image and video URLs included	$0.0025 (= $2.50 per 1,000 ads)
advertiser-profiled	one advertiser summary + verdict row	$0.005
landing-page-checked	one distinct landing URL that answered (any HTTP status)	$0.004
google-ad-found	one Google Ads Transparency Center creative of the advertiser's domain	$0.002
image-text-read	one ad image that carries readable text	$0.001
video-transcribed	one video ad that has speech	$0.015
ad-creative-analyzed	one distinct ad creative read by an AI model (the same text and image text is read and charged once)	$0.003

Not charged: notice rows, duplicate ads, pictures without text, videos without speech, and anything we could not fetch (our side). A page that answered with 403 or 404 is an answer and is a row.

Example, 50 ads of one advertiser with the default layers: $0.005 + 50 x $0.0025 + $0.005 = $0.135, plus landing pages, Google ads and image texts that were found.

What you get

1. Collect: every ad of an advertiser or keyword from the logged-out Ad Library (US by default, any country list), with creative text, headline, call to action, dates, run days, platforms, spend and reach when the library shows them, EU data, paid-for-by, report counts, image and video URLs. Creative files are linked, never copied.
2. Dedupe: one row per ad id, one row per landing URL (click-tracking parameters removed), one row per creative file.
3. Enrich (switch layers on or off with enrich):
   • landingPages: final URL, HTTP status, title, and product name and price when the page sells something (not only shop pages).
   • googleAds: the same advertiser's ads in Google Ads Transparency Center (format, first and last shown, advertiser, preview link).
   • imageText: the text printed on image ads (OCR, confident words only).
   • transcripts: speech in video ads (off by default; slower).
   • creativeAi: one row per distinct creative with hookType, offer, targetAudience, angle, emotion, claims and language, read by an AI model from the ad text, headline, button, image text, video speech and landing page. The same creative (same text and image text) is read and charged once. A creative with nothing to read, or a model failure, adds no row and no charge.
4. Judge: one advertiser row: active ads reported and collected, new ads in the last 30 days, longest-running ad, formats, platforms, top calls to action and landing domains, whether the advertiser also runs Google ads, and a verdict with a verdictBasis sentence. With creativeAi on, the row also carries topHooks (top three hook types and their share), dominantOffer (the offer that appears in at least two ads, otherwise null) and audiences, counted from the readings.

Every row carries sourceUrl, checkedAt, confidence and basis, so a person or an AI agent can see where it came from and why.

Input

• pageUrls (required): one per line. A Meta Ad Library URL, a Facebook page URL (facebook.com/nike), a page ID, or a search term.
• maxAdsPerTarget: default 200.
• enrich: extra layers; default landing pages, Google ads, image text and AI creative analysis.
• Advanced: countries, adStatus, maxTranscripts, adsAfter, adsBefore, onlyNewSinceLastRun, proxyConfiguration.

The run starts on its own network and switches to the residential proxy when Meta rate-limits it. Keep the proxy switched on (default).

Example rows

{"rowType": "ad", "adArchiveId": "1016451784037642", "pageName": "Nike", "isActive": true, "startDate": "2025-06-20",
 "runDays": 495, "publisherPlatforms": ["FACEBOOK", "INSTAGRAM"], "displayFormat": "DPA", "ctaText": "Shop now",
 "landingUrls": ["https://www.nike.com/us/en_us"], "imageUrls": ["https://scontent.../a.jpg"], "videoUrls": [],
 "sourceUrl": "https://www.facebook.com/ads/library/?id=1016451784037642", "checkedAt": "2026-10-29T04:42:29Z",
 "confidence": "high", "basis": "Read from the public Meta Ad Library page of this ad."}

{"rowType": "advertiser", "pageName": "Nike", "activeAdsReported": 73, "adsCollected": 30, "newAdsLast30Days": 29,
 "runsOnGoogleToo": true, "googleAdsFound": 20, "verdict": "scaling",
 "verdictBasis": "29 of 30 collected ads started in the last 30 days.", "confidence": "high"}

{"rowType": "creative-analysis", "pageName": "Gymshark", "adArchiveId": "1749316556368411", "adArchiveIds": ["1749316556368411"],
 "hookType": "social-proof", "offer": "Free delivery on orders over $75",
 "targetAudience": "Women who want gym and workout clothing and are inspired by Megan Thee Stallion's fitness lifestyle.",
 "angle": "Celebrity athlete endorsement paired with the idea that hot girl confidence comes from hard work at the gym.",
 "emotion": "pride", "claims": ["Megan Thee Stallion does gym and has the receipts to prove it", "Free delivery on orders over $75"],
 "language": "en", "inputSources": ["adText", "headline", "linkDescription", "cta", "landing"], "model": "claude-haiku-5-5",
 "sourceUrl": "https://www.facebook.com/ads/library/?id=1749316556368411", "checkedAt": "2026-10-29T06:05:04Z", "confidence": "medium",
 "basis": "Read by an AI model (claude-haiku-5-5) from: adText, headline, linkDescription, cta, landing."}

The creative analysis is a model's reading of the ad's own text, not a fact about the advertiser: confidence is medium and basis lists what was read. If the model is unavailable the layer adds nothing (a creative-ai-unavailable notice explains why) and nothing is charged for it.

Notes

• Unofficial tool, not affiliated with Meta Platforms or Google. It reads the public, logged-out Ad Library, the public Google Ads Transparency Center and the advertisers' public landing pages.
• You are responsible for a lawful purpose, for complying with the terms of those services and with the rights in the advertisers' pages and creatives, and for how you use the output.
• A run that was restarted by the platform does not charge any row twice and does not write any row twice.
• Two runs that start at the same time with the same input each deliver and each pay for their own rows.
• Memory: the default layers (landing pages, Google ads, image text, creative analysis) peaked at about 270 MB and 100 ads with the video layer at about 920 MB in our measurements, so the default run needs 2048 MB (the manifest default); choose 4096 MB for several advertisers with video transcripts at once. Pictures larger than 12 megapixels are not read for text (no row, no charge).