Google Ads Transparency Scraper + OCR

Collect public Google Ads Transparency Center creatives by business domain or advertiser ID. Get advertiser names, creative IDs, ad formats, dates, preview links, available ad text and video URLs. Turn static ad screenshots into searchable text with optional OCR.

Use the results for competitor research, creative archives, reporting and scheduled monitoring. Export the dataset as JSON, CSV or Excel through Apify, or run it from an API, integration or AI agent.

Quick start

1. Enter a business domain such as shopify.com, or an exact AR... advertiser ID.
2. Choose a country and how many ads to return.
3. Enable preview decoding and image text extraction for richer text results.
4. Run the Actor and open the dataset. Check SUMMARY for coverage and errors.

{
  "targets": ["shopify.com"],
  "region": "US",
  "maxAds": 5,
  "maxAdsPerTarget": 5,
  "includeDetails": true,
  "decodePreviews": true,
  "extractImageText": true,
  "ocrLanguages": ["eng"]
}

Start with five ads and 512 MB memory for OCR. More ads, variants and OCR take longer. For fast metadata-only collection, disable details, preview decoding and OCR. API defaults keep preview decoding and OCR off unless explicitly enabled.

No Google login or cookies are required. Company names are not accepted because they are ambiguous. Exact advertiser URLs without query filters are also supported.

Output

Field	Meaning
advertiserId, advertiserName	Actual advertiser identified by Google
creativeId, adUrl	Creative identity and source link
format	TEXT, IMAGE, VIDEO or UNKNOWN
headline, description	Text extracted from supported source fields
productTitle, merchantName	Available shopping-renderer fields
ocrText, ocrStatus, ocrConfidence	Separately labelled text recognized from images
videoUrl	YouTube URL when exposed by a supported renderer
previewUrl, variants	Source preview links and recognized variants
firstShown, lastShown	UTC dates from source search results
regionsShown	Countries/region IDs available in creative details
detailStatus, previewStatus	Whether enrichment worked or was incomplete
changeStatus	Optional comparison with earlier observations

Domain searches can include affiliates and resellers. Use an advertiser ID for one specific advertiser. A display URL is not a verified landing-page URL. Preview links can expire.

Image text and OCR

Google sometimes serves TEXT ads as screenshots with no extractable headline. Enable extractImageText to return recognized text in ocrText. The Actor does not guess which recognized line is the headline.

Choose up to three models: eng, tur, heb, fra, deu, spa, ita, por, nld. For Turkish and English, use ["tur", "eng"]. Language detection is not automatic.

OCR reads at most the first two eligible variants and reuses identical images within a creative. It runs inside the Actor without a paid external AI service. It can misread characters, include logo/photo noise or miss small text. Confidence is an engine score, not measured accuracy. Statuses include ok, low_confidence, partial, failed, no_text and no_eligible_image.

Preview scripts are parsed as text, never executed. Video transcription, landing-page scraping and image-pixel change detection are not included. Some renderers remain unsupported.

Filters and coverage

Supported country filters: US, GB, DE, FR, IL, CA, AU, NL, BE, ES, IT, or ANYWHERE.

startDate and endDate apply an inclusive overlap filter to the source first/last dates after collection. They do not prove daily or continuous activity. Page limits can leave incomplete coverage.

SUMMARY reports returned rows, requests, duplicates, date-filtered rows, failures and the reason collection stopped. complete=false means the scan was capped or interrupted. A capped sample is not all ads for an advertiser. A failed source request is never reported as proof that no ads exist.

Use maxAds, maxAdsPerTarget, maxPagesPerTarget, maxRequests and maxRunSeconds to bound work. Image downloads and retries count against requests; OCR respects the runtime limit. When per-result pricing is active, the Actor also stops at the customer's spending limit.

Optional observed-change monitoring

Reuse monitorStoreName with identical targets, region, dates and enrichment settings. Run only one job at a time per named store.

• baseline: first observations in a new store.
• newly_observed: not previously seen by this monitor; the ad may have existed earlier.
• unchanged / content_changed: comparison of supported extracted source text, format and video fields.
• unverified: insufficient or uncertain source content. OCR-only text does not prove unchanged content.

Missing creatives are never labelled stopped, removed or inactive. Partial scans retain earlier observations. OCR text, image pixels, signed URLs and date updates are excluded from the hash. Use a new store when changing query/enrichment settings. Stores are limited to 50,000 creatives.

Pricing

The active price is shown in Apify's Pricing tab. With per-result pricing, one result is one unique advertiser/creative pair saved in the run's default dataset. Details, preview decoding and optional OCR do not create additional result charges. A base record may have missing text or failed enrichment; inspect its statuses.

No result event is charged for a filtered duplicate, date-filtered item or source failure producing no record. Any configured Actor-start fee still applies. Each separate monitoring run charges for its returned records, including unchanged records. Reports in the key-value store are not result rows.

Start a fresh run after interruption. Resurrecting a run with an existing dataset is refused to preserve output and avoid duplicate charges. Automatic continuation is not supported in this version.

Validation and support

Private validation covered 1,000 metadata records, two-page detailed collection, country filters, monitoring and image OCR. In one 29-row US sample, source text was available on 5 rows and source/OCR text on 28. This is sample-specific availability, not a guarantee of accuracy or coverage.

For support, include the run ID, input and SUMMARY. Never include passwords, cookies or API tokens. Google can change its endpoints or restrict requests. Bounded retries and redirect diagnostics help explain failures but cannot guarantee access.

This is an independent tool, not a Google product. Protocol attribution is in THIRD_PARTY.md.