# Changelog of Google Ads Transparency Scraper | Ad Text + Domain Check (`zrk1212/google-ads-transparency-ad-copy`) Actor

- **URL**: https://apify.com/zrk1212/google-ads-transparency-ad-copy/changelog.md
- **Full Actor documentation**: https://apify.com/zrk1212/google-ads-transparency-ad-copy.md

## Changelog

### 0.1 (2026-11-18, first private build)

- Ads mode, domain check mode, monitor mode (`baseline`, `new`, `stopped`), ad text from Google's preview data and OCR.
- Routing: Google flags direct connections from Apify containers for this site (Step 0 gate, 2 runs: 7 of 60 and 6 of 60 HTTP 200), so search calls go through Apify residential proxy (US, one sticky session per run, rotated on a block). Static files (screenshots, preview scripts) are fetched directly.
- OCR is stricter than the plan: a screenshot is accepted only when mean line confidence is 80 or more (plan said 60) so noisy local/map ads are not charged.
- Date filter history window measured: ads from Dec 2025 return, Oct 2025 and older return nothing (about 12 months).

#### Measured on the platform (2,048 MB, build 0.1.9, 2026-11-18)

| Run | Runtime | CU | Cost (USD) | Peak memory |
|---|---|---|---|---|
| Prefill (20 nike.com ads, 3 runs) | 11.6 s, 10.9 s, 19.2 s (one earlier run 49 s after a residential /sorry/ and slow OCR) | 0.006 to 0.011 | 0.0012 to 0.0015 | 181 to 198 MB |
| 3x input (60 ads, 52 OCR) | 82.8 s (OCR mean 1.49 s per image that run, noisy platform CPU) | 0.046 | 0.0069 | 220 MB |
| Cost run (300 text ads, 269 OCR, 287 with text) | 484 s | 0.269 | 0.0528 | 258 MB |
| Metadata-only (same 300 ads) | 12.5 s | 0.0069 | 0.0030 | 106 MB |
| Domain check (220 domains) | 150 s | 0.084 | 0.0311 | 95 MB |

- Base cost per 1,000 ads (metadata-only run, includes residential proxy): $0.0099. Residential traffic is about 425 bytes per ad.
- Cost per 1,000 OCR texts (cost run minus metadata-only run, divided by 269 OCR images): $0.185. Per 1,000 `ad-text` events (287): $0.174.
- Cost per 1,000 domain checks: $0.141 (about 16 KB of residential traffic per domain).
- OCR time per image: 0.43 to 0.75 s on quiet runs, up to 1.7 s on busy runs (grayscale + 0.6 scale).
- Residential `/sorry/` seen: 1 of 221 domain-check calls, 1 of 2 calls in one prefill run, 0 in the other runs. Direct route: not used by default.
- Linear scaling (3x input) did not hold within 20% because OCR speed varied between runs, not because of memory or request growth.
