Summary of an article published on The Web Scraping Club on 2025-06-20 (for paid subscribers). Read the full article

More and more people in the TWSC Discord server and on Reddit are struggling with ReCAPTCHA, especially V3. Google held around 98% of the CAPTCHA market as of 2022, so sooner or later a scraper runs into it. The three versions are very different targets, and it pays to know how each one decides who is human before choosing a tool.

I start with the history, from v1’s book-digitizing word pairs to v2’s checkbox and v3’s invisible risk score, and then explain how each version works technically. After that I cover the alternatives: hCaptcha, Cloudflare Turnstile, Friendly Captcha, Geetest and the bot management vendors. Finally, we test open-source packages and commercial solutions against a real-world use case.

What’s inside

  • How v3 turns telemetry into a score from 0.0 to 1.0, what site owners do with a threshold, and why Google suggests embedding it on every page.
  • What ReCAPTCHA Enterprise adds on top of v3, including the fallback to a v2 challenge when the invisible score is not confident.
  • Why Cloudflare dropped Google in 2020, and how Turnstile now relies on rotating background challenges and Private Access Tokens developed with Apple.
  • Different ways to avoid a human puzzle: proof-of-work in Friendly Captcha, sliders in Geetest, and silent fingerprinting in bot management products.

In the full article: The real-world use case, the open-source packages and commercial solutions we tested with their results, and the scripts in the 87.RECAPTCHA folder of the GitHub repository.

Read the full article on The Web Scraping Club