A perfect scheme for providing Recaptcha v3 solutions looks like this: we periodically test workers on special pages with a Recaptcha v3 widget and record their scores. If they get 0.3, 0.7, or 0.9, we add them to one of our corresponding solving queues. Customers send their v3 tasks, workers load widgets, special software intercepts the g-response tokens, and we send them back to customers.
In reality, this doesn’t work absolutely perfectly because Google's AI measures user scores differently across different website domains and even within a single domain name.
For instance, 2 pages like
might provide different user scores.
Obviously, their AI measures not only the probability of single-user spam but also how often probable spambots visit target pages, and even the page content.This is why we can't guarantee a 100% correct score from our workers.
You have to periodically compare different captcha services, check v3 score quality, and switch between them.
The good news is that you can complain
about tokens that a target page wouldn’t accept and receive a refund. This way, you only pay for correctly solved Recaptchas.
And you can also report correctly
solved tokens with reportCorrectRecaptcha
method. This will bookmark the worker and put him or her in the front line for your next tasks.
At the moment, the percentage (%) of success/fails+success is considered OK if it’s higher than 20%.
If it’s lower in your case, consider reviewing your code – your automation is being detected in some other way.