“Helpful” is in the QRG 188 times
The raters are given so much instruction on understanding intent and which content is likely to meet it.
When raters are given SERPs to evaluate they review every page and, considering different user intents, they give their opinion on whether the page they are rating is likely to be helpful or not.
That rating doesn’t directly impact websites though.
It’s used to help train Google’s machine learning systems.
If the scores of the raters show that overall, search is improving at ranking content that is helpful, then the systems are working well.
If not, then the system can be fine-tuned.
It is suspected this is done with information in the grid the raters use to record their observations about a page.
Just like an LLM predicts how to end a sentence, Google’s machine learning systems predict how to present results that are likely to be helpful.
And just like an LLM, it’s not always going to be perfectly correct.
But the system continues to learn and improve.