I think you're right: some sort of personalisation could well be contributing to the discrepancy. P.S. Driven by curiosity, I repeated my experiment from an in

I think you're right: some sort of personalisation could well be contributing to the discrepancy. P.S. Driven by curiosity, I repeated my experiment from an incognito window and it yielded the same results. This time I looked at the results beyond page one, and at least the first few pages of results are 100% relevant.