[gregallan]

I think this is my biggest problem with how Google now works. It's always been disappointing when you didn't find what you were looking for. But you used to be able to examine the results and see how your search terms might not have been optimal, and adjust accordingly. It was the expectation that you'd have to tweak. Now, changing your exact search terms hardly seem to make a difference.

I think the major difference is that the algorithm used to highly weight matching of specific words and phrases from the search terms, so adding a word, re-ordering, and swapping for synonyms would drastically change the results. Now it seems they're using ML and natural language processing to try to actually understand what you're looking for and give it to you. You can change your search terms, but the language embedding doesn't change much, so the system is actually working as intended. I could see that this might actually be desirable for a large segment of the population who wants their search engine to "just work" in response to natural language queries. If the corpus being indexed was high quality, maybe this would be a good experience. But due to the ads, affiliate marketing, and blogspam that make up a large part of modern internet content, it's simply frustrating.

I wouldn't be surprised if they've done user testing that validates their approach. Programmers tend to be comfortable with the concept that a computer will do what you ask, even if it's not what you meant, but most people want to get the right results on the first try. The natural language/ML approach may be much more intuitive and forgiving in that regard. It's just not an approach that's compatible with the low average quality of the content being indexed, in that it takes away the authority of the user to improve their search results.

I think there's somewhat of a tradeoff in search performance between quality of results on the first try and ability to improve the results on subsequent tries, and google is now optimizing for the former at great cost to the latter. And honestly they're failing at both.