I feel the same. Looking for specialized topics with Google is now very difficult. Now is impossible to look for phones, uncommon words or looking for anything that is not the mainstream result.
I'm not sure if the culprit is BERT or using neural ranking. But in the last years I feel that is more common that I leave Google search without useful information. The worse part is that all the competing search engines are using the same algorithms that are only useful for mainstream results.
I noticed this in my country when searching for somewhat less common parts (electronics, car parts, tools, etc). The first few results are for online retailers in my country, and then after that it's full of domains with paths such as /sale_12345678. The domain sounds somewhat promising, and the description sounds good - other than it often being a quantity of 10 - but when you click the link it just redirects to AliExpress.
I find that using another search engine in that kind of situation is extremely useful! If I'm searching for more mainstream stuff google usually is great but when I'm going for more specialized topics duck duck go will usually bring up some different links!
Search engines seem to be stuck between serving two roles: 1. An easily accessible directory of mainstream information, and 2. A specialized tool to find the diamond in the rough. It seems like it has to be a tradeoff, it can't serve both roles equally well.
"The alteration or outright disappearance of inconvenient or embarrassing documents, photographs, transcripts, or other records, such as from a web site or other archive. Its origin comes from George Orwell's "1984", in which the memory hole was a small incinerator chute used for censoring, (through destroying), things Big Brother deemed necessary to censor."
> I'm not sure if the culprit is BERT or using neural ranking
Tools are not to blame here, it's like blaming the compiler for the behaviour of an application. Starting with the training data and ending with how the model is used in deployment it's the blame of people who made it, not of the neural architecture. The architecture itself can learn anything you throw at it, good or bad.
I'm not sure if the culprit is BERT or using neural ranking. But in the last years I feel that is more common that I leave Google search without useful information. The worse part is that all the competing search engines are using the same algorithms that are only useful for mainstream results.