> google seems to want full sentences instead of just keywords now. "How do I do X?" seems to get me better(?) results then "X + some relevant keyword"
Correct. Google will give you better results with full-sentence-like input. After several years of refining results, Google concluded that there's more benefit in teaching the machines to understand how humans ask questions than to teach humanity how to keyword like a computer expects (especially when you factor in that they get as many queries via voice these days as via text, and voice recognition in general always benefits from more information to disambiguate on). There's an entire semantic-analysis layer in front of the keywording layer these days to determine some semantics of the query to try and guess what category of thing you're looking for.
I generally have no problem with a few keywords for software engineering searches. I usually go general-to-specific (for example, `react unit test useState`).
You can drop the video results by adding `-youtube` to the query.
> "Sal dulu a" would both recommend "Sal dulu antasma" and list it but now unless i search for that particularly, it doesn't show up.
I'm not sure, but it's possible Google dropped 'a' as a signifier because of the semantic query support (as a single particle, it doesn't add signal to a sentence-like query). `sal dulu songs` gives me a list where Antasma shows up as item 3.
In general, my advice for Googling these days would be "don't try to keyword it out." Think more like how you'd ask another human for a random fact they might barely remember.
There's also still some symbols that are specifically understood by Google for tuning queries, listed here (https://support.google.com/websearch/answer/2466433?hl=en). Worth noting: the `+` modifier got killed when Google+ came and went. To force a word or phrase to be part of the results instead of "fuzzy-matched," put it in "quotes". Quotes these days do double-duty as both "I want this literally matched" and "results must include this token."
`shirt without stripes` definitely fails (and I see it is a natural-language meme because that's what is returned most prominently against that phrase. Hilarious.)
`shirt -stripes` as an image search is also not foolproof. It does a better job, but some leak through (I would assume for two reasons: if the text just doesn't mention it, the negative key filter has nothing to key on, and we have different words for "stripes" that Google doesn't consider synonyms. Is a plaid shirt striped? What about a checked shirt?).
But hilariously, all of the ad results are stripe-heavy. Looks like there's a bug: negative keywords in image search aren't passed as negative keywords to the ad engine, but instead as positive keywords. Wasted money for those advertisers; that's definitely not what the user wants.
For this sort of search, I'd actually use Shopping... Except that while there is a "Striped" category, I know of no way to do a negative category search on the Shopping UI, so I get the opposite of what I want. I can get all the striped shirts ever, but non-striped would be harder.
I suppose I should amend my previous comment to say "If you don't try natural language queries on the fringe of known unsolved problems, it works better than keywords." ;) And since Google is forever targeting the common use-case, perfect may be the enemy of good here for getting most people what they're looking for most of the time.
Correct. Google will give you better results with full-sentence-like input. After several years of refining results, Google concluded that there's more benefit in teaching the machines to understand how humans ask questions than to teach humanity how to keyword like a computer expects (especially when you factor in that they get as many queries via voice these days as via text, and voice recognition in general always benefits from more information to disambiguate on). There's an entire semantic-analysis layer in front of the keywording layer these days to determine some semantics of the query to try and guess what category of thing you're looking for.
I generally have no problem with a few keywords for software engineering searches. I usually go general-to-specific (for example, `react unit test useState`).
You can drop the video results by adding `-youtube` to the query.
> "Sal dulu a" would both recommend "Sal dulu antasma" and list it but now unless i search for that particularly, it doesn't show up.
I'm not sure, but it's possible Google dropped 'a' as a signifier because of the semantic query support (as a single particle, it doesn't add signal to a sentence-like query). `sal dulu songs` gives me a list where Antasma shows up as item 3.
In general, my advice for Googling these days would be "don't try to keyword it out." Think more like how you'd ask another human for a random fact they might barely remember.
There's also still some symbols that are specifically understood by Google for tuning queries, listed here (https://support.google.com/websearch/answer/2466433?hl=en). Worth noting: the `+` modifier got killed when Google+ came and went. To force a word or phrase to be part of the results instead of "fuzzy-matched," put it in "quotes". Quotes these days do double-duty as both "I want this literally matched" and "results must include this token."