For instance, if I search for "a-r" I receive results for "ar".

I hate this. It makes it impossible to filter irrelevant results.

Or try this query: "a-c" -"ac"

This will return 0 results.

See "re-elect" -"reelect"

Then there's the "what you need, when you need it" category, and then there's the "put your google search in the title even though the page has no relevant results" category (mostly software download sites - ie. try searching for "application to use nokia e72 with itunes" you get this site filebuzz in the top two spots that has a whole lot of ads and a bunch of crappy non-related downloads).

So if you add a "uniqueness" index - ie. find ways to "semantically tag" not just the textual content, but the layout and font choices etc. of particular sites, that will catch the blatant affiliate spam bullshit (diy solar homes, what you need when you need it etc.) and then just figure out a way to prevent those "file buzz" type sites from sticking my search term in the title tags (I actually have no idea how this is done) you'll eliminate like 95% of the spam.

Likewise http://www.treehugger.com/ looks like it has genuine content - although it's obviously a little bit thin on the ground.

These guys are clearly a legitimate business selling a product (well, legitimate website anyway):

http://www.supremeheating.com.au/pool-heating-top/solar-pool...

Now compare those three sites, to these four:

http://www.diy-solar-power-for-homes.com/

http://www.energy4living.hottipsonly.com/solar-power-for-hom...

http://www.diysolarpower4home.com/

http://www.solarwindpowerguide.com/diy-solar-heating/

this one, too, links back to earth4energy - strikingly similar to "earth4living" above:

http://greenerhomediy.com/create-solar-electricity-build-diy...

then we have this incredibly reputable and highly respected forum whirlpool:

http://forums.whirlpool.net.au/archive/1539413

As a human I find it relatively easy to pick out the massively spammy sites amongst those results that are on the first page for the search "diy solar homes" - and I think that any time you were to get several sites that are similar in some set of ways to each other that rank for the same search term then they should be "de-ranked".

So for example, thousands of people use the same wordpress themes, you can't just say "they're all spammers". But if the top 5 results for a particular search all share some measurable characteristics, you could safely say "hmm there's something spammy going on here".

If you click anywhere on the page, you'll get a popup asking you to buy some book of theirs.

They must just have compiled huge lists of relatively specific search terms and have pages against each? But I would think this would be easy to identify and downrank..

It is a puzzle :/

The obvious problem is that spammers would attempt to game the feedback mechanism. But a combination of things like captcha to defend against robots, limits on how many times you can flag/upvote sites in a month (feedback credits), and exposing the feedback mechanism to only real, active-for-a-long-time users above a karma threshold (Google can definitely figure this out looking at the search history, gmail account etc), might be strong enough to beat the spammers.

You could start this as a Labs feature, and see if it works well.

It happens when one tries to use allintext: and is identified as a robot (why??!?)

People should understand how to use a search engine instead of have machines (second-)guess what they're thinking.

People learn to drive; if people can't drive a car we don't give them a car that drives itself!

-- Oh, wait.

Perhaps there is a better way.

http://www.nytimes.com/2010/10/10/science/10google.html

Edit: actually, all I really wish for, is for allintext: to work all the time. Why would I be a robot if I'm logged in, on an account of a normally active Gmail box?

Also, why is it such a secret? Why not make it more visible? A checkbox on the home page...

I wouldn't mind answering a captcha from time to time, but instead Google just bans the use of allintext for several dozen minutes (or more). Really frustrating.

When I search for [2010 ira contribution limit], all the results are spammy. The real official answer (on irs.gov) doesn't even show up on the first page.

BUT, if I use Google Instant, it does show up as the first result. As soon as I hit Enter, it again disappears and only the spammy results remain.

The Instant guess suggests that it's because the IRS website ranks for the plural term with "limits" [2010 ira contribution limits], not the one with "limit".

The first result I get is is irs.gov. Then again, we know what a bunch of shysters they are, so you still might be onto something. ;-)

http://www.google.com/quality_form?q=foo

The problem is that it gets polluted by lots of people who have no clue that Google is not the Internet, or (for that matter) their neighborhood handyman that they found through Google and who did an awesome job repairing their windows. A lot of people don't make a distinction between "I found what I was looking for" and "What I was looking for worked out for me", which makes a lot of the feedback a little less than useful.

This thread is as good a place as any - I'm guessing that the URL will get passed around, the appropriate teams will read it and adjust their algorithms, and if DuckDuckGo gets to improve their algorithms too, great - it's one more good search engine that people can use.

Also, google is one of the worst companies, if not the worst company, at dealing with user feedback. You can't just expect people to give you feedback but never return the favor in any way. People have the feeling that giving feedback to google is like throwing things into a black hole - like talking to a machine - if you know no one will answer you and you will never know if anyone even read your feedback, not even a little thank you note or a clue that the feedback was useful, there isn't much incentive to give feedback, now is there?

As a result, I enjoy reporting issues to Google Maps, because I know they will be addressed. Maps are dealing with a lot more tangible and unchanging dataset than Search, though.

Create a better mechanism through which people can submit bad searches for human review.

I wonder if a change towards "human input" might improve things more.

For example, what if the Chrome browser had a big feedback button so that if users wanted to help improve the Google search results they could rate the usefulness of the link they just followed?

Back then (and with "Remove result" and SearchWiki) it had issues because we were trying to get people to recognize spam, and people weren't that good at recognizing spam techniques like hidden text. The more recent complaints we've had are more like "here's content I don't like." So maybe it's time that we tried something similar again.

People are pretty good at recognising "scams", but you need to tell them that's what they're looking for, not just how the content makes them feel. It seems like the report spam button was a large factor behind gmail's spam filtering success. Would love to see the same approach applied here. I know I'd be hitting my report spam button in chrome pretty often :)

I just know I see a lot of worthless junky sites on the first page, and I wouldn't think it would be hard to recognize them using ML techniques, which requires training data.

Recognizing that not all users are created equal is (IMHO) an incredibly powerful insight that Google and many other companies overlook. Qualified, technically literate users will be happy to volunteer, but you have to ask.

I think the cases that Google breaks is that some people figure out some trick/technique that lets them become a small power circle, which they then obviously abuse. When Google works it's because they're able to algorithmically spread the power around and at scale see what is quality and what isn't. Therefore I think the better strategy isn't to concentrate power even within an "elite" class, in fact it's the complete opposite, it should be how to make sure the power is spread within the masses quite evenly.