I missed most of the content on this ... page ? Exhibit ? Installation ? whatever it's called, because it told me to scroll, I did, and I scrolled through a bunch of what looks like empty space and arrived at the end ("and that's how search works"). The user is apparently supposed to stop and watch some animation at certain places, but it's not clear where to stop scrolling.

Perfect example, near the top there's some text about "It's made up of over[........] 30 TRILLION[.........] INDIVIDUAL PAGES[........] and it's constantly growing." But there's nothing to indicate that I should stop somewhere and wait for some more text to show up.

Maybe they should limit how far down you can scroll by setting the height of some element, and only increase it when the animation is finished.

Edit: the key problem here isn't the "scrolling makes things happen" gimmick that's popular lately. the problem is that it starts certain animations or fade-ins some time after I've already skipped past an apparently blank space.

The most interesting thing there is the live view of the most recently deleted webspam. I wonder what blackhat SEO firms can learn from that to better avoid the filters.

It's nice overall, but the timing for making items appear is a little slow. I was past most headers by the time they appeared, and I don't think I scroll too incredibly fast.

thx matt and the google search team for doing this. it's nothing new for technically inclined people, but every little bit helps. helps for what? teaching people to worry about the right aspects of search and the impact on their business, instead of worrying about bullshitphrases that were planted in their head by a SEO agency key account or a blogpost from 2008. so well yes, thx for doing this. i will send it to my clients (and tell them to click on the bubbles, even though they don't look clickable)

now an anecdote (because i feel like telling one): this week started for me with an interview that finally got published http://werbeplanung.at/news/marketing/2013/02/interview-mit-... (it's german) in that interview i claimed that

* 80% of everything written about SEO and Google is bullshit

* that all the rumors, tipps and trends are actually hurting business

* that we should treat SEO as a numbers based craft of constant optimizations

* instead of the esoteric bullshit art it is currently

* and, if search traffic is important for the success of a business, they must rid themselves of external (agency) dependencies and develop internal structures

nothing to far fetched i think. everybody knows the SEO vertical is full of bullshit, i just took some time to estimate a number (based on a random sample of collected blogposts (that at least one person tweeted about))

yeah, i got a lot of angry emails, skype messages, linkedin messages, xing messages after the interview was published.

most of them mentioned at least one of these words

  * pagerank
  * whitehat
  * blackhat
  * grayhat
  * linkjuice
  * panda
  * pinguin ...

so yeah, thx google for educating people about search. keep up the good work.

  "80 Prozent von allem, was über SEO geschrieben wird, ist Bullshit"

Has anyone deciphered the fat-mustache diagram in the "Query Understanding" circle? It's in the Algorithms section.

At first I thought it was supposed to represent a Gaussian-like probability distribution. But when I clicked on it, the resulting animation showed a series of such distributions getting flattened by some kind of distribution-flattening hydraulic press. The accompanying caption: "Gets to the deeper meaning of the words you type."

If I was confused before, now I was completely lost.

How is deeper meaning represented by distribution flattening? I'd think it would be just the opposite, raising probability mass around the likely meanings, not spreading it out into a uniform distribution over all meanings.

Baffling.

If anyone has figured it out, please do share.

(Maybe I'm taking the diagrams too seriously.)

EDITED TO ADD: New option: If you don't have any clue what it means either, come up with an entertaining yet plausible story that fits the hydraulic-press-vs-mustaches animation and share that story instead.

EDITED TO ADD: Example: At Google’s new eco-friendly data centers, NLP computations are performed by genetically enhanced inchworms. Difficult queries, however, can cause the inchworms to get cricks in their backs. In such cases, Google’s innovative back-massager descends and restores the inchworms to their preferred position (prone), from which they can return to their computations with renewed vigor.

I don't know what to take from this.

That search is very complex (I knew that, but not with this technical detail).

Or...that Google is trying very hard to maintain user interest with gimmicky shows of why it's cool and cutting edge and necessary.

Not that Google isn't those things...this just seems like an unnecessary expenditure of time. We know it's complex Google. Improve some other features and stop shutting others down instead of making these web 2.0 animations.

Interesting that they show the approximate number of searches / second at the bottom. Is that an otherwise publicly available number?

I was halfway through before I realized that some of the content was clickable.

Very nice page, though.

Their characterization of their spam procedures is grossly misleading. They do not send emails to most people that have been penalized, nor do they give clear instructions on how people can fix their sites.

Thousands of small sites were killed by Panda for no good reason, and have little hope of getting their traffic/incomes back. Google's spam policy is skewed heavily in favor of large sites and their own properties.

I keep checking every so often, but searching for "this phrase" or +absolute +requirement is still broken. Even "Verbatim", isn't. If they can't even get simple search right, who would trust them with anything more?

Scrolling is really becoming the new thing in UX design. It's an interesting contrast to the 'movie-like' flash animations of a few years ago that required no interaction on behalf of the user.

They left out the part where they index your emails and choose items you agree with over items you don't :)

38,800 requests/second according to their estimation.

Is this just PR for Google? Would rather see a more technical approach - although great for forwarding to clients when asked :)

Whoa... really, 100 MILLION gigabytes to store "The Index"? Wow. That's big.

There are some good facts and numbers hidden in rather toy explanation:

1. Spam detection is automatic

2. There 6 types of spam

-Unnatural outbound links (link selling)

-Content copy/manufactering

-Keyword stuffing

-Forums/user generated spam

-Parked domains

-Sites hosted on spammy DNS

-Different content humans and bots

-Hacked sites

3. Google is removing as many as 50K spam sites per month, they get 8K reconsideration requests

4. Google's machine learned relevance model may be using about 200 features

> By the way, in the 47 seconds you've been on this page, approximately 1,813,260 searches were performed.

Aren't these just some random numbers that they pull out of the air?

A beautifully designed page more than anything else

Nice scroll-UI! Took some time to see the clickable items. Interesting bits about spam pages.

An awful way to learn anything.

The better people understand their tools, the more effectively they can use them.

"We write programs & formulas to deliver the best results possible."

No kidding.

Some of the live listed 'spam' pages appear to be genuine to me.

Answer: It uses a bunch of skip lists.

Source: I do hacking on top of lucene.

Here's the unminified JS on the site responsible for the numbers updates.

   var kd = function () {
    function a() {
        e = e || Q("number_of_seconds");
        d = d || Q("searches_count_num");
        f = f || Q("searches_count_unit");
        var a = ~~ (((new Date).getTime() - h) / 1E3 % 86400),
            k = a * b + "";
        f.innerHTML = " " + c[Math.ceil(k.length / 3)] || "";
        e.innerHTML = a;
        d.innerHTML = k.replace(/(\d)(?=(\d\d\d)+(?!\d))/g, "$1,")
    }
    var b = ~~ (1E11 / 2592E3),
        c = " hundred thousand million billion trillion quadrillion quintillion sextillion septillion octillion nonillion decillion undecillion duodecillion tredecillion quattuordecillion quindecillion sexdecillion septendecillion octodecillion novemdecillion vigintillion".split(" "),
        e, d, f, h = (new Date).getTime();
    return {
        hc: a,
        rb: function () {
            a();
            setInterval(a, 100)
        }
    }
   }();

It's just running on an interval and doing in-page calculations, so it's entirely estimated. The value of "b" in this function evaluates to a little over 38,000 (https://www.google.com/search?q=1E11+%2F+2592E3) which they're using as the basis for the calculation.

Not sure why you're being downvoted, as it seems like a legitimate question, but...

I don't think so. It seems logical that Google's been keeping statistics about this sort of thing, so it doesn't surprise me that they keep track of such things as 'average queries per second'.

That would be about 38K searches per second. Does this include Google instant searches?

Google search results show a time value for each search. E.g.: About 2,210,000,000 results (0.12 seconds). Is this time machine time per search? This number is often around 30 ms, give or take a factor of two. If so, each machine can handle about 30 searches per second. If so, 38K searches per second need about 1000 machines. Sounds a bit too low... so my interpretation must be wrong at least somewhere.

It probably just increments by some fixed amount each second, but it seems like a statistic they would have an estimate of.

    function a() {
        e = e || Q("number_of_seconds");
        d = d || Q("searches_count_num");
        f = f || Q("searches_count_unit");
        var a = ~~ (((new Date).getTime() - h) / 1E3 % 86400),
            k = a * b + "";
        f.innerHTML = " " + c[Math.ceil(k.length / 3)] || "";
        e.innerHTML = a;
        d.innerHTML = k.replace(/(\d)(?=(\d\d\d)+(?!\d))/g, "$1,")
    }
    var b = ~~ (1E11 / 2592E3),