Google Updates and SERP Changes - March 2013

[gyppo]

< continued from [webmasterworld.com...]

I'm seeing daily switching between 2-3 datasets across around 600 keywords. Seeing roughly 10-20 keywords swapping in & out of positions.

[edited by: tedster at 2:47 pm (utc) on Mar 1, 2013]

[TheOptimizationIdiot]

@TOI, yes, almost smiliar in mind :). I almost thought that you are a google worker cause the way you defend the serps.

LOL Nope. I just try to look at the bigger picture and realize how tough it is for them to do what they do. Helps keep me sane (or relatively sane) lol

I mean really, a trillion+ pages and a trillion+ searches a year? They're gonna miss a few, which relative to the numbers they work with means if they're right on 99% of the time (for 990,000,000,000 queries) they miss on 10,000,000,000 queries.

Yes, that means 10 billion queries are off if they have 99% accuracy and here we are complaining about a few results. Kinda makes me chuckle a bit sometimes when I think about the near insanity of people thinking every result set should be "right on every single time" and "Google's broken" if that's not how the results are right now today, because even with 99.9% accuracy, they're "missing it" for 1 billion queries.

[edited by: TheOptimizationIdiot at 9:01 pm (utc) on Mar 7, 2013]

[claaarky]

Tedster, it feels almost like a daily Panda refresh. Lots of small adjustments instead of one big monthly shuffle.

I'm seeing sites moving in both directions on an almost daily basis, but the ones going up keep going up and the ones going down keep going down.

[Martin Ice Web]

TOI, especially when I look at the search strings that poeple write into our own searchbox! Sometimes my head thuds to the table when i see it. But 99%! Without complaining but most results for thinks i search for are very terrible so i mostly use bing now, especially when it comes to coding examples/help. In g i miss the good old forums that have been vanished with penguin/panda.

[TheOptimizationIdiot]

Yeah, I switched to Bing a couple years ago, and it took me a while to get used to, but I like it better now even though the way I search when I've done comparisons the results are close to the same, I guess there's something about the #0044CC blue I like! lol

* We should probably get back on topic though or we're gonna make tedster yell and pull out his golden snippers.

[TheOptimizationIdiot]

Something I've noticed the last couple/few days is new pages are slower to get indexed and generally when I start seeing that it means they're running something "big" in the background and not increasing the size of the index while it's running.

I really don't know what it is though. It could possibly be a Panda/Penguin run that seems a bit larger than normal and they could be pushing portions of it as they complete, which would explain some people seeing flux and others seeing stability, but the results I've been watching are relatively stable. Some flux, but nothing huge I've seen.

I've also see two seemingly different result sets based on site searches every-other-day for a few days. Indexed page count is +/- about 10% from one to the other.

[edited by: TheOptimizationIdiot at 9:16 pm (utc) on Mar 7, 2013]

[getcooking]

@tedster

We've been seeing this same pattern with our Pandalized site since the Dec 21 Panda update. Each week traffic drops a little more. Nothing that we can trace to a specific date or change to the site. I've been wondering if they are integrating Panda updates into a more real time type process instead of the once-a-month-ish updates.

[TheOptimizationIdiot]

Maybe we'll get lucky and their purging the index of the stupid 18 results in a row from the same site garbage they've been doing.

I like it when I have 4 or 5 results from one of the sites I'm working on in a row, especially on page 1, but it's really too much from a "user perspective" for me anyway.

I wish they'd just show the best 1 or 2 results they can find from a site and let me decide if they're what I want or not again like they used to.

(That's actually one of the biggest benefits I've found lately with using Bing. I don't have to go to page 5 to find a result from a 6th site I might want to visit, because they don't put 20 results from one site and 15 from another on the first 4 pages.)

[HuskyPup]

Yes, that means 10 billion queries are off if they have 99% accuracy and here we are complaining about a few results.

However it's not a few results is it? Every sector I look at is a mess and don't even go down the images route, that's so screwed-up it's unreal.

For me Google used to be pretty accurate for most searches however these days it is inaccurate for most searches. This has all come about with constant change after change after change.

Why is it most users, and I include Joe public in that, can see their results getting worse and worse every day yet they seem to believe they're getting better. Are they actually seeing what we are seeing, I very much doubt it because if they were they'd do something about and darned pretty quick as well.

[claaarky]

I'd say the "daily Panda" effect started January 31st.

[Dymero]

Traffic is flat, which for me is an alright thing, because for a while it was skyrocketing and pushing down CR.

But SERPs are definitely all over the place. Even some rankings that I've been able reliably expect in the same place from week to week are jumping five to six positions or more.

[TheOptimizationIdiot]

However it's not a few results is it? Every sector I look at is a mess and don't even go down the images route, that's so screwed-up it's unreal.

Depends on how many sectors and queries you're looking at. I know someone said they track 600 queries here and it sounds like a bunch, but out of even 1 million unique queries it's 0.06% or barely even a blip and out of 1,000,000*1,000,000 (1 trillion) overall queries a year (some are obviously repeated in that number but) 600 queries out of that many searches is nothing even close to statistically relevant: it's 0.000000006%.

Even if we take the 600 queries and multiply it by 10 results per page to say they're tracking accuracy of pages returned in the results, then look we at the number of overall pages to choose from the portion being reviewed and tracked for accuracy is 0.00000006%.

[blackpixel]

@TheOptimizationIdiot

By using statistical determination of sample size, I believe that if a big chunk of us see the messed up results on keywords we track that means it is a big problem and google is broken.

For example I use USA google and see sites in turkish for popular english query not related to Turkey by any means. That is not normal.

[TheOptimizationIdiot]

Count the names of people in this thread who are saying things are messed up. How is that a big chunk of 250,000+ members? (It could even be 350,000+ I think I heard 350,000+ but I'm not 100% sure so I use the lower number I know I read a few years ago about membership here.)

And there have been Tons of reports here over the years about how broken Google is, but somehow, someway, within a couple weeks of an update when things settle down it turns out it's not really broken after all.

[edited by: TheOptimizationIdiot at 10:14 pm (utc) on Mar 7, 2013]

[blackpixel]

I mean statistical determination of sample size of queries.

[blackpixel]

It would be pretty easy to check by taking a broad one word keyword search volume and do a sample size long tail keywords check for accuracy and quality :)

[frankleeceo]

I think if we really do look at the long term picture. Google is in the works of "scaling" to the web. They are testing to "explode" or "focus" on the sites that are bad. So bad that they can get a consistent "bad" sample of these "bad" sites. The intention is not to bury the bad samples, rather letting them float to the surface, and let the crowd or userbase teach Google. But because it is so cheap and fast to create sites, it creates a constant flux of bad results.

Many sites may and do get reshuffled based on some "random" or hidden metrics. I think signals are no longer signals, they are more a mixed goo of mixed signals with loops and tons of if -then statements. Like, if you are an informational site your logo should be X size, should have comments enabled. If you do not your site is "bad". It can be something completely random simply based on "machine observations".

I foresee more shuffling and pain in the future. Many sites will die, and only few will flourish. The algorithm is certainly going in that direction. If somehow your site exhibits signals either accidentally or intentional to that of other "bad" samples, your sites will die. And no one can help you.

Feels like terminator is gonna take over the world, but in this case it is Google's machine learning capabilities. One day "Google" is going to scrape all the knowledge in the world, but we all just need one single website. I look forward to the day that we can order pizza via Google. (being sarcastic)

"1 large pizza, 3 toppings, cheapest price, under 15 minute". "I am feeling lucky"

Google: "Please give me your credit card."

Done ordering, pizza gets delivered. Google WILL decide which pizza you like the most based on your browsing behavior and what pizza you recommended via G+.

Result - all pizza website dead. Only one pizza left. Hell, Google should become a pizza restaurant too. They can cache one in their databases with unmanned vehicles for delivery. No Tipping necessary. More Win.

[TheOptimizationIdiot]

I mean statistical determination of sample size of queries.

There's 34 names of people posting in this thread.
(35 counting franklee, but I did the math on 34)

Many are complaining about their specific site's rankings, but only 6 have said anything about noticing only results filled with spam.

Being generous and assuming the 6 people saying they notice spam filled results track 1000 unique queries a day each that's .6% out of 1 million, .0006% out of 1 billion, .0000006% out of 1 trillion.

Being super generous and assuming everyone who's posted sees only spam (that's not the case), because I for one don't, but assuming they do and all track the generous number of 1000 unique queries per day, it accounts for 3.4% of 1 million, 0.0034% of 1 billion, 0.0000034% of 1 trillion.

To have a trillion+ queries in a year they have 1,902,587 queries a minute. How is 1.8% (0.001241% of daily queries) of the queries they have in 1 minute close to large enough to be representative of even the daily number of queries (2,739,726,030) when the algo is updated 500+ times a year or 1.4 times per day? Never mind thinking it's representative of the year.

[Wilburforce]

@TOI

You are looking at the question backwards. The number of possible sequences of a 52-card pack is much bigger than the numbers you are talking about. What is the probability that 6 people with six separate packs all draw four aces?

[TheOptimizationIdiot]

The same as the probability 6 people with 6 separate decks will draw the same 4 cards of the suit specified.

What is the probability 6 people out of 250,000 will see 'only spam' in the SERPs they check? The same as the probability 6 people won't see 'only spam' in the SERPs, so if 6 do and 6 don't whom should we trust?

If 6 do and 10 don't whose information is better and more accurate?

6 people are stating they see 'spam filled results' in the SERPs. 28 aren't saying that. Whose got the better/more accurate sample size?

BTW: I think your correlation to a deck of cards and drawing aces would be better suited to "what's the probability all 6 people would search for the same terms".

[Wilburforce]

Whose got the better/more accurate sample size?

That isn't the question.

Even very small samples are statistically significant if the probability of the result occuring by chance is infinitesimal.

If I look at 50 different phrases and see the equivalent of the same card sequence every time, it doesn't matter that they are only a small fraction of all possible phrases: it is statistically very unlikely that it has happened by chance.

If, in your sample airline, six aeroplanes crash and ten don't, it is more likely than not that the next flight you get on won't crash, but it is still certain that the airline isn't safe.

[TheOptimizationIdiot]

LOL it's also statistically unlikely you'd be arguing with someone who's not only flopped a spade royal playing texas hold 'em but turned the 9 to make it a 6 card royal (I had the KT ... AQJ flop 9 on the turn) or has seen the flop come JJJ turn J river Q for a split pot when one of the two players in the hand to the river was holding a Q and had the pot won until the river, because the other player was holding a T, or a who's seen a flop of KKK twice in a row with 2 black and one red K on each from two separate decks, but you are.

It's also more statistically unlikely if the 6 people's results were actually representative of the whole somehow 28 other people could manage to search and not see the same type of results, than it is for the 28 people's results to be more representative and 6 seeing something different, so when you look at it logically it really comes down to the sample size of which group is more likely to be representative of the whole than anything else.

[TheOptimizationIdiot]

And the answer to your question about drawing 4 aces is:

52*51*50*49*6 if your sample size is only 6 people, but it's
52*51*50*49*6/4.25 when you increase the sample size of those drawing to 34, because, as the number of people drawing (searching) increases, the likelihood of a statistically low probability event occurring increases.

[edited by: TheOptimizationIdiot at 12:21 am (utc) on Mar 8, 2013]

[Wilburforce]

It's also more statistically unlikely

I have no idea why you are arguing with me about statistics.

If 6 of a sample of 34 see a problem in Google SERPS, it is not "more likely than not" that they are OK generally. If a surgeon kills 6 of every 34 patients he operates on, he is not an OK surgeon.

Also, like HuskyPup, every sector I look at is a mess. It is more likely that the mess is widespread than that I am able to keep finding only mess. I don't need to ask a few billion people before I can be confident of it.

[TheOptimizationIdiot]

I have no idea why you are arguing with me about statistics.

Because you're wrong about which is more likely to be representative of the whole, but if it makes you feel better to think you're correct and Google's broken or something because your site doesn't rank as well or you're seeing a mess in whatever searches you're conducting, then feel free.

If a surgeon kills 6 of every 34 patients he operates on, he is not an OK surgeon.

But you're 4.25 times more likely than not to survive the surgery. The same as "the random searcher" is, based on the numbers we've been using 4.25 more likely than not to find the results they were looking for.

[edited by: TheOptimizationIdiot at 12:29 am (utc) on Mar 8, 2013]

[Wilburforce]

And the answer to your question about drawing 4 aces is:

52*51*50*49*6 if your sample size is only 6 people

No, the chance of six people draxing 4 aces is one in

(52*51*50*49)^6.

[TheOptimizationIdiot]

(52*51*50*49)^6

Neato. Does the likelihood of 6 people drawing 4 aces increase or decrease when the number of people drawing increases?

For some reason I doubt you'll answer that probably because you didn't correct that 28/6 is 4.7 not 4.25 when you corrected my other math. But the exact math really doesn't matter, the simple fact an event of low probability happening increasing relative to the number of people participating is enough, because just as someone is more likely than not to survive a surgery in your example, so is someone to find the results they were looking for in Google, even if some people aren't satisfied with those results or the results are not 100% "great quality" for every query. I just don't get why you know the math better than I do and refuse to admit it?

[Wilburforce]

Does the likelihood of 6 people drawing 4 aces increase or decrease when the number of people drawing increases?

If 12 people were drawing, it would be twice as likely, i.e. 2 in (52*51*50*49)^6, or 1 in ((52*51*50*49)^6)/2.

[TheOptimizationIdiot]

Okay, you proved me wrong about answering lol.

How about we leave it at: It's probable some people will find what they're looking for and probable others will find spam and since there's 500+ algo changes a year the number of who finds what and what % finds each is likely to change 500+ times a year too.

I do think it's helpful for people to actually think about "the maths" of things regarding rankings and changes, which is part of the reason I kept answering, because when I think about the difference in "overall score" for pages from #1 to #50 for a query where 1,000,000 results are possible, the % difference in "score awarded" is tiny and if the non-spam results are lurking on page 5 it's entirely possible they'll "disappear" with the "slight twist of some quality dial" and if not then "overcoming the spam" on or off site shouldn't require a "huge aggressive change", because of the small difference in "score awarded" to pages it takes to influence rankings when a page is anywhere near the top of the results (like in the top 100 even).

[garyr_h]

I think it depends on what you define as "spam". I see many, many, many results pages filled with Facebook pages and Pinterest boards. In many cases, those take up 5 or more spots in the top 10. If I'm doing research for a paper, what good does that do me?

It's not technically spam, it just doesn't belong there.

For that matter, I don't think Pinterest boards should be included AT ALL. A huge portion of those images are stolen. However, even if they were all unique, why is the page with a picture in it showing up in the search results instead of it simply being in the image search? It doesn't make sense.

[JesterMagic]

There is a lot of spam in the search results (been like this since April). Lately I am seeing multiple sites from free web providers popping up on the first page or two with just crap on them.

In my niche for a popular brand you can type in their domain name and the second result is a scrape of the brands login page which contains malware. It has been their almost 2 months now.