I noticed an engaging blog post from Stoney DeGeyter over at SearchEngineLand this evening (or rather, morning/afternoon for our American & European readers) - The Ranking Roller Coaster Cause & Effect. It's definitely worth a read, but I also wanted to point out one specific area that we see causing the "rollercoaster" effect all the time - temporal data.
Temporal data for a search engine can include:
- When content was first spidered
- When a new link was first discovered
- Time frames for influxes of links
- Time frames for large amounts of content on a specific subject
The engines can use this data in all sorts of ways (everything from knowing what to put in the "news" results to determining potential spammers), but it really affects the rollercoaster ups and downs of rankings, too. Stoney mentions three things that can cause the coaster:
- Changes you make on your site
- Changes to search engine algorithms
- Changes made by your competitors
I'd add temporal fluctuations as a critical fourth. In a way, this falls under "changes to search engine algorithms," but the algos aren't really changing, they're just absorbing new data in the ways they always have. What we usually see is that Google and MSN, and Yahoo! to a slightly lesser extent, give priority to new documents on trusted sites and to even small clusters of inbound link influxes. Thus, the following scenarios happen quite a bit:
- You're ranking great, when all of a sudden, a Flickr page or a Technorati tag page or a page at Wikipedia overtakes you. The page is new, has little to no external inbounds, and you're flumoxed by how it can rank well. Don't worry, amigo - that's almost certainly the fresh boost, and it tends to die out after 5-10 days at most.
- You're ranking in 10th or 20th place behind some heavy hitters, but your domain is pretty tough and all of a sudden, 5-10 new links point your way. Voila! You're at the top of the results, ranking in front of pages you were sure you'd never overtake this quickly. Once again, it's fresh boost, giving a little bit of "extra credit" to your newfound popularity. I liken this to the search engines almost making the assumption that "whoa! this page got a lot of link love quickly, it must be super relevant/popular for this query, let's give it some juice." The problem is when the engines don't find lots more new links, you start falling down in the results fairly rapidly. Should we call that the "stale drop"?
- Rollercoaster mania hits - you're trading places atop the SERPs with 2-3 pages almost every day. I almost want to call this the "tie" flux - new links, and possibly refreshed content on your page and your competitors is making it a really tough call for 1st place, so the tiniest of changes can bump you ahead or leave you behind.
All in all, I like Stoney's post, particularly for his last few lines of advice:
Almost every site owner will, at one time or another, find themselves face to face with significant ranking drops. Panicking should be the last thing that you do. Sometimes the best course of action is nothing, however you can never go wrong with a bit of research.
Many people, when seeing sudden drops in rankings, make drastic changes in their website in order to compensate. For the most part, this is a bad move. The first thing you need to do is to research the issue, identify what (as much as can be determined) caused the problem and then carefully plan out a course of action, if any, which needs to be taken.
However, make sure you're thinking carefully about temporal data the engines use and how it might be impacting your rankings/results.
For those interested in some background on temporal link analysis, check out the following paper by IBM researchers: Trend Detection Through Temporal Link Analysis
The paper discusses (amongst other things) the timestamping of links and also analyzes a series of experiments where a list of sites considered 'authorities' (under the HITS algo) were ranked twice, once while considering temporal data and then again without temporal data. Interesting results ;)
Understanding the timestamped link profile (TLP) of a page can be very beneficial for a search marketer. For example, by observing TLP, you will understand:
(edit: I'd be remiss not to include Rand's analysis of Google's Information Retrieval Based on Historical DataPatent)
Thanks for the link to the paper. Very interesting info.
Thanks shor for both the link and the info.
No matter how many people tell a website owner not to panic if they fall foul of a Google algorith update and lose rank, they panic! I've even seen one very successful guy who stopped most of his optmization activities after Google wiped his site completely. So sad.
I completely agree. I personally believe in the rational approach and not over-reacting, but convincing a client to sit on their hands after their rankings (or traffic, online sales, etc.) drop is a lost cause. I've found the best I can do is try to get them data and explanations in the short-term and help them see how that leads to eventual action.
Definitely. The forums are filled with people panicking over even the slightest change. I'm with Pete on taking the rational approach and give things a little time to sort themselves out. Sure sometimes there's a problem, but not always.
Overreacting to something small could lead to a self fullfilling prophecy where you end up with the big problem you were hoping to avoid.
Reason #237 I love seomoz: things I have been suspecting but can never convince my boss of for months are confirmed on a really prominent website.
You've made my job easier. Good work.
If you change Changes to search engine algorithms to Search engine changes it will cover both code and data, including temporal data.
A piece of temporal data that I like to keep an eye on is last cache date and, related but not so temporal, chache frequency. Some of the queries I compete for have well ranked sites that swap rankings based of which one was most recently cached. When I see this occur I figure that the documents involved are all pretty equal. If mine is one of the sites involved I expect that for a little effort I might reap a large reward. Otherwise I will often examine the websites involved to see if I can discern commonalities and clues that might help rank for that query.
Very interesting. I like your angle and thoughts. You do a great job of making these technical processes easy to understand.
It's got to be hard for the search engines to determine valid temporal data such as the quality of the links/fresh content, right? That must have something to do with the drop after so many days once the jump occurs. It is your opinion they assume at first, no matter the quality of links or content that it must be relevant, then once they take a deeper look you fall back to your "actual" placement for awhile?
Nice follow up to Stoney's article. I remember when I first built my site I would notice a page ranking very well quickly. I knew they didn't deserve the rank at all, but kept wondering why they did rank until the soon disappeared.
I've observed the phenomenon with other sites since and knew it happened, but never completely understood why. I have a better hendle on it now.
Patent sleuth Bill Slawski has a tremendous update on this topic. He's found what looks like 2 months worth of reading material - Google has updated the original Information Retrieval Based on Historical Data patent with 5 newer patent applications.
Read more on this grab pack over at SEO by the Sea.
Thanks, Shor
The important part of those new patent applications are the claims sections.
Each of them focuses more narrowly on one area, and expands on those areas, using some new language to differentiate them from some previous patents and papers that cover some similar ground.
The discussion areas and examples within the patent applications themselves contain mostly the same text from one patent application to the next, so it might only be a few hours reading instead of a couple of months. :)
Don't get caught "flummoxing" in China, I heard it carries a heavy penalty! Good post.....
I like the term 'stale-drop' - it's a good descriptor - thanks for the insights Rand!
Yes, this is very true. I've seen new sites in my category zoom past me for certain keywords, only to fall back in the rankings a few weeks later.
I've started to become much calmer when the rankings ebb and flow...it's so much easier.
Thanks for the post reference and added information. I also celebrate the use of "flummoxed."
I think it would be a good idea if you sent over this week's "whiteboard friday" from the Eastern hemisphere incorporated with those cool site-to-sound delays that can be found in the 80's saturday afternoon kung-fu theater days- ahh to be a kid again.
Sorry contentmuse, but you'll have to put up with the peon Mozzers hosting Whiteboard Fridays for the next couple weeks. Rand is sans camcorder in China.
Oh stop the modesty you sagacious mozzstress. I can't wait to see what you mozzsters will craft up. :)
The absence of Rand won't spell catastrophe like Jackee leaving 227 or when Lenny "mysteriously" disappears from Laverne and Shirley- never the same...
Are you going to use the Transformers gun or would that be sacrilegious?