Forum Moderators: Robert Charlton & goodroi
Think about it... the Google Toolbar, Google Analytics and click monitoring on the SERPs give Google an incredible picture of where people are going, what pages they stay on, what sites they frequently return to and where they go when they leave.
We know that Google is pushing the toolbar onto consumers. They're paying Dell a billion dollars to install it onto 100 million consumer PC's. Imagine what the behavior patterns of 100 million Internet users could tell Google about a particular site's value.
What scares me is that this will push the blackhats from link spamming over to the busy spyware world. Imagine if I could pay some shady company to have the web browsers of 100,000 pc's randomly click on my #10 ranked link and stay on my site until Google decides that I should be #1. Who cares if these users buy anything on my site. I just want Google to THINK that they're using it. Will Google start bundling anti-spyware with the toolbar to stop this?
Am I on to something, or has this been going on for years?
[edited by: tedster at 8:38 pm (utc) on April 6, 2006]
We had a related discussion recently. Two of us (including me) have an average ratio of 2.5, which seems very few compared to what I personally perform on my visits at webmasterworld, for instance.
Low pageviews per visit and even short visit-time could also indicate a very good usability on the website in question.
I do believe that google does analyse and use visitor behaviour, but I also think the level of mathematical abstraction and analysis is far higher than suggested here. E.g. neuronal networks and statistical correlations we can't think of, simply because we have no insight into that data.
As has been pointed out several times elsewhere by MC and others, big daddy's new infrastucture is very much about improvement on search quality on non-english websites. Any idea whether/how this might be related to traffic patterns?
A site with poor navigation, and/or empty content would have a higher average pageview than a site which gives its visitors exactly what they're looking for within a click or two.
Conversely, lots of pageviews can mean a site has a lot of interesting pages, which encourages visitors to dig deeper.
So, lots of pageviews can mean a site is crappy, but not so crappy visitors leave immediately, or it can mean that the site is relevant and has a lot of interesting content.
Low page views can show that a site is well organized and gives vistors exactly what they're looking for in an easy-to-find format, or that the site has nothing much of interest.
Therefore, pageview statistics should be taken with a grain of salt.
Wouldn't the rate of return visits be a more accurate demonstration of a site's relevance?
If google were to put emphasis on pageviews as an indicator of relevance, would that not encourage webmasters to make their sites harder to navigate?
Someone earlier mentioned that in a Meet the Engineers session a Google engineer had mentioned that they wouldn't use this as a direct method of deciding rankings, but might use it for indirect influence of some unspecified sort.
Here's my take, based on usage analysis of a Fortune 10 company website: Google is not using Toolbar stats to generate a rank of a page (yet), but they are likely using it as a component of their quality assessments to pull bad pages/sites out of rankings.
Consider: if Toolbar and Urchin usage data was used in combination with methods which were proposed in the "Combating Web Spam with TrustRank" paper which was published out of Stanford, Google would have a strong tool for suppressing spammish/low-quality pages from their SERPs in an automated fashion. We know that they, like other SEs, use a number of staff members to manually assess the quality of search results and to identify spammers and other black hats. But, the combination of methods mentioned in their "Information Retrieval Based on Historical Data" patent and the Stanford paper would pave the way to improved automated assessment methods.
Simplistic explanation of how this works: the TrustRank study suggested that human assessors could rate a sample set of internet pages on whether they were good results for a keyword search or not. An algorithm could then be applied to those pages which were ranked as bad, and link structure patterns could be used to take all the pages associated with the "bad" sample set and suppress the low-quality pages at those sites and their related network sites.
Toolbar or webbug usage data could then be used to flag sites with high abandonment rates or links which users never chose to click, and then suppress rankings of all sites with pages associated to the bad ones. This would be fairly trustworthy data, and not as prone to manipulation by black-hats. After all, it would be using negative info rather than positive -- pages not visited, and pages which are not stayed-on longer by users could be automatically dropped in rankings. It's not as prone to manipulation by black-hats because they couldn't generate a negative -- they'd have to stay OFF sites and leave sites in droves, and compared with all the other browser users who are visiting a site, they wouldn't be as able to artificially cause competitor sites to drop in rankings.
So, if my theory is correct, Toolbar/Analytics data is not being used to make your site rank higher, but it could be used to drop you down in the results or smack you out of the indices entirely (if you've got pages that users hate).
Trustrank is my favorite google product, mainly because it seems to once again find a way to reward websites that are actually trying to create unique valuable content. It's very hard to emulate the trustrank link schemes, although I'm sure blackhats are hard at work on the problem, and have been for quite a while. But as long as the initial pool is kept relatively uncontaminated of seo sites, it will be hard to break into that level, although you won't see as much benefit I'd guess in the heavily commercial categories.
It makes total sense to treat this type of data as a negative page ranking factor, when people leave a page in a consistent manner, it's fairly obvious that it's not useful to the searchers.
When I first saw signs of trustrank it gave me a lot of hope that google could start working out a way to both maintain its income and the quality of its serps, that's a fine line, hard for them to walk.
chadlerh: yes, absolutey, we've had test sites spidered because of that issue. No question at all. But that part is just the google indexing system being notified of a url that someone with the googlebar visited. Even if Google isn't currently using the user data in all the ways it can, it's not throwing it away, that's for sure.
The fact that it was too big as a free service and they stopped handng out free accounts tells me they don't care about getting all that data or they are simply not ready to use it all. Otherwise they woul have found a way to make it work so they could continue to get the massive infux of data.
Once it becomes free again then I will muse over what their true intentions with that tool are.
Examples:
A user clicks on a SERP and leaves quickly leaves the site and moves on. Is the site "relevant" and valuable?
Maybe the user thought the site sucked and left. Or maybe they found their information quickly (price comparing, answering questions, solutions, definitions, etc.).
In virtually every scenario with click metrics, there are completely inverse scenarios that perfectly explain user behavior.
I saw an engineer from MSN speak on this and said their research shows that click metrics has far too many assumptions built in.
site a ranks for green widgets, number 3
This term is searched for 1 million times a day.
If a large percentage of searchers return to the search page and click on the next option, it's quite clear that site a is not what the searchers wanted. Individual behavior isn't that important, but it's clearly one way you can measure how close the serps are to user expectations, that's why google included this in their recent patent application.
It's not the only way, of course, but it is one tool you have. And if I read the msn search javascript correctly, they are tracking clicks, although that script is hard to read, google's is radically simpler. About 1 line versus 50 or so... typical Microsoft.
If a large percentage of searchers return to the search page and click on the next option, it's quite clear that site a is not what the searchers wanted. Individual behavior isn't that important, but it's clearly one way you can measure how close the serps are to user expectations, that's why google included this in their recent patent application.
Suppose a representative subset of them are doing SERP comparisons? It isn't clear (for all searches) that hitting the next link is a sign of dissatisfaction.
Average searchers just know that they typed in a search term, they got some pages, they click on them, if they are right, they stay, if they are not, they go back to the next search item until they find one they like. Google's goal is to keep people using their product, which means to give users what they are looking for. This isn't particularly earth shattering information if you ask me.
Try to step out of the seo box, seos have very little impact on search patterns, we're talking about far too many searches a day for seos to influence things, though of course that won't stop them from trying. Obviously they can and do influence pages landing in the serps in the first place, the trick here is to get rid of garbage and replace it with reasonably good quality results.
Most people type in a search, then go to the site. If they don't come back to the google search, the result can be considered successful, if they do, it can be considered as a failure. There's no need to make simple things complicated.
Keep in mind, if this is being used, it's simply one factor among many, nobody knows what weighting any one factor has, so stressing about one component really isn't worth bothering about.
To keep it ultra simple: it's a good thing if your page gives the searcher what they were looking for. It's not necessarily a bad thing if it doesn't, since it's hard to know what they are looking for. But once the search numbers get big enough this stuff all becomes pure statistics, stuff you can study and integrate into other components.
