There is an interesting article over at Formatur.de (in German) that talks about Apple’s release of Safari 4 beta and its potential impact on web analytics. The contention is that due to Safari’s new Coverflow-like feature to display the most visited sites, Safari makes regular “hidden” visits in the background to update its screenshots in the Top Sites. This means a website that is listed in the Top Sites section gets visited even if the user does not actively visit the site. The implication for web analytics, Formatur goes on to explain, is that Safari 4 users will have more page views, higher bounce rates and more ad impressions (therefore lower conversion rates) due to these hidden visits.
According to Formatur, there is no known workaround to this issue for web analytics providers as there is no method, using JavaScript or otherwise, to differentiate between a virtual visit and a real visit.
I tried to replicate this phenomenon on my site using Safari 4. On Wednesday, I visited my site several times to ensure it appeared in the Top Sites in Safari. On Thursday I used Safari for my web browsing but made sure never to visit my site. I then pulled up some reports in Sitestat and Google Analytics to check if there were visits from Safari 4 on Thursday to my site. I then segmented these reports to isolate my visit – in Sitestat I used my IP Address and in Google Analytics I segmented by City and Provider. While the segment in Google Analytics does not necessarily exclude other potential Safari 4 users in my city, the data from Sitestat validates.
Top Sites in Safari 4 Beta
Safari's hidden visit to Datadriving.com with in Sitestat
Safari's hidden visit to Datadriving.com in Google
These reports confirm Formatur’s suspicions that Safari 4 is making hidden visits to update its screenshots and these are being tracked by analytics tools like regular visits. The question for me is how often does Safari make these hidden visits and how will it affect web analytics if Safari 4 gains market share or, as Formatur predicts, the Top Site feature is copied by other browsers?
Hopefully the major analytics providers (Omniture, WebTrends, Coremetrics) are already trying to get in touch with Apple to see if the virtual visits can be measured some how. I’d be surprised if these companies aren’t trying to talk to them already. After all, when it came out that Google’s AJAX SERPs were not good for analytics, these companies were on the phone with them.
Hi,
The Firefox FastDial plugin does the same thing, as does Opera’s Speed Dial.
Cheers,
Jan
Possibly to get true bounce rate these users have to be filtered out, at least in the early stages.
Possibly, bounce rate as a statistic is just going to have to go away. And stats will have to be adjusted to only report numbers for visits greater than 1 hit/event/pageview/whateveryoucallit.
We’re already doing this kind of reporting (remove the bouncers) — it’s actually a fruitful way to look at the numbers.
Not only are stats misleading due to coverflow/top sites updates, but possibly also due to a new prefetching feature Apple calls “speculative loading”. I’m unclear as to whether the beta precaches linked pages or simply prioritizes non-image data before rendering, but if it’s the former, then we should expect distortion.
[...] The Data driven Blog und <formatur in ihren Blogs melden, soll der Safari 4 Browser “virtual visits” oder [...]
@Jason Egan: The problem looks to be more general: web analytics is working only because of the blindness of the majority of the users. They have no clue about collecting of the data on a site or over different sites at all. Additional all used technics are designed to prevent the user to notice, what is really happend.
On the other side the clients of the web analytics companies do accept these technics of obfuscating. They are interested in the data. This means the web analytics business in the current situation is based on the agreement between two of three parties that it is acceptable to hide the real activities against the third party. In Germany we call this “Vertrag zu Lasten Dritter”.
In conclusion: This business needs a clear and easy to use technic for opt-out or much better opt-in. After that, the data will become more scruffy, but they will be still comparable. This is the exact situation which we have at print and tv.
If you say Omniture have to find out a way to remove the ghost measurements from Safari you say also: o man, it is so cool that the users have no clue about web tracking, Apple, please let us stay at this situation.
Try adding a bit of JavaScript to check the document.URL of the page in question. If your domain isn’t visible in the URL box then make sure the tracking doesn’t occur. That should reduce the “passive” requests from such a browser.
Then again, you might want to see such requests. In that case use a custom label in Sitestat to signal passive=yes during data collection so that you can segment on it. That also lets you capture if your pages are in an iframe of another site as that too can mess up numbers.
Makes sense?
/F
@Fulton: sounds good! I’ll try it.
@Fulton: Thanks for the comment. I will be sure to try that out. At this point it makes sense to see these requests to gauge the impact of the virtual visits.
Better yet, use server side checks of the URL if you want to reduce JavaScript dependency.
Based on the results the appropriate coding can be applied.
@Fulton: the JavaScript solution isn’t working. The document.URL in Top Sites are always the real URL.
To block the incoming requests based on the ‘Public Beta’ part of the user agent string will only help a few weeks. First: it will be gone after the final release (probably with Snow Leopard end of June). Second: it is a typical mistake of the Apple developer, so probably it will be fixed with next beta of Safari.