Google’s Vince update, which first affected the US results early in 2009 and then later the UK results in July, is still a mystery. It looked at first like an update designed to boost brands, as Aaron Wall first mentioned – but how could Google, a search engine that relies on algorithms, possibly determine which sites were brands? Matt Cutts later announced that “within the search ranking team we don’t really think about brands”, and mentioned that “we try to return whatever we think the best results are for users”. Which makes sense, but doesn’t help to explain how brands might have received a boost.
There was some talk about how the large brands that gained a boost after Vince also frequently appeared in the related search results. And this is where we think we’re close – here’s the science part (concentrate):
Theory #1
The related searches are now driven by user data. In the past a search on something like “hotels” used to bring up results that looked very similar to the “hotels” keyword, almost as if it had been run through the AdWords keyword tool. Search terms like “hotels ireland”, “hotels in scotland”, “hotels in newcastle” and “hotels england” appeared in the related searches. At some point, quietly, around December of 2008 this changed significantly. Looking at them now for the UK (and using Google’s fairly new expanded ‘related searches’ option) I’m seeing more brands being listed like Travelodge, Last Minute and Premier Inn. The most likely idea here is that Google is now using more user-data to populate these related searches. Theory #1 is that for high volume terms that have a large number of people refining the query, Google records all queries that are refined and the most often searched terms following the original query form the related searches. That sounds complex, but really it just means when people search for “biscuits”, don’t click on the results but instead search for “mcvities”, Google records that and will list it in the related searches if enough people refine their search like that.
Theory #2
When people search for “X”, can’t find what they are looking for within the results and instead refine the query to “Y”, and then one of the results gets a much higher click-through rate than expected (e.g. the result in position 1 gets 70% of the clicks instead of the usual 50% or so), then that listing becomes a candidate to rank for “X”. To put that into context, if people search for “biscuits” and then refine the query to “mcvities”, and one of those results gets a huge click-through rate (which will probably be UnitedBiscuits.com who own McVities) then that result gets a chance to rank for the original search term “biscuits”. So, in short, related search results that do well may be brought into the original search.
From there, Google will also have to check how well it fares ranking on the (probably) competitive term “biscuits”. I’d expect it to place it anywhere in position 6-10 and then see what the click-through rate looks like. If it looks good, or was better than the result it replaced, it can stay or move up. Else, it might move down or be removed completely. So there should be a stronger focus on click-through rate for these terms, although Google will also need to make sure that the site is authoritative and that the page is relevant to the original term so it’s very likely that there’ll be a stronger focus on both authority links and keyword-rich links, as Richard Baxter discovered.
Can we prove these theories?
It’d be incredibly difficult to prove for sure either of those ideas because it involves user data that we don’t have access to (which is probably why the Vince update so far remains unsolved), but we can look at things that can indicate that we’re on the right lines.
Firstly, there’s some evidence that Theory #1 is right, and that Google now strongly uses user-data (and particularly what people searched for after the original query) to form the related searches. Here’s what happens when people search for “time in new york”.
Mixed into those related searches are “dialling code for new york” and “telephone code for new york”. If Theory #1 is right, that implies that people are often searching for the time zone in New York, finding out that it’s waking hours and then calling someone they know, which is classic user behaviour.
Theory #1 is possibly best proved though by searching for anything that brings up a Google one-box. If someone searches for something, then finds the information without having to click through at all – the next search they make will, in many cases, be completely different. They’ve got their answer and moved on. Let’s try “weather” with its lovely, helpful one-box.
Facebook, Hotmail, Yahoo, Argos, Ebay, YouTube, Bebo – nothing to do with the weather and everything to do with user data. Awesome.
Theory #2 is slightly harder to demonstrate, but not impossible. If Vince is all about mixing in results from related searches back into the original query then Google would most likely need to place them in a lower position (probably positions 6-10) while it tests out their click-through rates. Positions 1-5 should, in theory, be more likely to remain static so that the user experience is relatively unaffected. Tracking rankings for the generic terms “car insurance”, “shoes” and “hotels” over a period of 7 days has led to the creation of the below graphs, days at the top and positions on the left. (I’ve lovingly stolen the graph idea from SEOMoz and Distilled).
Car Insurance:
Shoes:
Hotels:
It’s clear that the SERPs have been bouncing around massively in the lower half of the results, while the upper listings are hardly touched.
And now for an example – something that has been puzzling the iCrossing team since Vince was the appearance of direct.gov.uk in the car insurance SERP, it wasn’t a bad page but there were probably better results that could have been there. And why did it keep dropping out and then reappearing? Check out the related searches for “car insurance”:
“Car tax” is an interesting related search but I can imagine why it’s there. It’s possible that there’s a huge number of people that, on buying a new car, search for car insurance and then after looking at the search results decide to instead get the car tax sorted out first. And Direct.gov.uk has the first two positions for “car tax”, including sitelinks, so it’s click-through rate must be massive.
So, with Theory #2 in action, Google keeps trying to bring it somewhere around position 6-10 for the original search “car insurance”, finds that the click-through rate is too low and then drops the result.
And that, folks, are some of our theories on Vince.
A joint piece of iCrossing research from Liz Ayers and David Peiris
Gregory Lyons on 
Jeremy Head on 
Digital Marketing on 
Ali Ahmadyar on 
Dinesh Thakur on 
Dave Oliver on 
Art Rogue Studio on 
Rob Pearson on 
September 23rd, 2009 at 4:25 pm
Really solid work guys – love this post. Sphunn: http://sphinn.com/story/127873
September 23rd, 2009 at 4:35 pm
Nice bit of research, and certainly highly plausible – particularly given the level of both user data and behavioural data that Google have to hand are are able to draw on.
I would also add its in line with their perspective on other channel thinking ie behavioural retargeting etc.
Again great post….
September 23rd, 2009 at 4:45 pm
Interesting post – nice angle and some good theories. One interesting point is that the upper half of the SERPs have always been more stable than the lower half so this fluctuation isn’t new (that I’m aware of). Even going back to Will’s post that you link to shows that the lower results fluctuate more than the top results.
I’d love to see some more data to back this up but great post!
September 23rd, 2009 at 4:55 pm
Theory 3 looks really interesting. A break through!
September 23rd, 2009 at 4:56 pm
That was a great read and clearly explained.
September 23rd, 2009 at 4:59 pm
Interesting thoughts guys. Fantastic post.
September 23rd, 2009 at 5:00 pm
Great lateral thinking
September 23rd, 2009 at 5:00 pm
Click through’s have been dismissed in the past as too noisy, and I’ve not seen anything to change my opinion on that. Have you even considered that brands could still be easily identifiable through normal indexing activities – McVities and Biscuits are likely to appear somewhere together on the web.
September 23rd, 2009 at 5:17 pm
Hi Chris,
I disagree that click throughs are far too noisy – it’s a good enough metric to base a good proportion of Adwords Quality Score on. Moreover, if Google are constantly testing positions, constantly refining, then some noise doesn’t matter as it will be ironed out over iterations.
September 23rd, 2009 at 5:23 pm
Hi Chris, I don’t remember any mention of click-throughs being dismissed before as too noisy (although I think Matt Cutts once said that bounce-rates were tested once, but were too noisy a signal – I need to check on that). I would guess that click-throughs would be a clearer signal to Google than looking at which terms are often used together. Also, Vince was described as a simple change, and your idea (while not impossible) would be a fairly complex task.
September 23rd, 2009 at 6:16 pm
Amit Singhal, a Googler who likely knows at least as much about the ranking algorithm as anybody else, if not more, gave a presentation in 2004 about how clicks are an ‘incredibly noisy’ signal. You can find the presentation here and he mentions it specifically on slide 32.
It’s worth keeping in mind though that just because a signal is noisy doesn’t necessarily mean it’s not being utilized. The link graph is getting more noisy every day, but it’s still obviously being used in ranking results.
Great post by the way, very insightful.
September 23rd, 2009 at 9:08 pm
Awesome, love this kinda stuff
Good work!
September 24th, 2009 at 9:12 am
@DaveShapiro – thanks for the link, I’d not seen that before. Really interesting stuff
September 24th, 2009 at 10:44 am
Dave/Liz – I had a theory that the volume of searches (or % traffic) from “brand + keyword” terms might be having an effect here as well.
I have seen a site whose 2nd highest traffic driving term is “brand + keyword”, and post-vince the page that ranked for “brand + keyword” now ranks very well for “keyword”.
Do you guys think that this has any/much significance?
September 24th, 2009 at 10:46 am
I also disagree that click-throughs are too noisy. Just because a metric can be noisy doesn’t mean that the metric is useless. Noise can always be filtered and the more specific the connection between keywords the less noise there is to filter in the first place. Which is perhaps why strong brands with a focused “message” have done so well in the update: Their improvement is not down to the fact they are household names but because their brands are known synonymously for what they do. So brands doing well in the Serps become a symptom of the update not a specification of it.
Nice work guys!
September 24th, 2009 at 1:39 pm
Excellent work guys. Completely agree with this line of thinking
September 24th, 2009 at 1:53 pm
Fantastic analysis. Very interesting read, Now I’ve got to go play and try some of this out for myself.
September 24th, 2009 at 2:25 pm
I think Google just wants to see websites and businesses really utilizing all components of the internet. If you are writing articles and marketing videos amongst other things than you are on the right path to showing Google you are trying to brand.
September 24th, 2009 at 3:14 pm
For what it’s worth, Google released the search based keyword tool (for PPC) some time around November, which they hyped as being based on user data as opposed to the data in the regular AdWords keyword tool. It leaves me wondering if there wasn’t a larger general shift toward relying on user data that impacted organic results around the same time.
September 25th, 2009 at 9:40 pm
Thank you very much for that magnificent article
September 25th, 2009 at 10:02 pm
A larger shift to using user data, sure, but I still think that even with that shift user data is a very, very small piece of the pie. We might see that grow more and more in the coming years, of course, but for the moment, I don’t think it affects things on a large scale.
September 27th, 2009 at 8:42 pm
Authority “brand” sites have latent traffic potential – they are not using up all their juice budget, so Google just split tests to get a balance picking winners from related search terms.
September 28th, 2009 at 12:29 pm
Excelent analysis! Nothing further! Great job.
September 28th, 2009 at 9:25 pm
Another (more simple) theory is that Google is filtering all kinds of affiliates that implemented iframes in their sites (for example booking.com affiliate program) giving more credit to the actual ‘brand’ site while dumping the copies. That explains the huge fluctuations in rankings in the travel sector in the past 3 months.
September 29th, 2009 at 9:17 am
I’m convinced ‘brand sites’ are simply determined by the number of search queries for the brand name. I think user data will play a much bigger role in the algo in the future though.
September 29th, 2009 at 4:08 pm
It is an interesting theory, although it seems incredibly complex to do and could be defeated by click-farms. Without wanting to elaborate too much, our analysis seems to point to a re-evaluation of the value of brand and product mentions, just a clean up of backlink spam and better attribution of (product) content to the original site.
September 30th, 2009 at 9:20 am
@Jeff – you’re right, that’s very interesting. There could be something in that.
@Gareth & Lorenz – I don’t think Google are looking to boost brands – I think they were just after more relevant results for users. Connecting brands names with products would involve Google being able to understand what words could be considered a “brand”, which I would imagine to be far more complex (in my opinion). I also think that Google will easily be able to protect itself from click-farms, don’t forget they’re able to detect click-fraud within the paid results.
@Bert – While Google could do that, there’s a huge number of affiliates that don’t use iframes (and sites that *do* use frames but aren’t affiliates) which would throw a spanner in the works.
Thanks for the comments guys.
October 10th, 2009 at 11:41 am
In the depression of the 30s, people retreated into buying mostly from established brands. We’re now in a similar recession/depression, so Google are giving appropriate search results. These results highlight what people are known to prefer and be looking for in a period of financial restriction, namely, established brands.
BB
October 15th, 2009 at 5:04 pm
“direct.gov.uk in the car insurance SERP”… Random!
Nice post. I’d like to see a graph of the impact of these “tests” to the revenue of the websites involved…
I’m sure the ups and downs are quite significant!