The story so far: In a post and follow-up blog post this week we observed that Wikipedia desktop visits from search (as a measure of traffic volume) had fallen from an estimated 2.5Bn per month in March to just 2.0Bn visits per month in July. Our assumptions included a couple of hypotheses. One hypothesis was that the reason for this significant drop in traffic from search was the result of recent changes in the way Google surfaces Wikipedia related information in search results gives immediate information to Google users and has therefore driven fewer visitors to the actual Wikipedia website. We also hypothesized that the significant drop from May to June could be related to the switch from HTTP to HTTPS protocols for Wikipedia.
Data from Wikipedia showed an increasing proportion of traffic from Google to Wikipedia but on further analysis (see below) both the data from the Wikimedia Foundation and SimilarWeb show similar trends, it is the absolute number of visits which is in decline.
Following feedback from Jimmy Wales we looked further back and it is now clear to us that this is indeed not a sudden development, but a longer trend that has intensified this year.
From May to December last year Wikipedia traffic slowed from 3bn monthly desktop visits to 2.7bn.
So as to rule out seasonality we looked at traffic for the second half of 2013 and saw the opposite trend, a net increase in visits.
In 2015 the trend continued with traffic falling all the way down to 2bn monthly visits in July.
Adding Wikimedia’s data into the picture
As a result of our previous blog posts there was significant media interest and the Wikimedia Foundation joined the discussion by contributing their own data showing recent traffic trends. Our initial observations after looking at this data from Wikimedia and comparing to our own data are as follows:
- YTD Google accounts for a larger % of PVs increasing from ~33% in Jan to ~36% in July.
Graph #4 – Proportion of Page Views from Google to Wikipedia (Source Wikimedia Foundation)
2) We see a similar pattern in terms of the visits we are tracking where Google accounted for 81% of visits in Jan and has been higher than that every month since then.
Graph #5 – Share of traffic from Search Engines to Wikipedia (source SimilarWeb)
3) The significant declines in PVs reported by Wikimedia in recent months (around 20% drop) which accelerated in June and recovered a bit in July. The drops are far larger than the increases in Google’s relative share (3%) which suggests total PVs from Google have indeed fallen in recent months by more than 15%.
Graph 6 – Daily Page Views (Source Wikimedia Foundation)
4) Our PV data for Wikipedia shows a similar pattern of declines accelerating in June with some recovery in July
Graph #7 – Daily Page Views Desktop Devices (Source SimilarWeb)
5) This is consistent with our observation of a drop in visits to Wikipedia from Google of >10% over the same period, especially since SimilarWeb estimates show that Google accounts for >80% of all visits to Wikipedia on desktop.
What are the reasons?
This is a long term trend that seems to be continuing full force. The question is why is it happening?
It is not clear to us whether there is indeed a disagreement as to whether overall traffic from Google to Wikipedia is declining even while search traffic as a proportion of overall traffic is increasing. SimilarWeb data shows a trend happening since at least mid 2014 i.e. small increase in search engine traffic as a % of total traffic against a backdrop of large declines in traffic and especially traffic from search engines which account for more than 80% of visits to Wikipedia, while Jimmy Wales has mentioned a trend in declining traffic to Wikipedia in his discussion of our earlier post.
Both sets of data showed that in June a higher than seasonal average drop in traffic occurred, most likely because of the HTTPS switch that occurred at that time. We will have to wait to see if this has a lasting affect on traffic.
The big open question seems to be what the reason is. We have 3 proposed hypotheses which can explain some or more of the traffic declines:
- Google’s Knowledge Graph boxes which aims to provide direct answers within the search results. Our hypothesis is that a significant part the declines are due to changes made by Google which reduce the traffic sent to Wikipedia through providing information from Wikipedia in the search results. We would ask if this is fair – does improving the user experience on Search justify cannibalizing traffic to the publishers who produce that content – thereby disrupting their ecosystems (for profit or otherwise). It’s possible that the Knowledge Graph is also decreasing direct traffic as people may prefer using search to get direct to the data they want vs going to Wikipedia.
- Seasonality – normally they expect to see a drop of around 5% in traffic from May – June as students finish up school for the year and go on vacation. This seems to explain around a third of the 15% drop seen this year.
- Wikipedia has been kind enough to share their thoughts that their shift to HTTPS might have resulted in drops in traffic overall and from search engines :”This switch to HTTPS impacts traffic in a number of ways. For example, it affected a large number of bots that crawl Wikimedia pages. It also affects how search engines index pages, leading to lower measured traffic.” However, this shift to HTTPS happened in June and could be part of explaining the acceleration in recent months of the decline in traffic. It doesn’t explain a trend which seems to have begun more than a year ago.
What’s clear is that there was certainly an unusual drop in traffic overall which included the majority of visits which come from Search Engines. We will be looking into the HTTPS question to try to determine how much if any was due to that.
We also want to take a moment to apologize to Jimmy Wales. We were pretty excited to have the founder of one of our favorite websites join the discussion. While we can have a tendency to write controversial headlines we always aim to be solid and adding value with our data. We never intended to say that Jimmy Wales confirmed our report, only that he confirmed the trend. We made a mistake and are sorry.
It was really great that the Wikimedia Foundation reached out with data and it certainly helps us get to the root of this issue.