Paid Search Conversion by Position

To skip straight to the results click here. It's been a little over a month since Google added the {adposition} parameter to its ValueTrack capabilities and we were quick to implement it in order to see what insights we could gain from the ability to see the specific position of all individual ad clicks. Position data available previously was far less granular -- about the best you could easily do was see the daily average position for each keyword in your program.  This view is still valuable, but it masks the variability in position an ad might see in any given day due to a number of factors including: broad matching, day-parting, quality score calculation at auction and other SERP personalization. Our initial review of this new data focused on how click volume varied by position for a sample of keywords.  This added clarity and perspective to what jumped out to many upon Google's introduction of Top vs. Side segmentation and earlier with Bid Simulator data.  Namely the extent to which click through rates are higher at the top of the page and, by extension, how Google's impression-based average position figures don't necessarily represent well which positions are generating clicks. Knowing that CTRs and traffic levels rise exponentially as an ad moves up the page is all well and good, but if we're already bidding what we can afford, spending more per click is not likely to work out, unless we believe we can make up for smaller margins with volume increases and/or we expect conversion rates to improve. We've long maintained that conversion rates don't vary much by position, although we haven't had such a clear view of the position of each click.  We were happy to see Google's Hal Varian confirm our findings in 2009, but he's just their Chief Economist, right?  Kidding aside, Varian didn't offer a lot of detail about the small (<5%) differences they did observe in conversion rates and didn't address potential order size differences by position.  So, I thought I'd take another look now that we have a more robust data set. Methodology We looked at four weeks of click and conversion data for a sample of clients.  The data was restricted to ads appearing on the Google.com domain and keywords that did not include our clients' brand names.  We grouped each ad based on position only, ignoring top vs side information.  Different instances of the same keyword (on different match types, with different targeting options, etc.) were kept separate. It's insufficient to simply aggregate all keyword performance by position because of the bias introduced by bidding (terms with the best conversion rates get bid up the page while poor performers get bid down). We need to account for this as best we can, so for each ad we determined its overall conversion rate and sales per click and normalized those metrics for the various positions in which the ad appeared based on those overall totals.  Here's a hypothetical example: For position 1, the normalized conversion rate of 0.95 means that the keyword performed 5% worse in top position than it did overall.  Conversely, the conversion rate was 19% better than average in position 3. We then weighted these normalized results based on click volume, aggregated them across keywords by position and then unweighted them to get relative conversion performance by position. Results After the number crunching, we end up with a graph of conversion and sales per click that looks like this: Note that we've grouped performance for position 6 and lower on the page.  In lower positions, the data thins out pretty quickly, becoming noisier and inconsistent across the clients we surveyed. Not surprisingly, we found that an ad in top position is likely to perform slightly worse than it would on average elsewhere on the page.  This effect is probably the result of less careful users clicking the first link they see without considering whether it is likely to provide the information for which they are looking. Conversion rates improve over the second and third position on the page, drop at position 4 and then revert to the mean over the lower positions.  This pattern was consistent from client to client, but the potential causes aren't so obvious.  Position 4 will often be the highest listing on the right rail of ad results, particularly for the most competitive and high traffic terms, and it could suffer from the indiscriminate clicking that appears to hurt conversion for position 1. Positions 2 and 3 could benefit from a perception of authority and trustworthiness, a result of their more prominent display, without suffering from careless clicks.  Users might price compare by clicking a position 4 ad, but favor making their ultimate purchase with the companies listed higher on the page.  The farther we move down the listings, we could be seeing a shift to a different type of user who is less concerned about these notions. Next Steps While we did see some differences in conversion rate and sales per click from position to position, they were still relatively minor.  The largest deviation we saw from the average was about 7% -- not much higher than Varian's "less than 5%" pronouncement.  This is also the type of research that would benefit from experimental results with greater controls and randomization applied.  There are a number of other limitations with this data set and methodology that I think can be improved upon, but it's a start, and the results fit with a common sense understanding of paid search. If there's anyone still out there suggesting that conversion rates are significantly better in top position, this is another refutation of that.  There is no magic position in the ad listings and it's much more important to be able to predict your overall conversion potential accurately than chase minor gains by appearing in a specific position more frequently. As we work to improve the rigor of this analysis, we'll also look to expand upon it.  How often a keyword gets broad matched or syndicated on the search network can have a major impact on how well it converts overall and where its average position ends up.  We also disregarded considerations of top vs. side placement here, but that could have a significant impact as well.  Stay tuned!
Join the Discussion