Welcome to the second post in a ten-week series on how to generate CMO-level insights from SEM ad copy testing. This week, we eliminate customer acquisition cupid-killers and play match-maker between you and potential customers by sharing a critical SEM ad copy turnoff to avoid.

Worst-performing word in retail SEM ads

Back in 2009, OKCupid analyzed online dating conversations revealing the worst words to use in digital pick-ups (Spoiler alert: “ur”, “r”, “sexy”, and “beautiful”). Now, Boost Media’s data scientists have applied similar natural language processing techniques to retail SEM ad copy. Based on the more than 1.3 billion impressions analyzed, we know that “available” is a major buzz-kill, lowering CTR by 6% and CPI by 49%! (CPI, conversion-per-impressions is a metric that takes both conversion rate and CTR into account).

Don’t State the Obvious

Seemingly benign, “available” takes up a precious 11 characters (including spaces) that could be used to woo customers with product benefits. Customers presume that advertised products are “available” without it being stated. In a post-microblogging society of split-second attention, deadweight is a performance killer.

Lessons for your CMO

  1. “Eliminate deadweight” should be central to any cross-channel messaging strategy. Boost Media’s overwhelming data, the popularity of 10 second Snapchats and 6 second Vine videos all confirm this trend.
  2. Use SEM to identify specific performance-killers. Starting with “available,” rigorously test potential deadweight terms in SEM ads. Use Boost Media’s framework to test at scale, identifying words to avoid in upper and lower-funnel advertising contexts.

Tune in next Tuesday for more CMO insights plus retail SEM ad copy benchmarks.



We looked at data from over 1.3 billion impressions across 250,000 ads from 15 AdWords non-brand desktop search accounts from January 1 to June 30, 2014

To normalize factors such as bid and keyword differences, we limited comparison to ad groups that had at least one ad variation containing “available” and one ad not containing “available” where both ads were paired with the same keywords with the same bids over the same period.

When determining whether or not the performance impact of including a term was statistically significant, we used non-weighted calculations and a 1-sided paired student’s t-test with a p value of < .05. When examining the degree to which the impact was seen, we used weighted metrics and did not calculate statistical significance.