Sentences can be categorized by AI as positive or negative. You can test it out for yourself through Google’s AI part way down on this page.
Ed Leake said in this video that a more positive ad rank will lead to better ad ranking, is he correct?
As sentiment score goes up, the average position an advertiser’s headline goes up too
Here’s what the above graph means: Every advertiser has an average position in an auction, like 2, as well as a standard deviation like .4. That means, 68% of the time, they show up between positions 1.6 and 2.4.
Each time they showed up, I calculated the standard deviations away they were from their average, as well as the sentiment score.
I then grouped that data based on sentiment score to show the average standard deviations away a given sentiment score is from the average domain positioning.
A sentiment score of -.7 shows nearly -.2 standard deviations lower position than the domain’s average.
An advertiser has two headlines.
The headline with a sentiment score may be more likely to show up lower in an auction (-.2 standard deviations away from average) compared to a headline with a positive score of .9, which may show up .06 standard deviations above average.
When running a test of statistical significance, headlines with a -.4 sentiment score or lower show up in a .44 standard deviations lower position than headlines with a greater than -.4 sentiment score with a P value of .14.
Translation: We can say, with 86% confidence, that having a headline with a greater than -.4 sentiment score improves your ad position, and the average effect is .44 standard deviations.
Furthermore, having a headline sentiment score of .5 or higher, brings the average position up 1.77 times with a P value of .075.
Higher sentiment scores in ad copy are highly correlated with improved positioning in an auction, especially when compared to low sentiment scores.
A key factor in ad rank is relevance. This analysis did little to assess the lurking variable of “relavance” of a headline to a given query.
In fact, when isolating headlines that did not have any matching words from the query, it appears that very negative headlines showed up much higher! (Below, the blue bar are the observations where the headline does not have any of the keywords in the text)
There isn’t enough data to make use of this observation yet. Perhaps once I collect 500k ad results and 1 million headlines I will rewrite this section with better data.
Until that time, it’s been a blast.