For decades, PR people have been touting a superstition. We’ve been telling clients that stories that mention their brand in the headline or the first paragraph or photo caption were more likely to be remembered than those that only mentioned the brand in the text. It’s been a common assumption throughout the industry. But, to be honest, I have no idea where it came from. I always assumed someone must have done some research at some point to come up with that conclusion.
Dwyer & Oliveira prove a superstition wrong
Fortunately, Angela Dwyer, VP Measurement Sciences at Lippe Taylor, and Maria Oliveira, PhD, and Research Partner at Elevate1850, decided to actually test that assumption. In the paper they presented at IPRRC this year, they tested the following hypotheses:
- Brand recall will be greater for a brand in an article with a branded visual compared to one without a visual.
- Brand recall will be greater for brand mention in the headline vs. a mention deeper in the article text.
- Brand recall will be greater for a brand with multiple mentions in an article compared to a brand with a single mention.
To test these hypotheses they conducted a survey that measured brand recall for articles with photos, headlines, and multiple mentions. All articles were about the same brand, and respondents were asked first for unaided recall, then aided, and then for brand attributes and their purchase intent.
Bad news for headline and mention fans: Dwyer & Oliveira’s analysis showed that the only factor that significantly increased recall was a photo. You can read their results here.
Let’s mine some data to look for valuable article attributes
All of which got me thinking about all the various factors that we typically track for clients: tone, message presence, spokesperson quotes, myth dispelling vs. myth perpetuating, calls to action, and yes, of course, prominence of brand mention. Which, of all of those criteria, might be the most likely to drive the desired outcome—be it brand recall, messaging, web traffic or purchase intent? What article attributes matter most? (See our story on how to create your own Impact Score.)
The good news is that I have several clients that have lots of data, so I thought a little data mining exercise would provide hints at the answers. I got out my trusty Excel spread sheet and pulled in 39 months of earned media coverage for one client, a total of 16,408 articles. The articles included stories in a wide variety of media: online, print, broadcast, blogs, etc. Each had been human coded for the presence of one or more of the following criteria:
- Positive/negative/neutral sentiment
- Message presence
- Presence of a desirable visual
- Quote from a company spokesperson
- Brand mention in a headline
- Contained a call to action.
Each of those criteria was weighted according to client requirements and priorities. The weights were adjusted so that a “perfect story” would score a 10 and a “total nightmare” story would score a -10. (And if you’d like to create your own index of coverage quality, read “How To Develop a Customized Kick Butt Index of the Quality of Your Media Coverage.”)
Knowing that one of the client’s goals for the PR program was to drive engagement and new users, I then used Google Analytics to pull the following data points on the same 39 months for Goal Conversions, Unique Sessions, and New Users. I also pulled engagement metrics such as Pages Per Session, Session Duration, and Unique Page Views. I ran all that through a Pearson correlation test in Excel.
Results imply that visuals are important
While we all know that correlation is not causation, the analysis of my data lends support to Dwyer & Oliviera’s conclusions. By far the highest correlation to Goal Conversions and the engagement metrics are visuals. Headlines aren’t far behind, and messages came in third. The volume of mentions in an article appears to have no impact.
| Volume of Mentions | Tone Points | Visual Points | Spokespoints | Message Points | Headline | Call to Action | |
| Goals | 0.12 | 0.17 | 0.46 | -0.22 | 0.00 | 0.23 | 0.14 |
| Users | 0.04 | 0.26 | 0.29 | 0.00 | 0.33 | 0.32 | 0.03 |
| Session | 0.13 | 0.31 | 0.47 | 0.02 | 0.32 | 0.40 | 0.18 |
| New Users | 0.06 | 0.28 | 0.30 | 0.04 | 0.36 | 0.33 | 0.06 |
| Pages per session | 0.04 | -0.02 | 0.33 | -0.39 | -0.31 | 0.08 | 0.04 |
| Session duration | 0.26 | 0.32 | 0.48 | 0.04 | 0.32 | 0.40 | 0.26 |
| Unique page views | 0.26 | 0.32 | 0.48 | 0.04 | 0.32 | 0.40 | 0.26 |
| All sessions | 0.25 | 0.37 | 0.52 | 0.07 | 0.37 | 0.46 | 0.25 |
However, while visuals certainly have the highest correlation with getting readers to act, good messaging and headlines are nearly as important. Tone was somewhat less correlated, and the presence of spokespeople mattered not a bit. However, I also found that spokespeople’s quotes were closely related to positive sentiment of articles and the presence of a call to action. That’s a strong hint that, if you want to get your messages out, you should get your spokespeople in front of the media.
The fine print
Now here are the caveats for this data and analysis:
- I did not test for the impact of impressions. First, I don’t believe they’re relevant, and the client doesn’t either. They prefer to focus on the most influential media only, many of which have very small circulations.
- Secondly, the “behavior” data from Google Analytics was not specific to the PR program, and was as likely to have been driven by other elements in the communications program, however this organization does not spend heavily on advertising. Social and earned media are the major components of their communications efforts.
- Finally, let’s remember that all this is a data mining exercise, not hypothesis testing. All we can conclude from this analysis is that there are hints of possible relationships. To actually prove those relationships, we’d have to start with a hypothesis, then collect the data. As Dwyer & Oliveira did in their work reported above.
Lessons learned
There are several lessons learned from this exercise:
- Before you invent an “impact score” (read these tips on how to create one) you need to figure out which elements have impact, and which don’t. This data mining exercise is an example of how to do that.
- Different types of products and brands will likely have different elements that drive engagement and impact. While we didn’t find significant impact from spokespeople quoted, the presence of expert recommendations may be considered crucial if you are selling pharmaceuticals or airplanes. Endorsements by relevant influencers may be more important than headlines if you’re selling shampoo or movie tickets. The key is to test the distinct elements in your coverage to see which in fact have the most clout.
- Data is your friend. If you’re having an argument about what’s important to “good coverage” then it’s always more effective to use data to prove your point than it is raise your voice.
∞
