I am sceptical of this data - perhaps "Please ReTweet" was found in tweets of a few authors whose content was more likely to be retweeted? However, will try to use Please ReTweet. But if everyone does it, then the effect will disappear

I think you data shows one outlier that should be excluded for the analysis. It badly skews the averages. 

Great post Dan - it would be interesting to break these down into retweet types - e.g. was it charity / causes or just general tweets from general people?

Very interesting! Did the placement of the "Please ReTweet" make a difference? (ie either in front of the tweet, at the end, middle...)

Reading the above reactive comments to this article (all good, btw), I am reminded of the expression "lies, damn lies, and statistics" :-) Nevertheless, I will give the PRT technique a whirl, and collect my own analytics. Thanks Dan.

Good to see you publishing confidence intervals! 
 
This is good stuff, but we must always be careful with any statistical analysis. While I'm sure there are some mitigating factors if you dig deeply enough, these numbers seem pretty overwhelming. Besides, they confirm something that landing page optimization has been teaching us for some time now: If you want people to take a particular action, it's a good idea to ask them do take that action! 
 
And to the commenter who said that there are outliers, it's just a way to abbreviate the graph so that it doesn't have to extend the x-axis off of the page. They aren't outliers. Having said that, it is very deceptive to use an average with a power law distribution. I do not think it means what you think it means.  
 
One last suggestion... If you really want to characterize this sort of post as science then you should substitute the word "suggests" for "proves." Only experimentation can prove; statistics can merely suggest a relationship.

Where were these 10K tweets gathered from? @Hubspot? Or other accounts as well?

Also sceptical. This level of retweeting seems close to that needed for a runaway chain reaction. 
 
If these numbers are right, I would expect any very popular content, combined with a retweet request, to "go critical", with increasing numbers of retweets in every generation than the one before, taking down Twitter.

Very interesting. I will now have to use Please ReTweet a little more often when I'm wanting to share important info. 
 
I hope people don't start to use Please ReTweet all the time now though.

I'm curious about how you have measured the RT's for a specific tweet - I've only found tools that will measure clicks on a link in a tweet or give a total RT count for an account rather than on a tweet by tweet basis.

To me, "please retweet" or "please RT" seems somewhat desperate. Regardless of the data, I'm still a believer in RT'ing for and on a tweet's merits.

And for the past week, I have been annoyed by Hubspot saying "Please RT" 
 
Annoyance for Re-Tweets, no thanks.

I see no causal evidence yet. Tweeps may think (at least I do) that let's say only max 10% of tweets may have an above average 'RT-factor', so if I label my ultimate top-3 ones with PLS RT, than they gain more RT's than the #4-10

I'm not so confident in this data either. I'm apt to believe Gregory's comments above. Did all of these tweets contain links? For example, personal tweets that are not intended to be retweeted may have been included in your sample. Comparing the tweets I would post on my personal account with the kinds I would post on a company feed is like apples & oranges. 
 
I, personally, find the "Please RT" business irritating and desperate when I do see it pop up in my newsfeed.

@Chris, As I said in the post, all the tweets I looked at did not start with an "@" sign. 
And clearly there aren't enough people annoyed by please rt to make it not work.

@Dan 
 
I do understand you did your best to refine the tweets you were using, but were tweets without links included in these data? I think that in most situations, those tweets are not intended to be retweeted to begin with. 
 
My last statement was just a personal preference of mine. I personally do not retweet "Please RT" messages.

@Chris, you've chosen to not use a CTA in your tweets, even though you now know it works?

@Dan 
 
I'm referring to my own, personal (not for marketing) Twitter. Did the tweets you used in your sample only come from company- or marketing-related Twitter accounts? This would add a lot more meaning to the data for me.

I don't think this proves much.  
 
If the author thinks the content is worth retweeting it is more likely than average to be worthy. 
 
If someone is trying to get in to favor with someone then they will follow the request.  
 
I think all of this has to do with your audience. Test and repeat.

Some very perceptive comments. I'd like to know more about how the data was prepared. Did we screen for outliers? (e.g., did Lady Gaga or Oprah tweet 'Pray for world peace, please RT'). I think the only scientific method I would trust is to take x tweets and to send them out with and without 'Please RT'.

@Dan, the biggest difference between "Please ReTweet" and "Please RT" is in the rightmost bar (21+ retweets). The top graph implies that ~20% of all tweets with "Please ReTweet" got 21+ retweets. Is this true? If so, what were these extremely popular tweets about?

I think the Content of the tweet itself (and the Reason why the sender labeled it with a PLS RT) is the significant driver, and not the the type/form of the label

Was the content different when testing? I must believe the content of the tweet had a lot of influence on the retweetability of the tweet. Anything it worth trying, so I'll give it a shot and see if any of our stats change.

A lot of great comments here - clearly we're very interested in this correlation between requests & actual retweets. 
 
I wonder if there's not another factor here - I posit that a certain type of tweet tends to have a "please RT/retweet" attached to it, and that *type* of tweet is more likely to be retweeted in general.  
 
A good test might be to find multiple twitter accts with similar follower numbers/demos/behavior, and have them generate the exact same tweets, but with "rt" "retweet" and no plea attached. Providing the number of accounts activated this way were large enough to be a good sample, and overlapping followers were compensated for in some way, I'd be more comfortable with the data generated by something like that. 
 
That requires an actual pro-active experiment, though, not just crunching existing data looking for correlation (not discounting the effort or usefulness of the latter - just illustrating the addl work a controlled experiment requires).