Do you really know when the best time to tweet is?

It has become standard advice to anyone wanting to get the most out of Twitter: time your tweets for the most impact. In this case, the standard advice is also good advice – and so has spawned a mini-industry of tools which promise to tell you when the best times to send your tweets are.

I have tried out quite a few myself, and the more Twitter-scheduling tools I have tried, the more sceptical I have become.

The reason for this scheduling scepticism? The tools tell us very little about how they work, expecting us to take their quality on trust, and almost no-one evaluates the efficiency of different ‘best time to tweet’ tools against each other based on how well they select their times. (By almost no-one I mean, “no-one I’ve come across, but the internet’s pretty big and there are plenty of smart people out there, so I can’t really be the first, can I?”)

In fact, there is an odd logical break in the thought process when it comes to using a Twitter-scheduling tool. It goes roughly like this:

Hooray! I’m going to be really smart and make good use of data analysis to optimise my use of Twitter. And I’m going to do that by… taking a tool just on trust, without being given any data to show whether or not it is good.

Think that sounds harsh? Well consider this: how many times have you come across someone saying, “I’m not using tool X to schedule my tweets because its scheduling algorithm for Twitter isn’t very good at finding the best times”? Instead, we all (including myself in the past) pick a tool that suits for other reasons (cost, usability, other functions, and so on) and then basically take on trust that its analysis of Twitter and its conclusions about what times we should tweet are right.

Yet is it really plausible that every Twitter analysing algorithm out there which spits out ‘best time to Tweet’ answers is really as good as each other? That there are no turkeys and no stars?

Even if you buy the idea that everyone is as good as everyone else, with no brilliance or blunders to sort the field, there is still the question of what you really mean by the best time to tweet.

Is it the best time for readership? Or click throughs on links? Or replies? Or some mix of all three? There is no right or wrong pick from those options. It depends on who you are and what you are trying to achieve. Which makes those tools that simply offer one set of ‘best’ times all the more suspect.

Therefore in taking my first steps in systematically testing out different scheduling tools, I have three criteria:

1. Does the tool give me data that is specific to me?
   This rules out some of the great data published by Bit.ly and by Dan Zarella. Both deserve numerous thanks for sharing such useful data. It has its limitations, however, especially if you are not aiming at the US internet audience, which tends to dominate such data sets, or if you have content which reacts to a specific weekly pattern (e.g. sports which tend to have contests on particular days).
2. Is the tool fooled by Christmas Day and New Year’s Day?
   I want data I can trust. That means data that has a big enough sample size to be robust and algorithms that are smart enough to spot outliers caused by specific events and not to generalise as if they are typical. In other words, if in my testing a tool tells me Tuesday is a bad day to tweet, and it has done that just because Christmas Day and New Year’s Day were both on a Tuesday, then go to the back of the class. Understanding that bank holidays are not normal week days is a pretty basic concept which lots of people in all sorts of walks of life get right. Just because you’re writing a Twitter tool you don’t get a pass on such basic understanding of the calendar.
3. Does the tool distinguish between optimising for audience, click throughs and responses?
   If it doesn’t, “the best time” is simply magic black box technology that might be doing something different from what I want it to do. I’m all for magic and trust, but only as long as it has the same aims as me.

Three tests then.

The Bit.ly and Dan Zarella data fall at the first hurdle, as I said, although are both still well worth a look for general reference. The key messages from them about the best times to tweet are:

• Bit.ly: 1-3pm Monday – Thursday.
• Dan Zarella: 5pm (for retweets), Noon and 6pm (for click throughs).

Test two, handling bank holidays, sees the following Twitter scheduling tools fall by the wayside for deprecating Tuesdays based on the bank holiday mistake:

• SocialBro (I used the free version; the Pro version lets you set its timing algorithm to work on different subsets of users, which is useful if you want to distinguish between – say – journalists and friends). On timings generally, it went for 8am, 5pm and 8pm.
• Tweetwhen. On timings generally, it went for 6pm on a Thursday.

Test three, lumping together the different possible forms of optimisations clears out from the field:

• Hootsuite. Ouch. I am a great fan of Hootsuite, so it falling at this stage did make me pause for thought. However, the logic behind this test is a good one, so fall it does. (An an aside, it is probably a good thing it does for two other reasons too. First, Hootsuite is very tight-lipped about its scheduling tool, pretty much saying nothing I can find beyond ‘ooh, we’ve got one’. Second, when I’ve tried scheduling a large volume of tweets to see what time slots it suggests, it basically spaces things out around an hour apart from early morning to late evening. Better than nothing, and a feature I use, but not terribly advanced, at least so far.)

So who is left standing of those I tested?

• Crowdbooster. It offers only one optimisation, but is clear what it is for –“to reach the most people”. It breaks down by hour and by day, and the Tuesday data is free of Bank Holiday ills. Its conclusion: 11:55am, 4:55pm, 5:55pm (Tuesday-Thursday). One gremlin to note: its summary of this information strips out the days and ignores the minutes, without rounding off; e.g. 11:55am becomes 11am. Make sure, therefore, that you look at the full data.
• Followerwonk. Like Hootsuite, this is a great tool I find really useful. It simply gives one combined graph called “most active hours”. Unlike others, this one has the peak being in the morning 10am-1pm.
• Tweriod (Premium Analysis). It also gives just the one graph, this time for when “most followers online” broken down by day and hour. It has all weekdays pretty much the same. Unlike Crowdbooster it does not give worse scores to Mondays. Overall 4pm-6pm is best.

The conclusion from all this? Be careful. Be very careful. In the (albeit all free/cheap) tools I have first tried out, there is none that passes all three tests with flying colours. Those that do sort of pass through – Crowdbooster, Followerwonk and Tweriod – all give different answers beyond weekdays being better than weekends.

Are Mondays worse than the other weekdays? They disagree. So too on the best times, with quite different answers from Followerwonk (late morning) and Tweriod (late afternoon). For better or worse Crowdboster says late morning and late afternoon. The mixed pattern of times cited as the times to use gets even worse if you relent and let back in to the field some of the early fallers.

In other words, if you think one tool is making it easy and quick for you to optimise your timings, you are (on the evidence I have seen so far) taking a big gamble. For every answer a tool gives you, there is another supposedly similar tool that tells you otherwise.

Footnote: I have heard good things about SproutSocial but have not tried its Twitter timing analysis service as there is not a free/cheap option available.