Twitter artist metrics, and whether Twitter “deboosts” or “suppresses” tweet engagement
There has been suspicion among some artists that Twitter’s ‘algorithms’ suppress tweets with certain properties (e.g. tweets containing links, or containing certain words, etc).
I am going to explain why this is unlikely. I cannot *prove* that it doesn’t happen — to do so would require access to Twitter’s internal systems — but there are many other, better reasons to explain why this occurs.
This article is in two sections: the first section below responds to particular tweets, and the second section talks about general cases of deboosting.
It’s physically not plausible to have witnessed this, *particularly* on your own. I have access to Twitter’s developer tools, and I would have trouble ‘witnessing’ that even while monitoring thousands of artists — because the amount of data you would need to isolate out other factors to even a slightly satisfactory level is absolutely gigantic. Here is just a tiny fraction of the picture that you would need to build:
Time of day of tweets
Can affect tweet engagement if your audience is in a particular region.
Luck
Someone big retweeting your work can put anomalies in your data (this tweet got big, but this one didn’t) and can be hard to detect, so it can ‘look’ like something else is causing the lower engagement.
Composition of tweet text
How easy is it to read, does it have context, what language is it in, does it contain hashtags, and so on — all can affect engagement.
Number of images and ‘cropping luck’
More images means a smaller viewing space for each image before they are opened by the viewer, which means less chance of them being “easily browsable”. Cropping luck is relevant but there is honestly nothing you can do to change this — there is a VERY good reason Twitter doesn’t let you crop your images yourself.
Just imagine Donald Trump posting pictures out of context, intentionally cropping to parts of the image that push a narrative — most people don’t open images to view the full thing, and it’s all too easy to crop it to a point where it doesn’t look like the image has further detail to see. Similar abuse potential, by the way, is the reason Twitter has never implemented an edit button — most would not view the edit history, and it’s all too easy to distract viewers from looking at said history by implying the edit is just a typo, for instance.
Reputation, brand and profile of the artist/user in question
This can affect engagement a lot, but it’s difficult to quantify. For example, those drawing primarily sexy pin-ups with no tweet text tend to get a lot of engagement (there’s no language barriers, the art has mass appeal, and less time is needed by the user to evaluate if they want to retweet since there is no text).
Talking about controversial issues — even things that shouldn’t be controversial — can also reduce tweet engagement somewhat, such as politics or current events. This is a complex subject to go into, but I would speculate that artists who are outspoken on social issues gain a slower rate of followers, but a higher % of follower engagement. To give a hypothetical scenario:
Suppose “Artist A” is considered ‘left-wing’, and is outspoken with regards to climate change, BLM, and similar topics. This will cause less right-wing users to follow them, resulting in lower follow rates (while a slight increase in left-wing followers would be observed, many of them would follow anyway). While this artist has fewer followers, their followers are likely to have matching interests — and those with these interests generally take a more active role in engaging with topics, be it politics, art or anything else. As such, one would expect to see a higher % of engagement from those followers (though not necessarily a higher % of tweets seen — simply a higher % of engagement on those that are). This is, presumably, a particular trend for US artists given the highly polarised nature of politics there.
I’ll take this little moment to remind everyone that right-wing politics is malicious crap; I did not want this to look like a ‘neutral, both sides are okay’ paragraph. That side is very much *not* okay.
All of the above points make it extremely difficult if not impossible to reliably compare engagement of one artist versus another, which is a major hindrance in gathering reliable data for this. The logic, however, points very much to the conclusion that algorithms are not what is suppressing a given artist’s tweets.
This is, frankly — completely normal for social media. You shouldn’t be expecting even 10% of your followers to see *anything* you post — that’s already a very good amount, and involves no algorithms whatsoever.
I analysed my current followers on Twitter to give an example, and found that the median following count was 634. Even if each of those users they follow posted one tweet every two days on average, that’s over 300 tweets per day to scroll through.
Here’s how mobile Twitter looks on a large phone (I’m using a Galaxy Note 9, which is about as big as you can get screen-wise). As you can see, one or two tweets with images will take up the entire screen easily — so if for example, 25% of the 300 tweets per day are images, your followers would have to scroll an absolutely *huge* amount every single day to not miss your posts.
Most people don’t browse Twitter quite that often, and when they do are not always scrolling down to see everything — meaning that for each tweet, a majority of followers will likely miss it. Retweeting your art is a good idea in general for this reason, and isn’t anything to do with algorithms stopping people from seeing your work.
Deboosting by words
This is definitely not happening. There’s no reason whatsoever for Twitter to do so.
For many artists, a majority of followers are there to see your art and not much else (and to show it to friends via retweeting). When people consume social media posts, they skim over them to get a basic idea of what they are, and thus what to do with it.
For example: if you write ‘commission’ or ‘paypal’ in a post (particularly if it contains prices), a follower skimming over your post will assume you are posting commission information. Your follower will assume from this that the attached images are sample pieces, and that many of their own followers are not looking for commissions, and thus have a far smaller chance of engaging with that post unless they themselves are looking to commission you.
(By contrast, a post that has a single image, has the word ‘commission’ but doesn’t have any prices is more likely to be interpreted as an art post that was a commission for somebody, and will not have the same problem).
Also note that language barriers play a part in this; lengthy text in a tweet will often discourage those who don’t speak the language fluently from engaging with it (some will, using the translation feature, but especially on mobile Twitter this means leaving the Twitter timeline by clicking the tweet, then clicking translate etc — a lot of extra effort where social media is concerned).
Deboosting by links
The first major pitfall of assuming that “I put a link in my post, and it got less attention — therefore the algorithm is suppressing posts with links” is that much like the scenario described above, this misunderstands how your audience views your posts.
Links, by nature, are things you put in posts generally for three possible reasons:
- To advertise something
- To show the viewer something that you can’t post on social media because the format doesn’t suit it, such as a long essay or something with a specific design not easily put into an image file or text
- To source something you said in your post, such as when talking about the news
In each of these cases, the premise is that the viewer should go there either as a ‘call to action’ (e.g. subscribe to my <website>), or to read more information than the tweet can allow. Viewers, therefore, are less likely to engage with posts that seem like this because:
- In the case of a call to action, the viewer doesn’t necessarily want to spread said call to their own followers, and will feel the need to research it before proceeding
- In the case of a link to more information, it requires more effort on the viewer’s part to fully ‘understand’ the nature of the post and whether they approve of its contents. When unsure and not in the mood to go to the link and read the full information, people generally will just not engage with the post— they will read it, and move on.
The second major pitfall of assuming links ‘deboost’ tweets is that tests on this subject are almost universally badly flawed and do not isolate out other, far more relevant factors. Let us, for the sake of argument, take a hypothetical scenario.
Suppose an artist posts all of their work without links. One day, they post art with a link, and it gets significantly less engagement than usual, so they decide to test posting more art with links and get similar results. Conclusion: tweets with links are suppressed by Twitter.
This relies on the very faulty assumption that the usual nature of the artist’s posts are not relevant to the conclusion, when they are extremely relevant. Viewers generally will remember the pattern in which you make your posts; e.g. “Here’s my art! I hope you like it! <picture>”, or “I drew this, here’s some text about what I was doing while I drew it! <picture>”.
When that pattern suddenly changes, and there’s a link in the post, viewers who are not stopping to read the full post (aka, most of them) will assume that this is not a “normal” post, and thus will be less likely to engage with it (even if all the other text is exactly the same as normal — the pattern is what matters).
The same issue occurs with the opposite hypothesis (an artist who posts art with links, trying it without links). Even though the pattern has changed, the user has less to “process” — all they have to do is look at the art and decide what to do with it.
What matters here is not ‘whether you link to your Patreon/ArtStation/whatever’, but consistency.
Deboosting by hashtags
The primary reason this tends to look like deboosting is because hashtags are —for lack of a better phrase —going out of fashion. That may seem like a bizarre statement, but there are both technical and social reasons for this.
Social reasons: hashtags aren’t ‘natural’ language
Hashtags look very odd in text. No words in any known language I can think of begin with a hash symbol; as a result, when you see a hashtag in the text of a post it stands out like a sore thumb, especially if it’s in the middle of the text.
This is compounded by the fact that hashtags, in order to be more easily searchable, are hyperlinks and thus are coloured as such, separating them from other text.
This is relevant because it has the same ‘problem’ that links do: a viewer seeing the hashtag in your post will feel they need to ‘understand’ that hashtag before they will retweet your post, or may assume that the hashtag is for advertising purposes (for example, on a commercially commissioned artwork). This leads to lower engagement.
Technical reasons: hashtags are easily abused and prone to human error
Initially, hashtags were conceived as a way to make it easy to mark posts as being related to a particular topic. For example: a person searching for posts on “League of Legends” might be frustrated to miss out on all the posts that contain “LoL”, “LeagueOfLegends”, “League” and so forth; hashtags provide an unambiguous way to refer to a subject.
The problem with this is twofold. First of all, when it comes to automated social media accounts, hashtags are often abused: there are no end of accounts that attempt to push topics to the forefront of social media relevancy by spamming posts with a particular hashtag or content in them.
Secondly, humans — unlike bots and automated accounts — don’t write things the same way, or in the same context. For example, let’s imagine a trending topic is “Julian Assange”.
Some people will write their post with “Assange” in the text, some with “Julian Assange”, some with minor spelling errors, some with capital letters and some not — the list goes on, all of which are relatively easy for an automated algorithm to spot and group into one category (as a crude example, an algorithm could account for spelling errors by checking similar phrases for Levenshtein distance). A hashtag, on the other hand, is absolute — either it is the exact hashtag or it isn’t.
Due to both the automated aspect of hashtags and the fact they’re often associated with advertisement or pushing a topic, they tend to result in lower engagement for previously explained reasons.
However: there is one plausible method by which hashtags might theoretically cause Twitter to deboost a tweet, which will take a little explanation.
As previously explained, Twitter does not take kindly to bots that automatically tweet with hashtags. On a technical level, there is no discernable difference between a bot that tweets entirely on its own with auto-generated text and hashtags, and an application which lets a user schedule tweets, like Buffer or Hootsuite — to the Twitter servers, these look exactly the same, as the program simply tweets it on your behalf. Twitter cannot know if you wrote it or the program wrote it for you.
As a result, tweeting via third-party apps — rather than the Twitter web client — might plausibly cause a tweet to be deboosted if Twitter thinks it was auto-generated and not written by the user for the third-party app to post. This is speculation on my part, however: I have no evidence to suggest it is happening.
Deboosting by mentions
Deboosting by mentions is the same as other things: it’s a case of users seeing a blue-coloured mention and being less likely to engage with it, in a similar vein to hashtags and links (though usually to a lesser extent).
Advice on maximising art tweet engagement
The following advice is not intended to suggest that chasing engagement and followers should be an artist’s top priority: it is to help show what works and what doesn’t.
As such, your own preferences should come first, but the below advice is there if you want it.
Keep your tweet structure consistent
This sounds a bit off-putting — as if I’m saying you should tweet like a robot, but that’s not it. What this means is that your art tweets, at a glance, should be similar in structure.
For some artists, that means posting a picture and nothing else. For others, it means the name of the character followed by links on new lines, and for others still it means a sassy one-liner — it doesn’t matter what it is, but it matters that it is consistent.
When it is consistent, your followers will have an easy time recognising which of your posts are art posts and which are not, allowing them to more easily engage with them. It also allows you to have a more recognisable brand, so to speak.
Some good examples of this:
- cutesexyrobutts (NSFW)
- GUWEIZ
- Kajin-man (NSFW)
- Dandonfuga
Keep the text short
This is applicable to all tweets about anything: the more text it contains, the more likely people are to skip past it. This is because of the aforementioned issues for the viewer with having to understand/know whether they approve of the text before retweeting it, but also because it takes longer. Another significant reason is language barriers; e.g. if you are for example posting your art with 200 characters worth of text in English, non-English speakers and those not fluent in it will be much more likely not to engage with the tweet.
Remember that Twitter isn’t just people at their desks scrolling with all the time in the world; it’s also a lot of people on phones, browsing their feed on buses or trains or whatever and potentially scrolling through their feed quickly. Long text often leads to less engagement because it takes more effort to engage with it.
(Not a big deal) Avoid replying to your own posts
This is a difficult one, since many artists use this to e.g. post corrections to an already-tweeted artwork.
The problem with this is that generally, replying to your own post creates a conversation thread. In order to stop a single person’s giant thread dominating a feed for pages and pages, Twitter will not always show every tweet in the thread (and will not always show the first tweet). This is because it assumes the latest tweet is the most relevant tweet, which for many cases is true but for art often is not.
There are presumably internal Twitter metrics to gauge when this happens, but my advice is to try and avoid it if possible, as it runs the potential risk of accidentally hiding your main tweet from immediate view. Usually one sub-tweet is alright, but regularly having a reply to your tweet (because you e.g. don’t want to include certain info in the main tweet) is imo a very bad idea, as explained below.
I think it is a serious own-goal to tweet your art, then tweet your Patreon links or other links in a reply below it — replies are seen by far, far fewer people than the original post (as when your followers retweet your post, those seeing the retweet do not see the reply unless they look to see if there are replies, and then go far out of their way to click on the expand replies text — and most will not, partly due to effort and partly because they assume the reply is people praising the art or other things they won’t find relevant). As stated before, decide on a general structure for your tweets, then keep it consistent. There is no need to avoid linking.
Note this doesn’t apply to replying to someone else’s reply on your post — only to your own replies on your own post.
Absolutely avoid embedded links
By embedded links, I mean using a link without having images in your post (so this isn’t really applicable to art tweets most of the time).
If you have an embedded link, viewers can’t click on the image to view it in larger size without being taken to the site it links to, which will heavily diminish interest in looking at the link and the image at all.
Things that *do* get suppressed by Twitter
This is a very short list, largely because it only has one source: NSFW content.
As many people know by now (often by checking shadowban.eu), Twitter does put certain suppression on accounts it deems to tweet “mature content”: this applies to ALL tweets on that account, irrespective of whether the tweet itself contains mature content or not.
Search suggestion ban
If this is active (usually it is if your account is deemed NSFW), the main Twitter web page search bar won’t show you in its pop-up results.
Note that this isn’t a fully automatic “if you are tagged as NSFW, you never show up to anyone in the search popup” system; depending on tie strength, a measure based on e.g. how many people follow other people that follow you and so forth, they may still show up in the pop-up results. You will always show up in the people tab of the search results page, however.
Search ban
As Shadowban explains, this is generally a temporary status for accounts deemed to be e.g. violating rules or tweeting excessively, and thus is rarely something to be concerned about unless you’re doing things you shouldn’t be doing. It stops your tweets showing in search results at all.
The reason it exists is to protect Twitter search from, to take an example, bots trying to flood a topic with a particular narrative; a user tweeting excessively will be temporarily barred from search results to avoid abuse of the search system in this way. This is almost never going to happen to you unless you’re trying to do something like the above.
Ghost ban
Very similar to the above, this only happens if you are doing things you shouldn’t be. Generally this is applied if you suddenly start following loads of accounts or tweeting (replying) excessively — again to avoid abuse.
Reply deboosting
Reply deboosting stops your replies on tweets being seen unless people click the “show more replies” button. This isn’t universal; if a user follows you, your replies will never be hidden to that user. This often happens if you are excessively tweeting, replying, or other undesirable behaviour.
An example of why this is useful: there are lots of bot accounts that search automatically for posts that talk about “locked accounts”, then automatically reply with some scam about “this person can restore your account for $200” or whatever. Reply deboosting serves to stop such bots getting views.