Twitter Is Stealing YOUR Dopamine - Find Out How!

I take random guesses as to how the Twitter algo functions

Shadowbans have always been mysterious, mostly due to the “shadow” part. Twitter’s never discussed them or admitted to using them before. The average poster doesn’t even believe in them. That’s understandable - it sounds like massive cope. “Oh, my posts got four likes. I must be shadowbanned… There’s a conspiracy to hide my my subversive memes from the world! Possibly George Soros did this.”

Recently though, shadowbanning’s been confirmed as both real and ongoing. When Elon first took his position as head of twitter, several photos of the Twitter admin dashboard were leaked. Most were cropped in so far you couldn’t see dates, but a picture of

Libs of TikTok

‘s profile had a one visible off to the side - “last access 2022-12-07.”

This suggested the photos weren’t relics of some archaic moderation system long since discarded, but recent images of the current dashboard. Elon then commented all-but-confirming the system’s existence:

If you’re still unconvinced, he followed up with this tweet:¹

Okay, so shadowbans are real. But how do they work? How do you know if you’ve got one?

We’ve long since had tools like Shadowbird² that try and unravel this for us. They show four categories and how they’re impacting an account who’s handle you enter. You can’t test a private account because the program doesn’t have secret access to the Twitter API - it just looks at what’s externally visible.

What do these categories mean?

Search suggestion bans mean your account won’t autofill to someone searching your name, making it extremely difficult to find you. This isn’t absolute - It’s based on something called “tie strength,” a metric of how closely two accounts are related. If you have lots of mutuals with someone and interact frequently then you’ll have a higher tie strength to that account, and the chances they’ll be able to find you when searching are higher. The tool uses a dummy account with 0 followers, 0 following and no twitter interactions to test this, so it’s difficult to know how through the tool how closely you have to be tied to someone before they start being visible in your searchbar. Anecdotally though, I don’t even show up to my own girlfriend - so either you’ve gotta be interacting a ton to trigger it or my relationship’s in trouble.

Search bans prevent your tweets from showing up in search results entirely. This isn’t related to tie strength. While the tool shows that I don’t have a search ban, I actually kind of do - Check out this thread. Now try and find the thread by searching. Can’t do it, right? My tweets from the last month show up, but beyond that it’s gone unless you’re directly linked to it.

Ghost bans are made up of both a search ban and a god-tier gaslighting technique. Everything looks normal to the affected user, but to others their threads are completely shredded. If you reply to your own tweet, the reply won’t be visible to anyone else. It just doesn’t show up. They have to go to your profile and look at “tweets and replies” to see the full thread - they won’t do that though, because if there are no replies visible under the tweet they’ll just assume you haven’t replied to it.

Reply deboosting tests whether your tweets get hidden behind the “more replies” link. People who follow you can usually still see your posts, but there are conditions under which they can’t. Nobody seems to know the conditions.

Astute readers are probably saying, “Wait, they’ve got all that stuff but they don’t have a big red lever to simply turn down impressions?”

They likely do have that, and more. The tool’s extremely limited in its detection capabilities because it doesn’t have access to any more of Twitter’s data than you or I. Everything it checks is something you could verify yourself on an alt. It can’t look for algorithmic manipulation because there’s no way to positively confirm or deny it happening without explicit access to Twitter’s codebase. Looking at the picture of

Libs of TikTok

's profile gives us clues, though:

Right off the bat it’s clear there’s more than 4 modifiers applied to the account, which confirms what we already knew about the tool - it’s not testing for everything. It’s less clear what the visible modifiers do, or how many exist in total. It’s unlikely

Libs of TikTok

has every single existing modifier applied to their account, so I assume many exist beyond what's shown here.

Trends Blacklist seems pretty straightforward - your content probably won’t show up on the trending page even if it meets the requirements that would otherwise allow it to.

Recent Abuse Strike likely explains something all Twitter OG’s have known forever - after you get locked out of your account for a flagged tweet, your next month’s engagement gets turbo-nerfed. In some cases the nerfing persists forever. This aided my decision to appeal my flagged tweet instead of deleting it in my recent crusade. I assumed I wouldn’t get a strike if I won the appeal, but it seems even that doesn’t save you. My account was free of shadowbans prior to the incident but afterward it had the search suggestion ban and reply deboosting applied, and my engagement dropped off steeply. Post-strike my 6k follower alt now receives about as many organic impressions as my 47k main. Pre-strike my main was getting 4-6x more.³

Notifications Spike is beyond me. Maybe it silences notifications from your replies? I’ve found that replies from new, spammy or aggressive accounts get hidden from my notifications a lot. This might be it, though I’ve never seen it happen to an account with over 1000 followers.

Twitter Blue Verified unfortunately does not wipe existing negative modifiers off your account. I bought it to test this and everything is the same. That leads me to believe these greyed-out tags might be unrelated to engagement modification, functioning more as categorical identifiers. I feel a bit silly writing that though, because what’s the point of developing categorical identifiers if not to take action based on the categories? It’s possible the yellow modifiers are applied based on some combination of these.

I don’t have hard evidence to support this, but I’d be willing to bet a lot of money there’s also positive modifiers. When I was in my woman era the engagement I got simply made no sense given my follower count. It’s always possible every single thing I posted that whole month was a verified banger, but I find that hard to believe for a couple of reasons:

1. The period during which I got an unusually high number of impressions for how many followers I had lasted exactly one month before abruptly returning to normal.

2. A lot of those posts were simply terrible. Not even the type of terrible normies love, just… bad. It didn’t matter though. The algorithm sent every one. I experimented by tweeting total nonsense, even a 14 character keymash would get 300+ likes.

I obviously don’t have access to Marie's account because she’s an independent poster who definitely isn’t me, but from what I saw from the outside a similar thing happened to her. She had two months where her engagement-to-follower ratio became ungodly before things leveled off to normal again.

The picture of Charlie Kirk’s profile supports my theory - he has a Do Not Amplify tag. If they lacked the ability to amplify a profile, that tag wouldn’t exist in the first place.

I’d assume the Active tag function positively, too - it makes sense to reward the people who produce the most content to continue doing so.

It also makes sense to reward a person’s first post after a long period of activity to get them hooked and hopefully pull them back - possibly there’s an Inactive tag that confers this before leaving after a user tweets again.

The most surprising positive modifier I’ve experienced came from privating my account. After I got my strike, the number of non-followers liking my posts dropped to 5%. Before it’d been above 50%. This told me the algorithm wasn’t showing people my posts unless they already followed me - if Twitter wasn’t showing new people my posts then wasn’t I already effectively locked? Might as well make it official. The “Twitter OGs” said it’d increase engagement too. I admittedly didn’t believe that one, but they were right. I’m not sure whether this is achieved through a modifier or if there’s an entirely different set of algorithms for locked and public accounts. I’m currently leaning towards the latter.

I made a post commenting that privating offset some of the engagement loss I got from the shadowban and I received 2 distinct sets of replies. Ones like this, implying I’m not actually shadowbanned and it’s all in my head:

And ones like this, saying “Oh my god, you’re still alive?”:

I think this means they decrease the tie strength needed for your followers to see your posts when your account’s locked. If I had to say why they do this, I’d guess it’s a risk calculation. Private accounts have a low chance of upsetting someone into reporting because most people can’t see their posts. If a private account quote-tweets someone and their followers mock the person, they don’t even know they’re being mocked. This makes showing your content to more followers “safer” for Twitter if you’re private. The risk you’re a spam-bot is similarly low - why would you make a bot to spam people who can’t see your reply? Despite this, there’s another OG rumor that staying private for a little while results in a gradual engagement decrease. Not sure what Twitter’s reasoning for that would be, but I’ve learned not to doubt the OG rumors. Hopefully by next week I'll be able to update you on that one's validity.⁴

1. Bari also covered this in part two of The Twitter Files, but her thing is longer and more boring than my snappy three-tweet summary.

2. The domain changes frequently, so if the link’s broken you can just search “Twitter shadowban tester” and it should take you to it.

3. I test this by tweeting the same thing on both at the same time and looking at the impressions over the first few minutes, before retweets have a chance to seriously change the trajectory of a post. It’s obviously not perfect but I can’t think of a better way outside of capturing a Twitter engineer

4. Unless I get washed out by the follower churn.

Zero HP Lovecraft

suggested I stay privated for 31 days to see if it'll clean the shadowban off, but on private you lose followers so much more faster than you gain them due to low visibility - currently at about 100 a day. Staying private with casualties like that's gonna be a true test of my balls.