First Draft Club

The Algo’s Just Not That Into You

Something weird is happening on Farcaster, and it’s time we talk about it openly. This isn’t just “home feed doing home feed things.” It’s much deeper and more concerning. 

Some casts are skyrocketing with 10k+ views from accounts that have 1k followers. Meanwhile, longtime casters with 100k+ followers are seeing reach crater by 50%. The math doesn’t add up…unless something’s being boosted.

We’re watching artificial amplification play out in real time and this experiment needs to end now before more trust is lost. 

The algorithm cares more about engagement quality than raw follower numbers. There are several Dune queries who show the most popular casters by the numbers, but they do NOT align with the Farcaster Leaderboard. We know that not everyone’s engagement is weighted equally, and most of us have come to begrudgingly accept this fact. 

The home and trending feed have always favored certain accounts, and as little as two engagements with those accounts can amplify a cast to trending.  We see it and have come to terms with it. 

But recently, we’ve seen a new trend emerge: artificial algorithmic boosting…to levels never before seen on Farcaster…and we need to stop dancing around the subject and talk about it. 

A user with 1,000 followers might have a highly active niche audience, while someone with 250,000 might have a large but disengaged base…but we have Dune queries to show us how “active” each account’s followers have been over the last 7 and 30 days. 

Very large accounts (6-figure followings) tend to have about 50% of their followers be considered “active” under this metric (active=have casted, liked, or recasted).

Newer accounts skew higher because many are only a few weeks old, so most of their followers are obviously active. 

If the algorithm is boosting based on more active followers, it could explain the great discrepancies we are seeing in reach, so we need to look deeper. 

A new tool has emerged that lets us look into the number of unique accounts that interacted with a user's content in the last 30 days, and the data is very telling. This data counts everyone on the protocol, not just those with good spam labels. 

If you think your engagement has gone down a lot over the past few weeks, you aren’t imagining it. Usually, if 2-3% of your followers interact with your casts, you are doing great.  But what if most of the people who follow you never even see your casts in the latest version of the algorithm? Spoiler alert: they don’t. 

Let’s compare some of the biggest, most established accounts on the network to newer accounts who I believe are being artificially amplified. 

These accounts all have well over 100,000 followers each (even more on the protocol!), cast daily, and consistently engage back with the network. Their content and casting patterns have not changed noticeably over the last 30 days. 

The number of unique accounts they have reached in the last 30 days and how much their engagement is down is as follows:

15,400 (down 67%)
7,500 (down 49%)
1,500 (down 29%)
2,700 (down 36%)
6,500 (down 55%)
4,200 (down 54%)

Now let’s compare this to one of the new accounts that I believe are being purposely placed into many users’ home feeds (note that these accounts have 1,000-3,000 followers). 

10,500

The number of views on many of their casts are also between 10,000 and 18,000 views. Remember, the data above shows interactions, but even more people have seen these casts!

A user with 1,000 followers getting seen by 10,000+ people requires external amplification.

That reach doesn’t emerge organically. The algorithm chose to show them beyond their organic social graph. 

Observing a few examples of weird reach/follower ratios doesn’t necessarily prove systemic bias.  Visibility is complex and can swing based on timing, interaction graphs, and reposts by top casters, but the math doesn’t add up without boosting.

The people getting boosted didn’t necessarily ask to be. Now they’re targets of backlash; they are victims as well. 

Meanwhile, those who have not had this type of algorithmic boost feel invisible and demoralized, even if they’ve consistently added value to the community.

It’s creating a toxic imbalance that erodes trust across the board. We can politely say “vibes are off” and tell people to just mute what they don’t want to see, but that’s not going to solve the issue at hand. This is deeper than engagement FOMO.
We are watching something we’ve invested time into start to resemble the very platforms we came here to escape.

The algorithm may be silent, but its choices are loud. If it is hand picking winners, the rest of us should stop pretending we’re playing the same game. Invest your time here accordingly.

carwash enthusiast

@ceejifi everywhere else social connections are made

@jacque and @zwillow just did this experiment!

They may both have thoughts!!!

Honestly if I can figure out how to do @adrienne script I’m hopping on the unfollow trend. I know I spoke about it when @pichi did hers too. 

So be kind rewind ya know. 

https://github.com/atenger/farcaster-unfollow-refollow-everyone

I’m going to be refreshing all my following soon to see how the algo through removing who I follow. I have been talking about this for a while more than a few months and honestly the algo hasn't gotten better so now with this tag from @pichi I’m going to be just ripping off the bandaid. 

Please remember this is an experiment.

The app being used is created by @compusophy and I have downloaded the backup file

https://farcaster.xyz/miniapps/zuk8qhrAxueF/compusophy

https://farcaster.xyz/pichi/0x7ce2a7f3

https://farcaster.xyz/~/conversations/0xb1405e816ff3cfe9dceac8ac7491fed457517cbb

Pi-chi.eth Artist /animeoutcasts
Hosting: /sakura /kyoto /anime-manga /farcasterunion

https://portrait.so/pichi

stream of concious writer, musician, & artist | co-founder of /wac

NFToholic, Web3 native, Metaverse MC, hosting EZ’s Corner for 5+ years Monday at Tokensmart, hosting Farhouse Fridays, & Impact Concerts, through bear & bulls🤠

First thoughts idk what they are boosting here yet the algorithm is not good if you don’t follow folxs

Yeah I just need to make almost of my must followsnsonifni clear mine on man at least be quick to connect with like a couple hundo peeps I need to be in touch with!😉
(Maybe I’ll wait till Monday and get some feedback…)🤔

I have a list hopefully from @compusophy that I exported before this test. I will be cast documenting what I see.