Medium is Broken

Like a crooked poker game. So let’s fix it.

A.H. Chu

Published in

Quality Works

11 min readJan 8, 2016

I am not a good poker player.

However, by chance over the holidays, I was lucky enough to win a casual match between family and friends. Unbeknownst to me, the strokes of luck I benefited from were unprecedented according to my family.

For example, in one sitting, I managed to draw a straight flush (near royal), a flush on the river to beat pocket aces and ended the game with three jacks (with two jacks being drawn on the turn and the river, respectively). My more experienced competitors were flabbergasted. I was officially deemed “the best worst poker player ever.”

Beginner’s luck to be sure.

But something occurred to me as I was playing. That poker at its core is a bit mean-spirited. Specifically, as my chip stack grew, I was encouraged to bully my opponents out of the match.

“That’s the right play,” my relatives would say as I forced others to make do or die decisions turn after turn, slowly strangling the life out of their ever-diminishing stacks.

As a game, poker clearly rewards those who hold a dominant advantage in resources over others. 99% of the time, it is a case of the rich get richer.

And so is Medium.

There’s a dialogue emerging on Medium. Kicked off by Kamil Stanuch’s piece on plagiarism, and then clarified by Ben Belser, something is becoming clear.

Medium is broken.

How so?

Well, like the poker match I played, a few players (i.e. writers) in this game have accrued dominant chip stacks (i.e. followers), creating a very similar dynamic of the rich get richer.

As user Angelique commented on Ben’s piece:

Ironic that I’m reading your article on Medium, because the kind of empty stuff you’re talking about is just about all that Medium is willing to show me. I have no idea how I finally got to see your article (four days late.) There is NO way to fill our Medium newsfeeds with posts by other Medium users that we have deliberately chosen to follow.

Compare this to the founding principles of Medium as described by founder Ev Williams in his “Welcome to Medium” mission statement:

Medium is not about who you are or whom you know, but about what you have to say.

Implicit in this statement is that Medium’s mission is to highlight ideas and writing based on the quality of (sorry Ben) its content.

But this is not what is happening.

To illustrate, I am going to take a real-time snapshot of the “top 3 stories” in my medium feed below.

What do these stories have in common (aside from the same author for two of them)? The two authors in question have collectively 78.4 thousand followers between them.

In this case, is it any wonder that they are accruing more views and recommends than other would-be writers?

Is the quality of their writing really that good? Or are they simply playing by the rules that we have all implicitly realized at this point? Have they effectively “hacked” the system so well that that in itself is now the driver of value in their writing?

More views begets more views. More followers creates more followers.

Without directly addressing this vicious cycle, Medium will continue down its current trajectory where quality is determined largely by the summation of views, recommends and comments which is undoubtedly correlated with the number of followers a writer has. Thus, quality, as manifested in the feed that Medium is willing to show you, actually is about whom you are and whom you know, NOT about what you have to say.

So how do we fix this?

I don’t have the silver bullet but here are some ideas:

1. Kill the Recommend Button

This seems like a drastic measure, but think about it. How often do you skim your news feed and either consciously or subconsciously equate the number of recommends an article has with the quality of its content?

I try not to do this but even I instinctively discount articles that have low numbers of recommends.

Instead, I believe Medium would be best served by removing visible recommends altogether. Or if it decides to keep it, to qualify recommends by the number of followers a writer has.

Something like:

# of Recommends / # of Followers = Medium Quotient.

So if Mr. Hardy gets 1k recommends on an article relative to his 18.4k followers, then he gets a Medium quotient of 1k / 18.4k = .05

Compare this to another article that popped up from Jenna Elfman “The Unthinkable is Happening in our ‘Backyard’” which has 204 recommends vs. Jenna's 74 followers. A phenomenal quotient of 204 / 74 = 2.75

Again, not a perfect answer, but at least in this case we don’t get an arms race where, rather than just writing good stuff, even well-meaning writers like Tim Cigelske are incented to write stuff that gets attention and can bolster his followers.

2. Highlight “Dialogues” rather than “Comments”

I am a firm believer that quality is not an intrinsic thing but one that exists within the relationship between two things.

By basing a measure of quality on static “snap-shot” metrics like recommends or even # of comments, Medium will never transcend the world of 140 character posts like it wanted to since inception.

So what may be a more accurate measure of the quality of the relationship between writers and readers?

Comments are a good start, at least this shows engagement based on the reader. However, even comments alone I believe are insufficient. This is because it is largely a measure of one-way engagement.

It’s a reader in effect “paying” mindshare to the writer. In this paradigm, the writer is then incented to “extract” mindshare from its readers regardless of whether a true relationship was formed.

This is the impetus for endless listicles and the like. Certain writers don’t really care about establishing a relationship with their readers. They just want to add to their chip stack.

Case in point? Remember that top article “How to Become the Best” by Benjamin Hardy? It has an admirable 44 comments.

Curiously, how many of those comments do you think Ben has responded to in turn?

Zero.

Guess he is too busy trying to figure out how to become the best at something besides engaging his 18.4k readers.

Now imagine instead of tracking comments, we tracked “Dialogues” whereby a comment was posted to an article and the writer actually responded. What a concept huh?

In this case, we not only get a measure of reader engagement but we also get a measure of writer engagement with their readers.

Again, while not perfect, I believe this is a much more accurate measure of the quality of the relationship between readers and writers. It also challenges readers to not only “farm” for likes and comments but to actually think critically about the responses and feedback they receive.

The Future and Fate of Medium

So look, I know there is a Medium meme going around that everything is either inspiration porn, a rant or a listicle.

Although this is closer to a rant, please don’t take it as such. I am writing this because I think Medium is really close to being something transformative.

I think Medium really can reach its original vision of creating a more informed citizenry and a deeper depth of understanding. I truly believe in the importance of this vision.

However, for it to truly transcend the “echo chamber,” its definition of quality needs to transcend the current paradigm of “likes”.

If it fails, it will never be more than a 140-character+ Twitter with awesome formatting.

If you like this article please consider commenting instead of recommending.

If you leave a comment, I promise to respond so we can kick off a dialogue, the way Medium should work. If Medium can’t fix itself algorithmically, let’s do it organically.

Medium is about the dialogue not the diatribe. Please join me in making it one!

PS — thanks to those who have left comments thus far, we have already come up with some other noteworthy ideas. I’m going to try to keep a running log of these below for easier reading and brainstorming!

PPS — will try to start organizing now as well into broad categories.

PPPS — Life imitating Art? Just got this a few min ago. Weird!

Quality Metrics

Andy Burke suggests consider a composite metric which includes recs/views as well as recs/followers
Charles Frank posits that stats and recommends also takes into consideration what people (and how often I presume) highlight —
(incidentally Greg Gueldner responded and said this, in fact, is one of the tweaks they are currently looking into.)
john seeker suggests a grading system from 0 to 5 for both articles and comments to avoid the oversimplified binary nature of the current “like” system.
Nenad Ristic has a critique of the MQ idea in that he wonders if this will unfairly advantage new users (for example someone with only 1 follower). Does it swing the pendulum too far the other direction?
Myddera has two suggestions on quality metrics, one is a visual color-coded indication for users’ “Medium Quotient” from say green to red. So a new author who has low view counts but exhibits a strong MQ could have a green color code. Likewise for users who have both high views and still maintain strong MQ’s. Myddera’s other idea is to highlight which commenters are “within” ones network vs “without”.
Steve Moraco also suggests a re-weighting of the “quality metric” but takes into account several more variables as follows: read time x likes x x shares x author replies / total comments x followers = “Advanced MQ”. This metric would encourage authors to write back and also gradually cull articles that have a lot of comments simply due to a large follower distribution list without much actual two-way engagement.
“Click-through” or “Follow-through” metric: I have another thought on a metric of relevance. Taking into account whether a reader clicked through to read any of the author’s other works. This is a pretty sure fire sign that the story was relevant in my mind. To facilitate this, Medium should consider posting the Most Popular articles by an author below their story rather than or in addition to the most popular articles with that tag like currently below:

Dheeraj Dhobley has some very thoughtful suggestions. On quality metrics, he posits that there is a certain vicious cycle that takes place along the lines of “garbage in garbage out”. If he recommended stories are not relevant, than any recommended story that gets propagated to that users’ followers is potentially even less relevant. Thus, a randomized article with no connection to the user may actually be more relevant than one recommended by a followed user. Steve Baker also adds his vote to randomization as a possible idea.
Both janetgas and Steve Baker have an interesting thought which was to bifurcate the current recommend button into more specific callouts. Steve suggests one designed moreso for comments (an “agree” or “thank you” button) and one for actual stories (a true “recommend” button). Janet has a similar thoughts on an “agree” or “thank you” button and adds an “intriguing” button to highlight stuff that stands out as exceptional work.

Search and Discovery

Joskua would like a full searchable tag index instead of just featured tags (for example, being able to search related tags that may show Hispanic content) — from Joskua (I believe the home page does show your subscribed tags below the featured tags but there is no way to visualize the full “tree” of tags which would be really interesting to see)
Anna Schulten suggests simple “no-fuss” digest that gives here a broad but interesting array of reading. This reminded me of Spotify’s Discover Weekly, an algorithmically generated mixtape that is recreated from scratch every week for every single user on Spotify, that I discussed in a response to Joskua. If Medium could nail a “Medium Weekly” it would be groundbreaking.
Avery Korn Had a great idea to take a page from VSCO which “grid-ifies” user curated photos without applying so much as a heart or thumb or other appendage. Almost a pinterest-ification of Medium if you will. This could help wean us off the current rec system and also democratize the current publication based distribution system. (no publication = no views)
blainbovee has an idea of a potluck dinner digest which essentially takes articles at random and offers them on a less frequent basis. Perhaps lady luck can help us discover some quality writing?
Dheeraj Dhobley has a thought on discovery which is to first include randomized stories in a users feed. Then, if a user recommends a randomized story it gets propagated. However, if a user recommends a non-random story, it gets a tally but does not get propagated. In this way, as kind of a “blind taste test”, one can be more assured that a rec’ed random story is actually more relevant and not due to a “snowball effect” from mass distribution.

News Feed Optimizations

Lisa P. Smith would like an option on the homepage to just see articles from those she has recommended
Hope Atina suggests a re-weighting which does not completely exclude articles that were “similar” to what you liked but reduces the number of these “suggested” articles and clearly highlights them as such.
Nathan Zorndorf has two interesting suggestions, one is to have a tiered or weighted system of recs where we can manually push certain tags to have more presence in our feeds. The other is an inquiry into whether stories rec’ed by Medium are generally trending already and, if so, could they highlight some of the more “off the beaten path” stories that may also be interesting?
dp suggests that more robust filtering in terms of time, # of recs, # of views or specific tags would give users much more control over their feeds.
Kristine Kirby suggested the ability to “blacklist, “Mute” or “block” certain users from showing up in your feed.
Dheeraj Dhobley suggests that the news feed stress topics moreso than timing similar to how Quora deals with “tags” and “topics”. The key is that time-based propagation does not account for the slow process of writing thoughtful answers. By the time your thoughtful response to a “tsunami” article hits the wire, you probably already missed the opportunity to “catch a wave”. Instead, why bother to catch the wave at all if the feed instead properly stored and surfaced relevant answers and responses as well as original writings.

DIY Solutions

Matt Pfeffer has a more practical solution which is to simply prioritize quality over quantity in his own follow feed. Slow going but effective thus far.
Christopher Doiron suggests we suck it up and 1) learn to market ourselves better and 2) learn to use the search button (a la LMGTFY)
Daniel Blanaru and Guy Brodsky both had a similar suggestion which is to have the Medium team hold an open crowdsourcing forum to solicit ideas (kind of like we are doing here but in a more formalized and official fashion).
Dallas Clyde Moore was @Y suggests just going it alone and starting your own publication. Let your own selection of quality writing speak for itself!

Authorship Tools

Daniel Lee offers an interesting insight that content right now goes into a black abyss once you press publish. Which raises the question, perhaps more robust distribution stats in combination with existing readership stats would be helpful?