What should Tumblr do

I think rumors of Twitter’s demise (with Elon’s takeover) are greatly exaggerated. Yet, it has sparked a renewed interest in social products and that is exciting.

Tumblr was sold for $1B to Marisa Mayer’s Yahoo in 2013 and bought for $3M by the owner of Wordpress in 2019 (yes, those numbers are correct), and its time may have come … again. Here are two hard-to-ignore signs that Tumblr is teeming with life:

First: Tumblr users share a secret code to find other users in real life:

This tweet thread explains it. The last time I saw something like this was when TikTok was spreading like wildfire: TikTok users would shout out a line from a trending song in stores and see if anyone else would reply back with the next line.

The second sign that Tumblr is in its Cambrian Period is even whackier and is at the end of this post — scroll down!

Twitter always captured an inordinate amount of mainstream attention — undisputably greater than the value it captured [exhibit A][exhibit B]. Perhaps that attention was disproportionately greater than even the role it plays in popular and internet culture. Don’t get me wrong — it is extremely important, especially in political and current affairs discourse around the world (and why mainstream media pays it so much heed). But culture begins on the fringes. Reddit and Tumblr always seemed upstream of Twitter to me [pmarca on the subject].

We seem to be at an upwards inflection point in Social-with-a-capital-ess again at the end of 2022. There is an eagerness to venture forth and try out new social apps. What should Tumblr do to take advantage of the dynamics today, and inflect its own growth?

  1. Own your weird. Make it easy to “read the room”.
  2. Feed the machine. The algorithm is downstream of the product experience.
  3. Reward creation. Everything is downstream of creation.
  4. Ignore the graph. For now.
  5. A reality check on advertising. Find Ikigai in monetization.

Having worked on online media for a decade, a majority of which was on social products, including the very core of Twitter (the algorithmic timeline and discovery products), these are purely opinions and at best hypotheses: no one knows anything when it comes to consumer products; only data is truth.


1/ Own your weird. Make it easy to read the room.

Social products are like different rooms at a party: while individuals diffuse from one room to another, the tone of the conversation lingers on and is reinforced (or evolved) by newer participants. Great hosts actively design different rooms for different elements of the party. Tumblr has its own thing going — its users are on Snap and Reddit too, but they use Tumblr to share edgy, fun content they find online. For instance, reblogs are more important than likes here. Tumblr should lean into this.

Human beings are social beings. We are generally good at reading the room, whether to fit in or to stand out. But this is much harder to do online with many of the IRL cues missing.

As Tumblr grows, make it easy for new users to read the room. Explicitly call out rules of the platform (simple, but proven to be effective). Highlight content you want to reward — monkey see, monkey do. People will post the kind of content they see gaining acceptance (why you see so much thought leadership on LinkedIn and shitposting on Twitter). The look-and-feel can also reinforce the brand: for Tumblr, a degree of edginess.

It is hard to actively change the tone of the content and conversation on a social network, but easy to lose control over it. I believe most of the problems social products grapple with in this respect stems from it being hard for users to read the room (whether in sub-spaces on the same product or on the entire product).

2/ Feed the machine. The algorithm is downstream of the product experience.

By now, it is a given that your feed algorithm (yes, “feed the machine” is a pun) will drive most of the consumption in a social media app. However, the product experience drives more of what the algorithm will be capable of than the other way round.

The algorithm needs user input to learn and personalize content and learn again and adjust. User feedback is extremely extremely hard to get because most of us are lazy and will not lift a finger unless it provides a dopamine hit. Likes or shares on posts are hooks into the dopamine faucet. A “I don’t like this post, show me fewer of these” feedback button will barely get any clicks — it is easier to simply scroll away.

Rather than designing your algorithm around what parts of your product see usage, figure out what your algorithm needs to build the most amazing, kickass feed. Then design your product to feed into that algorithm a firehose those signals from users.

TikTok is a great example: if you pause and watch a video, that is an endorsement. If you skip to the next video, that is a negative signal. [Eugene Wei’s seminal post Seeing Like An Algorithm] Other social feeds will have an [X] button to allow users to indicate when they don’t like a piece of content. For the most part no one clicks on it, and this sparse signal is not enough to train a negative engagement model. This hampers how fast your algorithm can learn or how bold it can be in active learning.

There are countless other examples of this. In the early days of ranking Twitter’s timeline, we were severely limited in active learning because users expected the timeline to only (later, mainly) show content from accounts they followed. The product limitation hampered the algorithm.

Tumblr is an existing product, which may be hard to change wholesale (or perhaps not?), but there are a number of easy things to try: for example, gently snap from one post to the next while advancing the feed instead of a smooth continuous scroll to get the TikTok kind of negative feedback. It all depends on the “room” they want to build, the consumption they want to enable, and the content they want to reward. I’d encourage them to think of multiple rooms — most social apps have a giant feed where everything collapses and that only gets noisier with scale. Instagram is perhaps the only one to have created two, may be three, different rooms in their app.

3/ Reward creation. Everything is downstream of creation.

If people are not adding great content to your social platform, even the best feed algorithm will creak to a halt. There was a time at Twitter where early version of the feed algorithm had started growing usage again, but the number of users tweeting was still declining. There are many external and product mechanisms to grow this, but most surprisingly, consumption choices can reward and incentivize creation. Seeing good content made users more likely to create content. Finding friends made them more likely to create (for them). Giving newbies a leg up (like a golf handicap) by boosting their tweets and getting them engagement made them create more.

Social products are famously status-as-a-service. People create to climb the status ladder, or they create if they can capture value elsewhere — e.g. thread-bois on Twitter will shill their paid Substack subscriptions. I believe we are in the early innings of social platforms helping creators monetize directly. YouTube, of course, is the major game in town here, but with everyone trying elements of paid subscriptions or one-off tipping or web3 angles (NFT sales), this is a rich, mostly blank canvas.

In this, Tumblr is not behind the others. The ingredient features for direct monetization already exist in the app. But it is first and foremost a cold start problem, where simply building the product isn’t enough. TikTok in its early days of bootstrapping in a new market would pay creators large sums of money to make content for their app. What Tumblr needs to do is find the audience willing to subscripe or tip, and contract with creators to “supply” those communities in exchange for real $. They will be the marquee creators to trailblaze and others will follow their path. The World Cup (may be a bit too late though), Pokemon, Cooking and recipes (perhaps the IG influencers can reuse their content in exchange for tips or subscription revenue on Tumblr) seem at a cursory glance at their Trending page to be viable candidates.

4/ Ignore the graph. For now.

In an age of algorithm driven social networks, does the network even matter? TikTok has clearly proven it does not. But networks are valuable seeds and backstops.

If you expect creation on your platform to become “professionalized” through creators or influencers or what have you, a unidirectional graph (”follow”) is the right model. If you expect to have more of a peer-based relationship (like Whatsapp or early Facebook), bidirectional is better (”friend”). Tumblr is the former, but less in an Instagram influencer sort of way. While the unidirectional graph it has definitely suits it better, these are harder to bootstrap and it is best to start letting users find their friends and follow each other.

An explicit network also allows for new spaces and / or mechanics — direct messaging, friends-only interactions. It is an important signal to bootstrap the algorithm. However, in an algorithmic world, the network will be a slow build, and will be smaller (if higher signal) as users have less need to find and follow other users. It is more important to figure out the algorithm first.

If Tumblr is seeing old users returning, it might even worth asking them to reset their account to a new state and start over.

5/ Reality check on advertising. Find ikigai in monetization.

Yes, 3rd party ad networks abound, making Tumblr feel like yet another spammy content farm, and likely turns away new or light users. Without significant usage and not the resources to build and sell your own native ads product, that tradeoff is what I would make too.

Here is the problem with advertising: without great first-party data, you cannot build a direct response ads business. Without high reach, you cannot build a great brand advertising business. If you have a legacy ads product without either of these, it is momentum and relationships carrying the business and bringing you trial budgets (what the industry likes to call “social budgets”). In a recession these budgets will dry up, as we might see next year. I’m curious to see what the other mid-tier social platforms will do. 3rd party ad networks are a terrible tradeoff but may be the only option.

Charging to remove ads are not a good idea either: the uptake on it will be small (only the heaviest users might find it worth to pay for removing annoyance), and it caps the revenue you can make per user (most ad revenue is earned from the heaviest users and you just capped it at $5 per month). It will most certainly hurt total revenue.

Finding ways to monetize directly and leaning in on “whales” may be the only path to ikigai, if a product can have ikigai. Outside of this, getting smart about showing 3rd party ads can have a lot of juice: Amazon used to show Google and Yahoo sponsored links on its pages for years while it built out its own ad products. This provided disproportionate revenue for the number of people needed to run it, and more than funded the rest of the early ads team to build their own ads products.


What really matters

Much of the above are tactics, some large, some small. What really matters is identifying and leaning into your identity as a social platform.

Coolness is decided by other people and it comes and it goes; every social product is cool for a short while. But you pick your own weirdness and it is forever. People will adore you for it. Some might hate you for it too, but ignore them.

Tumblr skews younger, edgier, on the shores of pop culture. It should not try to become Twitter — political discourse, heated moral mudslinging and overall outrage are not its character, neither is the business / technology thought leadership. Although there are quirky, loveable parts of Twitter that would fit right in.

What is key is that the world is open to new social products right now. Tumblr can open up its weirdness to more people.

For example, did you know that one of the trending topics on Tumblr right now is Martin Scorsese? Do you know why? Because users have decided to, en masse, make fandom canon for a 1970s Scorsese film that … wait for it … ACTUALLY DOES NOT EXIST! This is amazing, authentic, worth taking to the rest of the world.