Seeding Creation

The views come from the author's own discussion with an AI; the write-up is AI-assisted.

Working on creator ecosystems, I say “inspire creation” almost every day. Say it often enough and you notice how vague it is. What makes a piece of content “inspiring”? What actually happens between a user watching a video and making one of their own?

A few mornings ago it occurred to me: this is just zhongcao.

Zhongcao (种草, literally “planting grass”) is the Chinese commerce term for seeding desire: content plants a little want in someone’s mind, and over time it grows into a purchase. Its partner is bacao (拔草, “pulling the grass”): the moment that want is acted on and the purchase is made.

With creation, what gets planted isn’t a product. It’s a thought: “I want to make one too.”

The more I followed that thought, the less it felt like a metaphor.

The same kind of growth problem#

Here is my view: getting users to create is the same kind of growth problem as seeding and closing a sale in commerce.

Both try to do the same thing. A user is scrolling, and the platform wants them, in the middle of scrolling, to take a more costly action. Commerce wants them to spend money on something; creation wants them to spend time making something. And in both cases, what drives that action is mostly what they see while scrolling.

Put the two side by side and the structure is almost identical:

One thing in this table is worth calling out: growth on both sides walks on two legs, content and initiatives. Content does the everyday, continuous seeding. Initiatives create concentrated moments that give users a reason to act now. Commerce has Singles’ Day; creation has trends and challenges.

If this is the same problem, then the funnels, playbooks and measurement methods commerce has validated over the years, along with the mistakes it has made, are all worth borrowing.

This class of problem has more members than these two: following someone, paying for a subscription, sending a gift in a live stream are all “moving from consuming to acting”. Commerce just happens to have studied it most thoroughly, and creation happens to be what I work on every day.

One funnel: seeding, closing, and self-sustaining#

Chinese platforms have turned seeding into a fine-grained craft. Ocean Engine’s O-5A model splits users into five layers: Aware, Appeal, Ask, Act and Advocate.¹ What I find most notable is how the third layer, Ask, is defined: not by how much someone has watched, but by proactive, high-intent actions such as searching and adding to cart.

The insight behind it is simple: seeing is not wanting, and wanting is not doing. There is a hurdle between every pair of layers.

Mapped to creation:

I use Ask, the moment a user is already tempted, as the dividing line. Seeding brings people to that moment and answers “seed whom, with what?”. Closing takes them from tempted to published and answers “how do we follow up on that want?”. Self-sustaining is the last layer: creators seeding more people in turn.

Looking at this table, my strongest impression is that in creation, most of the effort goes into the first two layers: getting users to see more inspiring content. Yet a large number of users are already stuck at the third. They saved a song, opened a template, maybe even shot half of something, and then nothing. In commerce, users with something sitting in their cart are the most valuable users there are. In creation, this group is rarely singled out.

Seeding: making people want to create#

I think seeding breaks down into six pieces of work:

Find content that seeds. Content that makes people want to create is often not the same content that makes people want to keep scrolling. Tutorials, fresh formats and creations ordinary people can reproduce are not necessarily the most “watchable”, but they are the most likely to get someone started. Timeliness matters too: a format that is just emerging inspires far more than one that has been done to death.

Seed the right people. The worst way to seed is to sprinkle it everywhere. The same piece of content does very different things when shown to someone who already creates in that niche versus someone who only watches.

Pick the right seed sources. Brand seeding works as a pyramid: top KOLs create buzz, while mid-tier creators and many ordinary users (KOCs) drive conversion. An early large-scale experiment at Facebook found that while a single strong tie is more influential, most new information actually spreads through the sheer number of weak ties.² This is even more true for creation. When a creator with a million followers makes something, ordinary viewers think “that’s their thing”. When someone like them makes it, they think “I could do that.”

Product initiatives. Commerce concentrates its bursts in sales events; creation uses trends and challenges to give people a reason to create now. The recommender’s job here is to deliver the initiative to the right people: a beauty creator won’t care about a basketball challenge, and an initiative pushed to the wrong people is just noise, however hot it is. Trend stage matters as well: early on, show it to people who like to try new things; later, show it to people who need something to copy.

Lower the barrier. Free samples are one of the oldest tools in marketing. Research found that the value of sampling is not only that people who would have bought anyway buy sooner; more importantly, it creates expansion: people who would never have tried the product try it once.³ Creation’s free samples are templates, the same song, pre-cut beats. Their biggest value is not serving existing creators, but bringing in people who have never made anything. And once someone is tempted, take them straight into the editor. Every extra page in between loses another group of people.

Social proof. A well-known experiment with an artificial music market found that letting people see others’ choices significantly changed their own.⁴ At the moment a user is hesitating, telling them how many people, or which of their friends, have already made one often works better than showing them ten more example videos.

How do you know the seed took? RED and Douyin often measure seeding by “search after viewing”: did the user go search for the thing after seeing the content? Creation can have a counterpart, “create after viewing”: did they save the song, open the template, start creating? But these attribution numbers are easy to overestimate. One study compared approaches directly and found that observational attribution methods often fail to recover the true ad effects measured in randomized experiments.⁵ To know whether seeding really works, you ultimately need A/B tests.

Closing: helping them finish#

After seeding comes closing. In commerce, closing is a whole set of infrastructure: carts, coupons, one-tap checkout, price-drop alerts. Platforms put enormous effort into making sure people who are already tempted can pay without friction.

In creation, closing means helping someone who is already tempted, or has already started, finish the piece. Almost no one works on this seriously today. I think it breaks down into five pieces of work:

Identify the tempted. Closing starts with finding these people. It can go one step finer: group them by the creation entry point they used most recently. People who came in through a music page, through a template, or by opening the camera themselves need different kinds of help.

Repeated same-interest exposure. Once a user shows intent, it is worth continuing to show them content from the same interest over their next few sessions. But commerce has hit a pitfall here: research found that repeatedly showing people the same product they had shown interest in was, on average, less effective than generic ads.⁶ So the point of repeated exposure is “the same interest, different content”, and frequency should be restrained. Seeing the same creation opportunity two or three times a day feels like seeding; seeing it ten times feels like nagging.

Broad to narrow over the lifecycle. To me this is the most interesting part of closing. Someone needs completely different things at different stages of creating, so the scope of recommendations should narrow as they go. Take someone who wants to film themselves making pour-over coffee at home:

This mirrors commerce too: show categories while someone is browsing, specific models once they know what they want, and reviews and comparisons just before they buy.

Advance, don’t imitate. Behind broad-to-narrow is a change in what the recommender is trying to do. Normally a recommender asks “what will they probably watch next?”, which imitates their past behavior. For someone in the middle of creating, the better question is “what can I show them that moves this piece one step forward?”. The first may show them things they’d drift off-topic into; the second shows them things that send them back to the draft to keep going.

Reduce friction, remind at the right time. Cart reminders work because they remind users of something they themselves wanted. Creation is the same: when a draft has been sitting for a few days, rather than a generic “come create something!”, show them a great piece related to their draft and let it bring them back naturally. An unwanted reminder is spam; a reminder about what they are working on is help.

How do you know the deal closed? Closing is measured by draft completion rate and time from start to publish. Completion rate matters more: most drafts don’t fail because they are slow, they fail because they are abandoned.

After creating: distribution is social recognition#

Everything above, seeding and closing, happens before someone creates, and that is also where most of our current work sits. But that in no way means what happens afterwards matters less. Quite the opposite: from the data I have seen, distribution after posting is the more important incentive for creation. What really decides whether someone will make another one, and even whether seeding works at all, is often whether anyone sees the piece once it is out.

So I think distribution after creation is the strongest incentive seeding has. At its core, it is a form of social recognition.

If I had to compare it to commerce, it resembles two things. One is the recognition a buyer gets after the purchase: you buy a new jacket, wear it out, and friends tell you it looks great. That compliment is what makes the purchase feel worth it, and what makes you want to buy again and recommend it to others. The other is reviews and reputation: buyers’ reviews slowly add up to a seller’s reputation, which in turn shapes whether the next group of people buys.

In creation, that compliment is the views, likes, comments and follows that distribution brings.

This recognition doesn’t only act after creating; it also reaches back to before. Many people are stuck at “tempted but haven’t posted” not because they can’t make it, but because they’re unsure: “Will anyone even watch?” When someone gets real feedback on their very first post, they’re braver the next time; and when the people around them see that “someone like me posted, and people watched”, that itself becomes another act of seeding.

Here is one way it differs sharply from commerce. A seller can’t control whether anyone compliments your jacket; but whether anyone sees your creation is largely decided by the platform’s distribution. In creation, the switch for social recognition is in the recommender’s hands. That is the platform’s biggest lever, and also a responsibility.

So for a recommender, cold-start distribution of new creators’ work is not only a content distribution problem; it is the last link in the whole seeding chain. A few things I think matter most:

• Find the right first audience. A hundred views from people genuinely interested in the topic produce very different feedback from a hundred views from people who happened to scroll past. The quality of recognition matters more than its quantity.
• Make feedback arrive fast. The first few hours after posting are when a newcomer cares most, and when they’re most likely to give up.
• Keep recognition real. Artificially inflating newcomers’ numbers looks good in the short term, but it degrades the experience for viewers, and creators eventually notice the recognition is hollow. What distribution can do is make sure good work isn’t buried, not manufacture applause for every piece.

Self-sustaining: creators seeding creators#

Back to the last layer of the funnel: Advocate. In commerce, advocacy is repeat purchase and word of mouth. In creation, it is someone who keeps creating, and whose work gets more people to create.

In my post on the creator flywheel, I drew a loop: consumers become creators, creators produce content, and content flows back to consumers. The part of that loop most easily overlooked is the “inspire” path: creation itself seeds creation.

Whether an ecosystem can turn on its own depends largely on this path. If every new creator has to be inspired by the platform, the ecosystem will always need the platform to push it. Only when a creator’s work reliably reaches people who are similar to them and on the fence, and gets them to start too, does the ecosystem begin to sustain itself. So distributing creators’ work should also be treated as seeding: show it not only to people who will watch it, but to people who will create because of it.

The social recognition in the previous section makes a creator want to keep going; the loop in this one lets their work spark more people. Together, they are commerce’s repeat purchase plus word of mouth.

Where it differs from commerce#

Calling them the same kind of problem doesn’t mean they are identical. There are at least three differences.

First, the goal diverges from consumption value. In commerce, products that sell well are often also things users like to look at. In creation, that’s not the case: content that makes people want to create and content that makes people want to keep scrolling are often not the same. Distribute purely on consumption value, and the content most likely to inspire creation gets buried. So creation needs its own objective; you can’t expect consumption-side recommendation to take care of it along the way.

Second, there is a tradeoff between creating and consuming. When users go off to make a video, they scroll a little less. In the long run, more creators mean more good content, and the trade is very likely worth it; but it should be seen clearly and chosen deliberately, not ignored.

Third, too much seeding makes content homogeneous. The more of a product you sell, the better. But when the same song or the same template is copied by a million people, the platform is left with only one voice. Imitation is how many people take their first step, and that is fine; but the end point of seeding should be more people starting to express themselves, not more people repeating the same thing.

Finally#

Commerce seeds desire, and the end point is a purchase.

Seeding creation, the end point is one more creator.