The avatar lottery is a confession

The numbers are real, and honest

A company running AI creators at scale published its own data recently, and credit where it is due: they showed the whole distribution instead of the highlight reel. 110 synthetic accounts on TikTok. One earned 17% of 414 million views. The top three earned 42%. The bottom half combined earned two tenths of one percent. Two thirds never produced a single video that cleared 100k.

Their read: you cannot know which avatar will hit, so audition each one across about twenty posts, cut the cold ones, and pour volume into the winners. Generation is cheap, so twenty throwaway posts is a fair price for the information.

It is a clean playbook. It is also a precise description of one specific problem, and it happens to be the problem Favente is built to not have.

A lottery is what you run when the ticket is worthless

You only reach for a lottery when every entry is disposable and interchangeable. That is the tell. An AI avatar has no history with the product, no reason to post past the prompt, and nothing at stake when the video flops. So you cannot predict which one lands, because there is nothing to predict from. Affinity, taste, the itch to say something true about a thing you actually use: none of it was ever in the pipe. What is left is dice, and a miss rate polished into a strategy.

Two thirds cold is not bad luck. It is the expected yield of starting from zero signal and manufacturing volume to paper over it.

Real users do not need an audition

Now start from the other end of the table. Your product has people who open it every morning without being paid to. Some already post about it on their own accounts, for free, today. A few have written you paragraph-long reviews, referred their friends, defended you in a thread. That behavior is the exact signal the avatar lottery has to buy twenty posts at a time to fake.

A user who already loves the thing and already creates in public is not a ticket. They are the one entry the lottery is quietly praying to draw, except they are sitting in your user table: reachable, warm, and pre-qualified by the single quality no prompt can generate. They care.

We agree that most people will not. That is what the filter is for

Here is the concession the honest version of this argument owes you. The power law is real. Most people, your users included, will never become creators, and of the ones who try, most will not hit. Pretend otherwise and you end up with 110 accounts and a spreadsheet explaining the churn.

So the disagreement was never whether the few carry the many. It is how you find the few. The avatar lottery finds them by brute force, minting and burning accounts until a winner falls out. Favente finds them by filtering for the two things that predict a real creator before a single frame is shot: affinity, they already use and love it, and revealed effort, they already post and review and refer in public. Rank people on that and you are not auditioning strangers. You are inviting the ones whose track record already answered the question.

A small deliberate gate finishes the job. Applying costs a real minute of effort, on purpose, because the effort is the filter. The users who would never make a second post are the ones who will not clear it. That is not friction to remove. That is the point.

What happens after the miss is the other half of the tell

In the avatar model, a cold account gets deprioritized and a fresh one gets spun up. Nobody learns anything, because there was nobody there to coach. The creator was a config value.

When the creator is a person, the miss is a starting point. Favente coaches the first video, edits it, and checks it against a Hook, Proof, CTA structure before it ever posts. A real user who almost hit can be taught to hit next time. An avatar that almost hit is just noise you already paid for.

That is the compounding the lottery structurally cannot reach. A coached creator gets better and stays. A burned avatar is gone, and its twenty posts went with it.

What you actually keep

Run the lottery and at the quarter's end you hold a pile of views and a roster of synthetic accounts you will cycle again next quarter. Run the other play and you keep something that appreciates. A bench of real customers who now make content. A record of which of your users convert when they create. And the piece most brands overlook: the exact language your best users reach for when they describe you, ranked by what actually worked. Your own audience's words, sorted by performance, outlast any view count.

Where to start

You do not have to buy the whole argument to test it. Pull the five users who already post about you. They are making content for you for free right now, carrying the affinity no prompt can fake. Brief them, coach the first video, pay on the result you approve, and inside a week you will know which of your own users become creators.

No avatars to mint. No twenty-post tax. You are not buying signal by the batch. You started the whole thing already holding it.