The $100 sorority girl and the end of buying attention by the view

Olivia Moore, a partner on the investing team at Andreessen Horowitz, spent about 30 minutes a day for a week animating "Janie," an AI-generated 19-year-old University of Alabama student built from a single ChatGPT image using tools including MiniMax and Grok Imagine ^[1][2][3]. By day seven the account had 1,300 followers and was averaging tens of thousands of views per video ^[4]. Forbes puts the total spend at roughly $100 ^[5].

The follower count is not the story. The cost line under it is. Thirty minutes a day across seven days is about three and a half hours of human labour ^[6]. Twenty videos were posted, which works out to roughly $5 of spend per asset ^[7][8]. On a per-follower basis, the account cost under eight cents a head ^[9]. Any marketer who has negotiated a nano-influencer package knows those are not comparable numbers to the ones on their rate card.

What makes this an arbitrage rather than a novelty is the pricing convention it attacks. Influencer budgets are still largely denominated in delivered attention: views, impressions, engagement rate. If engaging attention can be manufactured at $5 an asset, then attention is not the scarce good and cannot hold the price. The scarce good is the thing Janie did not have, which is verifiable human origin. Moore's own framing is that the cost curve on frontier intelligence keeps falling, supply of this content is about to explode, and creators are "about to be outnumbered" ^[10].

The governance data point is the more damaging one for anyone hoping platform machinery will sort this out. TikTok's automated systems eventually applied an "AI-generated" label to 8 of the 20 videos, leaving 60 percent unlabelled, and Moore reports the label had no visible effect on performance or on the comments ^[8][11][12]. A disclosure that carries no distribution penalty is a compliance artefact, not a price signal. Moore also notes that today a viewer can settle the question by zooming in on hands or a dresser, and soon may not be able to settle it at all ^[13].

Two honest caveats. This is a single self-reported experiment by an investor, in one narrow niche, competing against roughly 2,500 real prospective new members posting the same rush content ^[1][14]. And 1,300 followers is not a media business. The argument rides entirely on unit economics, not on Janie's audience.

The interesting wrinkle is that Moore came out ahead by disclosing. The reveal drew a largely positive reaction, with commenters comparing it to Black Mirror and asking her to run it again next year ^[15]. Her conclusion is that hiding the tools makes audiences experience AI as a trick, while showing the work creates a different relationship ^[16]. That is the outline of a market where provenance is a claim you make and can be held to, rather than an assumption baked into a CPM. The risk sits with actors who will not disclose, and the technology is available to them too ^[17].

Watch three things. Whether platform labelling rates climb above the 8-in-20 mark reported here, and whether a label ever starts costing reach ^[8][12]. Whether brand contracts begin carrying explicit human-authorship representations at the deliverable level rather than the campaign level. And whether the roughly $250 billion creator economy starts repricing off provenance, which is also where the Section 230 and platform-policy questions land ^[18][19].