Demand Discovery under Free Imitation: Generative AI and Creation Incentives in a Digital Doujin Platform

Abstract

Sales rankings on digital platforms broadcast which new ideas sell. The demand knowledge a pioneer discovers therefore spills over to every potential imitator, and generative AI has made acting on it drastically cheaper. On DLsite, the largest Japanese marketplace for digital doujin works, I measure how fast imitation dilutes the rent to discovering a new niche and how much AI lowered the fixed cost of creation. A policy sequence turned AI supply off and on (disclosure mandate, sales ban, reopening under caps), and a suspension of the major credit-card brands moved market size separately. From these events I estimate a model of demand and entry in which creators choose human or AI production under technology-specific fixed costs. Most newly discovered niches are imitated, and the lead time that protects a discoverer is eroding. The estimated median falls from about nineteen months before generative AI to eleven months with AI works back on sale. Removing lead time altogether would cost the median imitated pioneer about 11,000 yen, half of its six-month receipts. Half of discoveries meet no imitator within a year, so the removal would push only about seven percent of all discoveries below break-even. A six-month exclusivity window would raise that pioneer’s receipts by roughly 20,000 to 80,000 yen and lift about five percent of discoveries back above break-even. The median human fixed cost is roughly ten times the AI cost, though the human estimate carries a wide band. The AI cost itself is falling, down about 35 percent from its 2022–23 level. Buyers discount AI content heavily. Removing AI works would cost them a few yen per purchase, while removing the discount would raise the AI share of purchases from about 6 to 79 percent, so AI supply is held back by the buyers’ discount rather than by the cost of creation.