The AI Era and the Unreal Marketplace

The question

Generative AI arrived in force at the end of 2022, and the obvious assumption is that it flooded asset stores with cheap, machine-made content. We can test a version of that here. Plotting the major OpenAI and Anthropic model releases against our own continuously compiled picture of the Unreal asset marketplace, three honest findings emerge — and the headline one is the opposite of the assumption.

A caveat we lead with rather than bury: we can measure what AI did to supply — how much gets made, by whom — in detail through May 2026. We cannot measure what it did to demand: our buyer-activity coverage is complete only through September 2024, so nothing here tells you whether buyers actually *reward* AI-made or AI-themed assets. Where that matters, we say so.

Supply was already exploding — before AI

The catalogue grew from a few thousand listings to roughly 79,792 — about 7.3 times its 2020 size. But trace the curve and the inconvenient truth is plain: the steep climb was already underway well before ChatGPT's dashed line. The marketplace was on an exponential long before a single LLM shipped.

So the simplest story — *AI caused the flood* — doesn't survive contact with the data. New releases did keep accelerating through the AI years, but they were accelerating anyway. One genuine confounder sharpens the point: the period also saw the Unreal asset store fold into a single, larger marketplace, which lifts listing counts for reasons that have nothing to do with AI. Marking the models against the curve is useful precisely because it shows what *didn't* obviously bend.

A niche that didn't exist before ChatGPT

There is one thing that is unmistakably new. Listings that put generative AI right in their name — ChatGPT, GPT, OpenAI, Claude, an LLM, a diffusion tool — went from essentially zero to about 29, and almost all of it appears *after* ChatGPT's release. These are the in-engine bridges to the new models: dialogue plugins, prompt nodes, texture generators, local-LLM integrations.

It is a real birth, but keep it in proportion: roughly 29 listings against a catalogue of 79,792 is a rounding error in raw supply. Generative AI created a distinct new shelf; it did not remake the store. (This counts what names itself — a precise but conservative floor. We deliberately ignore description text, which is now full of “no AI was used” disclaimers that would muddy any count.)

The signal that genuinely bent: output per seller

Here is the finding that does hold up. For years, the number of listings an active seller shipped annually sat flat at around 3.4. From 2023 — the first full generative-AI year — it climbs sharply, reaching 5.4 in 2025: a lift of about 60%. Sellers who were already here started shipping markedly more.

That is exactly the fingerprint you'd expect from AI as a *production* tool rather than a product: faster texturing, faster code, faster iteration, more packs out of the same studio per year. It is the most AI-shaped pattern in the whole dataset. But correlation is not proof — the same window contains the marketplace consolidation noted above and the general maturing of a gold-rush category, and we can't disentangle them cleanly. We report the inflection; we don't claim a single cause.

More makers than ever

The doors are also busier. First-time sellers — measured by the year each account's earliest listing appears — keep setting records: 2,007 new sellers arrived in 2025, more than double the 966 of 2022. Whatever the mix of causes, the barrier to *starting* has kept falling, and the AI years are the steepest stretch of that climb.

A lower barrier to entry is the other half of the productivity story: more people making more things, each shipping faster than the cohort before them.

But most sellers still ship once

For all the talk of effortless mass production, the shape of the seller base is stubbornly human. The median seller has published just 2 listings, and about 43% have shipped exactly one and stopped. A long tail of prolific studios pulls the average up, but the typical account is a one-or-two-product hobbyist, not an AI content farm.

If generative AI were turning everyone into a factory, this distribution would be visibly fatter on the right. It isn't, much. The productivity gain is real but concentrated among the sellers already committed enough to keep going.

What we can — and can't — say

Pulling the threads together honestly: supply shows a clear AI-era fingerprint in *how much each seller ships*, a genuinely new but tiny generative-AI niche, and a record influx of new makers — all against a catalogue that was already growing exponentially before any of it. What we are not able to say is whether any of this is working commercially. Our buyer-demand coverage is frozen at September 2024, so we cannot see whether buyers prefer, tolerate or punish AI-made and AI-themed assets, and we will not guess. There are no unit-sales figures anywhere in this data; we never treat a price as revenue.

The measured conclusion: the generative-AI wave did not flood the Unreal store, and it did not invent its growth. What it most plausibly did was make the people already building build faster — and lower the bar for new ones to start. Whether that surplus finds buyers is the next question, and one the data can't yet answer.

Major AI model releases, against the marketplace

Catalogue size and per-seller output are from the MythicLemon Marketplace Index, as of May 2026. Model release months are the widely-reported public dates. The pairing is for context — it is not a claim of cause.

FAQ