How much of Adobe Stock is AI? We measured 1,333 keywords

74.7%

AI share for the median tracked keyword

82.2%

of tracked keywords are now majority-AI

1,333

keywords with a trustworthy AI-share reading

Everyone selling on stock sites has a feeling about how much generated work is sitting in the catalogue. Feelings are not numbers, and the argument never resolves, so we measured it.

Adobe Stock is the only one of the four marketplaces we track that publishes a first-party generative-AI flag on its own assets. That makes one measurement possible: query the illustration category for a subject twice, once unfiltered and once with the AI filter on, and take the ratio. It is Adobe's label rather than our classifier, which is both the strength of the number and its limit.

The headline

82.2% of the keywords we track are now majority-AI. For the typical subject, roughly three of every four illustrations a buyer scrolls past were generated.

By category

Two columns, because they answer different questions. The weighted share counts every asset, so a subject with four million illustrations dominates it. The median column treats every keyword equally, which is closer to what you experience choosing what to draw next. Where the two disagree, the gap is itself the finding — see the section below.

People & Lifestyle is the most saturated category we measure, at 81.2% weighted and 87.8% for the median keyword. If you draw people doing everyday things, you are working in the most crowded room in the building. Patterns & Backgrounds is the least saturated on both measures — for a reason that is arithmetic rather than aesthetic, and that the next section is about.

The pattern worth understanding

The strongest signal in this data is not category. It is catalogue size. The correlation between log(catalogue size) and AI share is -0.39 — bigger catalogues are consistently less AI-saturated.

The reason is dilution, not taste. Those libraries were mostly filled before 2023, so generated work has to outnumber a very large pre-existing pile before it can move the ratio. You can see it at both ends. The least-AI subjects with 2,000+ assets are broad, old, generic terms:

And the most-AI subjects are specific, recently-filled, scene-shaped phrases:

That second reading is the uncomfortable one. The descriptive long-tail scene phrases that stock advice has recommended for a decade — a family gathering, a swiss village, a scavenger hunt — are the ones sitting at 99%. They were easy to fill because they are easy to describe, and being easy to describe is now the same thing as being easy to generate. That specific strategy is closed.

Method and limits

Being explicit, because numbers like these get quoted back at us:

• ·Source: Adobe Stock's own generative-AI filter over its own illustration category. It inherits whatever Adobe's labelling misses, and contributors do mistag.
• ·Scope: illustration only. No photography — none of our supply collectors read photo catalogues, so we have nothing to say about that market.
• ·Sample: 1,333 keywords with at least 500 illustrations, out of ~29,000 phrases tracked overall. Readings only count from 2026-08-02, when we fixed a bug that had been pairing each AI flag with the following asset in the grid. Everything before that date was discarded rather than corrected, which is why the sample is this size.
• ·A snapshot, not a trend. These are current readings; we are not yet claiming a month-over-month direction from them.
• ·Shutterstock, iStock and Etsy publish no AI flag at all, so none of this extends to them.

Every phrase in the tables above has a public page carrying these same readings — for example haunted house — so none of this has to be taken on trust. Browse the whole corpus or start from what to draw next.