ASO used to be a fairly mechanical exercise: research keywords, place them in the title and metadata, test screenshots, repeat. That's no longer the whole job. Apple and Google have both started inserting AI directly into how apps get discovered — not as a tool developers opt into, but as infrastructure sitting underneath every listing, whether a developer has touched it or not.
What Apple actually shipped
At WWDC 2025, Apple announced App Store Tags, labels generated by Apple's own large language models from an app's metadata, description, category and screenshots. The tags went live in beta days later, and the mechanism is worth being precise about: Apple's AI is extracting information already buried in a listing — including reading screenshots directly, not scanning for keywords a developer planted there — then surfacing it as a browsable tag a user can tap into a curated collection of similar apps. Tags are human-reviewed before publishing, and a developer can remove an inaccurate one, but can't write their own. There's nothing to keyword-stuff here, because there's no keyword field involved at all.
What Google shipped, and how it's structurally different
Google's version puts more of the AI work directly in the developer's hands. At Google I/O 2026, Google announced that Gemini now generates app suggestions during Play Store search and, more concretely useful for day-to-day ASO work, can build an entire custom store listing automatically: click a keyword recommendation in Play Console, and Gemini creates a new listing variant tailored to that keyword, ready to publish in one click. Where Apple's tags are something that happens to a listing from the outside, Google's version is a tool a developer actively drives — same underlying shift toward AI-mediated discovery, opposite philosophy about who's in control of it.
What this means for how ASO actually gets done
Neither platform's AI layer responds to the old lever of writing more keywords into a description. Apple's tags come from what the AI can actually verify about an app — accurate, specific screenshots and metadata that plainly show what the app does, not marketing copy written to game a ranking. Google's Gemini tools work from the same starting material: a keyword recommendation is only as good as the store listing data feeding it. In both cases, the practical work shifts from keyword placement toward making sure an app's actual ASO foundation — screenshots, category, description accuracy — gives the AI something real and unambiguous to work with, the same underlying principle behind technical SEO work on the web: structured, accurate information is what an AI system needs before it can act on your behalf, whether that's citing a webpage or tagging an app.
This isn't a reason to abandon the fundamentals — ratings, conversion-optimized screenshots, and a clear value proposition still decide whether someone installs once they've been shown an app. It's a reason to stop treating metadata as a keyword-stuffing exercise and start treating it as the actual input an AI system reads before deciding whether your app belongs in front of anyone at all.