App Intents Are the New ASO: How Siri AI Will Discover Apps
Apple's app discovery is moving from keyword ranking to intent. Apps that declare their actions through App Intents get surfaced by Siri AI, while the rest fall outside assistant-driven discovery.
Apple app discovery is moving past the keyword-ranked search box. At WWDC26 Apple introduced Siri AI and rebuilt the connective tissue underneath it, and the result is a second discovery channel that runs on what your app can do rather than what words sit in your listing. App Intents, the framework developers use to declare an app's actions and content to the system, is the interface for that channel. Apps that expose their actions, entities, and in-app content can be reached by Siri AI, Spotlight, Shortcuts, and the App Store's new recommendation surfaces. Apps that do not will sit outside those workflows. Keyword ranking still works. It is no longer the whole game.
This is the same retrieval logic that already governs ChatGPT and Perplexity, arriving on the phone. An assistant takes a request, works out the capability that would answer it, and surfaces the app that declared that capability. App Intents is how you get into that index.
App Intents became a discovery surface the moment Siri AI shipped
App Intents is the only framework that connects a third-party app to the rebuilt Siri, which is what moves it out of the back office. Apple issued a formal deprecation notice for SiriKit at WWDC26. Apps still running on SiriKit keep compiling, but they will not surface when a user invokes Siri AI. The support window for the old framework runs roughly two to three years, per Phiture's WWDC26 recap. The migration is not urgent this week. The deprecation clock is running regardless.
Here is the mechanism. App Intents lets an app declare discrete actions: start a workout, add an item to a list, pull a date from a screenshot into a calendar. Once declared, Siri AI can call those actions directly, without the user opening the app, and chain them across other apps in a single request. Apple's framing in its newsroom release is exact: the App Intents updates connect apps to Siri AI capabilities like personal context understanding, app actions, and onscreen awareness. The apps that declared an action are the ones Siri reaches when someone asks for it. The ones that have not are not in the running.
That absence is the part most ASO teams are underweighting. A user who asks Siri to do something your app does, and gets routed to a competitor that declared the action while you did not, never saw your listing. No impression. No keyword match. No conversion event to optimize. The discovery happened one layer below the App Store, in the capability index, and you were not in it.
Siri AI discovers what your app can do, not what your listing says
Siri AI finds apps through three declared signals: actions, entities, and in-app content. Each is a retrieval target that you expose deliberately through App Intents. Apple does not lift these from your metadata.
Actions are the verbs your app performs. An App Intent declares a task as a unit the system can invoke and combine with tasks from other apps. This is the layer that lets Siri run a multi-step request across several apps and reach yours for its piece.
Entities are the nouns your app understands: a workout, a transaction, a saved place, a document. Declaring them gives the assistant typed objects it can reference, pass between steps, and act on. An app that declares rich entities is legible to the assistant in a way an app that only opens to a home screen is not.
In-app content is the larger expansion this year. Siri AI draws on a semantic index of the user's personal data, including emails, files, photos, and calendar, built on a rebuilt Spotlight indexing backend. Developers can expose content from inside their app through that index, so Siri AI can read and act on in-app content, not just launch features. Apple demonstrated this with a messaging app: the user asked Siri about specific messages and replied through Siri without opening the app at all.
Read mechanically, that semantic index is a retrieval index, and your declared actions, entities, and content are the retrievable units sitting inside it. Exposing them is the act of making your app addressable to the assistant. There is a real privacy control attached: new manifest APIs let developers declare, per intent, whether a Siri interaction can go to the cloud or must stay on device, which matters for apps handling sensitive data. The consequence for discovery is the through-line of this whole release. The unit that gets found is the capability, not the page, the same way a passage rather than a page is the unit that gets cited in web AI search.
Semantic indexing reorders the ASO playbook around capabilities
When discovery runs on a semantic index, the question the system asks about your app changes, and that changed question is what reorders the work. Keyword search asked whether your listing contained the words the user typed. Semantic matching asks whether your app's declared capability means the same thing as what the user requested. Those are different questions, and they reward different inputs.
The direction is not new to WWDC26. App Store discovery has been moving from exact keyword matching toward semantic interpretation, where metadata, creatives, and reviews are read together as connected signals of what an app is for. At WWDC 2025 Apple introduced AI-generated App Store Tags built from app metadata using its own models. Siri AI, Spotlight, and Personalized Collections run on that same kind of interpretation.
The operative property of a semantic system is that ambiguity costs you. The clearer and less ambiguous your app's purpose and core use cases, the more reliably any of these systems can resolve what your app does and place it against a request. Vague positioning that a human could charitably interpret is positioning a semantic matcher will mis-resolve or skip. This holds across App Store search, Personalized Collections, and assistant discovery at once, which is why the work compounds rather than splitting into three separate jobs. None of this retires keyword ranking. Typed search and top charts still run on the keyword index. The semantic layer runs alongside it and serves the surfaces where the user is recommended an app or asks an assistant for one.
Your App Intents names are metadata now
The names you give your actions and entities are public-facing positioning read by a semantic system, so they need the discipline you already apply to a title and subtitle. An action labeled in internal product jargon, or labeled inconsistently with the rest of your store presence, is a weaker retrieval target than one whose label matches how users phrase the task out loud.
The reason is geometric. A semantic matcher reads your metadata, your App Intents names, your entity labels, your reviews, and your creatives as one body of signal about what the app is for. When those sources agree, they reinforce a single clear reading. When they contradict each other or leave gaps, they pull the reading in different directions and dilute it, the same way a passage that mixes topics pulls its embedding toward nothing in particular. Reviews carry weight here because they are third-party corroboration in the user's own language: when buyers describe the app the way your metadata and your intents describe it, the signal is consistent and strong. Creative Assets are now part of this surface too. The new Header asset appears in the product page header and in organic search results, can be tailored to the query that surfaced it, and is managed in an Asset Library that lets you update it without waiting on an app release. It is one more place where what a user sees has to agree with what your actions and metadata claim.
Audit your capability surface before iOS 27 ships this fall
The work to do now is an audit, and it is specific. iOS 27 and the new Siri AI features are in developer beta now, with public release expected this fall. Six things to check:
- Core actions. Which of your app's primary tasks are declared as App Intents? If a task a user would plausibly ask an assistant to perform is not declared, it is not reachable through Siri AI, Spotlight, or Shortcuts.
- Task-based use cases. Map the phrases users actually say for each task, then check that your action names and entity labels use that language.
- Metadata clarity. Title, subtitle, keyword field, and description should state what the app is for in plain terms a semantic system can resolve without guessing. Ambiguity is the failure mode.
- Apple Business and App Store presence. Confirm your category fit and that your entity is described consistently across Apple's surfaces, so App Store Tags and recommendation systems read the same app everywhere.
- App Intents coverage. Inventory any remaining SiriKit dependencies and plan the migration. SiriKit apps keep running but will not appear in Siri AI interactions, including the chained, cross-app requests that are the point of the new assistant.
- Measurement. Instrument what you can. Apple has not exposed detailed Siri-sourced discovery analytics, so track the proxies available to you: Shortcuts adoption, intent donations, and downstream in-app behavior from assistant entry points. Third-party tools that measure AI visibility for apps are beginning to report assistant presence, and the early data is worth collecting now.
One regional note. Siri AI is delayed in the European Union on iOS 27 and iPadOS 27 under the Digital Markets Act, and EU developers cannot test it during development. The SiriKit deprecation clock runs anyway, so EU teams should still start the App Intents migration even though they cannot yet see it working in Siri.
What Apple has not published, and why the discipline matters
Apple has not released a Siri AI ranking formula, and it has not published a ranking framework for Personalized Collections, so any pitch that promises you can rank in Siri or guarantee a recommendation slot is selling certainty Apple has not provided. Hold that line, because the difference between what is documented and what is inferred is exactly where overclaiming starts.
What is documented, from Apple's own materials and the developer sessions: App Intents is the path to Siri AI. Declaring actions, entities, and content makes them reachable by the assistant. Personalized Collections surface apps across the Apps, Games, and Search tabs based on a user's interests, app usage, downloads, and other App Store signals, each carrying an App Note that explains why the app was suggested. These are mechanical facts.
What is a reasonable inference, but unconfirmed: the same metadata clarity and semantic precision that helps App Store search and assistant retrieval probably also helps placement in Personalized Collections. AppTweak draws this inference and labels it as one, which is the correct way to handle it. The reasoning follows from how the systems work. It is not a published placement signal, and it should never be presented as if Apple had confirmed it.
What is unknown: when several apps can fulfill the same request, how Siri AI chooses among them. There is no published selection logic. Anyone who claims to know the tiebreaker is guessing.
The move worth internalizing is plain. For a decade, ASO optimized a page so a person could find it. The new surfaces optimize a capability so an assistant can call it. The page still matters for the person. The capability is what gets you into the room the moment the person stops searching and starts asking.