How to Build Siri AI Skills With App Intents

On 3 November 2026, Apple opened its Siri AI Developer Preview to every developer through the iOS 27.3, iPadOS 27.3, macOS 27.3 and visionOS 27.3 betas. Until now, a third-party app could only act inside Siri AI where its task fitted one of Apple’s predefined categories. From this preview, any app can expose its own content and actions through App Intents, and developers can teach Siri AI to chain several of those actions together with a new layer called app skills. This guide walks through the build in order: the setup you need, the one tool intent to start with, how to write a skill, how to test it, and the limits that decide what you can ship — and when.

Quick answer: Apple's Siri AI preview lets any app expose actions via App Intents and chain them with skills. This guide covers setup, tools, testing and limits.

A plain-language warning first, because it changes the whole plan: this is a build-and-test preview, not a customer release. Integrations work only in builds a developer runs themselves or shares with internal TestFlight testers. They do not work in external TestFlight builds or in App Store versions, and Apple says App Store submissions for Siri AI experiences will open in spring 2027, with the first customer experiences arriving in a few months. Build now to be ready for spring; do not promise users a Siri AI feature in your current App Store app.

What actually changed on 3 November

Three changes sit at the centre of the preview, as reported from Apple’s developer announcement on 3 November 2026.

First, custom actions. Any app can bring its content and actions directly into Siri AI by using App Intents, the framework already used for Shortcuts and widgets. Siri AI can then draw on personal context understanding, onscreen awareness and natural language to carry out a request in plain words — Apple’s examples are checking a delivery, booking a reservation and placing an order.

Second, app skills. A skill is a Markdown file that teaches Siri AI how to handle a request too messy for a single action: which actions fit together, which to prefer, what to avoid, and any knowledge specific to the app. Apple’s shopping example walks through searching for items, adding them to a basket, choosing payment and delivery methods, reviewing the basket and confirming the purchase.

Third, open content. Siri AI can now surface anything an app has placed in the device’s Spotlight semantic index, not only content that matches one of Apple’s templates. That matters for search-style requests — finding a saved item, a past order or a document — even before you build a single action.

“Starting today with the iOS 27.3, iPadOS 27.3, macOS 27.3, and visionOS 27.3 betas, a new Developer Preview enables any developer to integrate their app with Siri AI, giving businesses and developers new ways to reach and serve their customers,” Apple said in its developer announcement, as reported on 3 November 2026.

Before you start: the requirements checklist

Work through this list before writing any code, because each item is a hard gate rather than a preference.

A supported device and language. Siri AI needs an Apple Intelligence-enabled device set to a supported language. The preview adds Traditional Chinese, Norwegian, Turkish and Vietnamese in the 27.3 betas, taking the supported count to ten, alongside English, French, Japanese, Korean, Portuguese and Spanish. Siri AI is unavailable to users under 13.
The 27.3 betas and Xcode 27.3. Install the iOS 27.3, iPadOS 27.3, macOS 27.3 or visionOS 27.3 beta on your test device, plus Xcode 27.3. The preview covers iOS, iPadOS, macOS and visionOS only; a tvOS 27.3 beta shipped the same day, but the Developer Preview does not extend to it.
Region check. Siri AI is currently unavailable in the EU on iOS, iPadOS and watchOS, and Apple says it is working with the EU on a possible route to availability. Do not plan a launch test around EU iPhone users for now.
Internal testing only. Plan your test group as yourself plus internal TestFlight testers. External testers cannot use the preview, and neither can App Store builds, even if they contain the new integration.
A single, safe first task. Pick one action a user would actually say out loud, with a clear success state and a reversible result. Checking a delivery is ideal; cancelling an account or moving money is not a first intent.

Step 1: Install the betas and open the Siri AI documentation

Install the 27.3 beta on a dedicated test device — not your daily phone — and Xcode 27.3 on your Mac. Read Apple’s Siri AI Developer Preview documentation before touching your app’s code, and note the dates for Apple’s Apple Intelligence Group Labs, held on 4–5 November 2026, where Apple engineers answer implementation questions. If you hit a wall on day one, that lab window is the fastest route to an authoritative answer.

Keep your production app untouched while you experiment. Work in a feature branch, and keep the Siri AI work behind the beta-only build configuration so nothing from the preview leaks into an App Store submission before spring.

Step 2: Expose one action as a tool intent

App Intents is the mechanism through which Siri reaches into your app’s actions (called intents) and content (called entities). The preview removes the old template requirement: a developer marks an intent with the Swift macro @AppIntent(.tool), which makes it conform to a new ToolIntent protocol and exposes it to Siri as a tool the assistant can call.

Build it in this order:

Choose the intent. Start with the single action from your checklist — for a delivery app, “check delivery status”. It has one clear input (an order) and one clear output (a status).
Mark it as a tool. Apply @AppIntent(.tool) to a new or existing custom intent. Do not mark every action in your app at once; each tool Siri can call is a surface you must test and maintain.
Write titles and descriptions as if Siri will read them aloud. Apple’s guidance makes this explicit: titles and descriptions for intents, parameters and entity properties help Siri interpret natural-language wording, distinguish similar actions and formulate responses. “Check delivery status” with a description naming the order number it needs is good; a bare “Check” is not.
Expose only what the task needs. Developers decide which content Siri may retrieve and which actions it may perform. If an action needs a payment method or an address, ask whether the spoken version of that task should really have it — and if in doubt, leave it out of the first version.
Test the single intent alone. Ask Siri AI for that one task, in several wordings, before adding anything else. If it cannot reliably pick the right intent and fill its parameters, no skill layered on top will fix it.

Step 3: Write a skill for the multi-step job

A skill is for the request that one action cannot finish. Apple’s own guidance is that a skill is a Markdown file, conventionally called SKILL.md, whose metadata — a name and a one-line description — helps Siri decide whether the skill matches a request. Siri consults the instruction body only after making that match.

Write the skill in plain language, in this structure:

Name and one-line description first. This is the matching surface. Make the description describe the job, not the app: “Reorder a previous delivery, choosing the same items, payment and address, and confirming before purchase” will match better than “Delivery features”.
Action order, as guidance. List the steps in the order that makes sense — search, add to basket, choose payment and delivery, review, confirm — and say which intent to prefer where two could fit.
Practices to avoid. Name the failure modes explicitly: do not confirm a purchase without showing the total, do not substitute an item without asking, do not use a saved address that has changed.
App-specific knowledge. Add only what Siri cannot infer from the intents themselves, such as how your app defines a “previous order” or which statuses mean “out for delivery”.
When to ask. State the points where Siri should ask a clarifying question rather than guess — a changed address, an unavailable item, an ambiguous order.

One caution from Apple’s documentation is worth keeping verbatim in your planning: “You can’t use app skills to enforce deterministic behavior.” A skill guides Siri; it does not script it. Siri will not simply run your listed steps in order. Design every multi-step flow so that a skipped, reordered or clarified step still ends safely — ideally by requiring an explicit confirmation before any purchase, booking or change takes effect.

Step 4: Index your content so Siri can find it

Separately from actions, place the content users ask about into the device’s Spotlight semantic index. Siri AI can now surface custom content from that index without it fitting an existing schema, so a saved order, a bookmarked item or a recent document becomes findable by description rather than by filename alone.

Treat this as a privacy decision as much as a technical one. Index the content that makes the spoken task work, and nothing else. Review what your app already indexes, and remember that the same content Siri can surface is content a spoken request can reach — so sensitive records should stay out of the index until you have a reason, and a safeguard, for including them.

Step 5: Test with the Siri Integration Advisor and Siri Evaluations

Apple provides two testing tools in the preview, and using both is the difference between a demo that works once and an integration that survives real phrasing.

Siri Integration Advisor, in Xcode, gives implementation feedback on your intents and skills as you build.
Siri Evaluations lets you script requests, measure responses and detect changes as your intents and entities evolve.

Build a small evaluation set before you call the work done: the same task phrased five different ways, the same task with a missing detail (no order number), the same task with an ambiguous detail (two recent orders), and the dangerous variant (a request that should trigger a confirmation, not an action). Re-run that set after every change to intents or skill text. If a wording change in a description breaks a previously passing case, you have found a regression before your users do.

Worked example: a delivery app, end to end

Here is the whole build applied to one app, as a concrete pattern to copy — this is a desk-designed example from Apple’s published examples and documentation, not a hands-on test of a shipped app, and it is labelled as such.

The job: a user says, “Where is my order, and reorder the same things if it has already arrived.”

Tool intents (two, not twenty): “Check delivery status” and “Reorder previous order”. Each has a clear parameter — the order — and a clear output.

The skill: “Reorder a delivered order” — check the status first; if it is still in transit, say so and stop; if it is delivered, list the items and total, ask for confirmation, and only then place the order. Never substitute an unavailable item without asking, and never use a changed delivery address without confirming it.

The index: past orders, so Siri can find “the order from last Tuesday” without the user reciting an order number.

The evaluation set: “Where is my parcel?”, “Has my order arrived yet?”, “Reorder what I bought last time”, the same request with two recent orders (must ask which), and the same request with a changed address (must confirm, not assume).

The ship decision: this integration is ready for internal testing now, and for App Store submission in spring 2027 at the earliest. The customer-facing launch waits for Apple to open submissions and publish the terms and usage policies that will govern Siri AI experiences — terms Apple had not yet published at the time of the 3 November announcement.

Our take: the decision checklist before you invest a sprint

This preview rewards a narrow, careful first integration and punishes a broad one. Use this checklist to decide whether to start now:

Do you have one spoken task users already try? If users already ask Siri about your app’s content, that task is your first intent. If not, survey support requests before building.
Can the task fail safely? A read-only check or a confirmation-gated action is a good first skill. An irreversible action without a confirmation step is not.
Is your test group internal? If your roadmap depends on external TestFlight feedback this month, Siri AI cannot be part of it yet.
Can you wait until spring to ship? The value of building now is readiness and feedback, not revenue this quarter. Budget the work as preparation, not as a launch.
Have you read the terms gap? Apple had not published the terms and usage policies for Siri AI experiences at announcement time. Do not build a business model on a Siri AI surface until those terms exist.

For teams weighing where Siri AI fits alongside other assistant surfaces, our guide to Claude Chat + Cowork for Enterprise (https://openaimaster.ai/claude-chat-cowork-enterprise-guide/) shows a contrasting multi-step agent pattern inside a workplace product, and our Anthropic agent skills explainer (https://openaimaster.ai/anthropic-agent-skills-open-source-explained/) covers the parallel idea of packaging reusable agent instructions — useful context for how skills differ from plain prompts. For a platform-access comparison, see our Gemini 4 Argon access guide (https://openaimaster.ai/gemini-4-argon-access-how-to-get/).

Frequently asked questions

Can I ship a Siri AI integration in my App Store app today?

No. The preview works only in developer builds and internal TestFlight builds. App Store versions cannot offer Siri AI integrations yet, and Apple says App Store submissions will open in spring 2027.

Do I need a new developer programme or entitlement to join the preview?

No. Apple opened the preview to all developers, with no new entitlement or separate developer programme required. You need the 27.3 betas, Xcode 27.3 and an Apple Intelligence-enabled device.

What is the difference between a tool intent and a skill?

A tool intent is a single action Siri can call, exposed with @AppIntent(.tool). A skill is a Markdown guide that teaches Siri how to combine several actions into one job, such as a full shopping checkout. Skills guide behaviour; they cannot enforce an exact sequence.

Will Siri AI work for my users in the EU?

Not on iPhone, iPad or Apple Watch for now — Siri AI is currently unavailable in the EU on those platforms, and Apple says it is working on a route to availability. Check Apple’s current regional availability before planning a launch.

When will customers actually get third-party Siri AI integrations?

Apple says the first customer experiences will arrive in a few months, with App Store submissions opening in spring 2027. Treat any earlier date in third-party coverage as unconfirmed unless Apple states it.

Sources and methodology

Sources: Apple Developer, “Siri AI is now open to all developers” (3 November 2026), as reported by Unite AI, “Apple Opens Siri AI Developer Preview for Custom App Integrations” (3 November 2026, https://www.unite.ai/apple-opens-siri-ai-developer-preview-for-custom-app-integrations/); Digital Trends, “Apple opens Siri AI to all app developers” (3 November 2026, https://www.digitaltrends.com/computing/apple-opens-siri-ai-to-all-developers-allowing-more-advanced-third-party-app-controls/); iClarified, “Apple Opens Siri AI App Integrations to All Developers” (3 November 2026, https://www.iclarified.com/103272/apple-opens-siri-ai-app-integrations-to-all-developers); MadRobot, “Apple opens Siri AI to every app, and lets developers teach it skills” (3 November 2026, https://madrobot.blog/2026/11/03/siri-ai-app-skills-developers-ios-27-3/). All dates, platform versions, testing limits and spring-2027 submission timing are taken from those reports of Apple’s announcement; where reports differed in wording, the article follows Apple’s stated positions and labels the rest as reported. This article is desk-researched, not hands-on tested — we have not run Xcode 27.3 or the 27.3 betas, and the delivery-app example is a designed pattern from Apple’s published examples, not a shipped integration. Claims were verified on 4 November 2026.

Govind Dheda

Govind Dheda covers AI tools, guides and prompts for OpenAIMaster — what is new, what is worth using, and how to put AI to work.