Overview: Navigating Apple's Privacy Questionnaire
Before any iOS application can be submitted for App Review, Apple requires developers to complete the mandatory App Privacy questionnaire in App Store Connect. The resulting disclosures are displayed publicly on your App Store product page as "Privacy Nutrition Labels," informing potential users about what data your application collects, how it is used, and whether it is linked to their personal identity or used to track them across third-party apps.
For developers who have built applications using AI coding assistants like Cursor, Claude Code, Windsurf, or Google Antigravity, completing this questionnaire can be challenging. AI agents frequently introduce third-party SDKs—such as Firebase, Supabase, Sentry, or RevenueCat—without developers realizing that these libraries collect device identifiers, crash diagnostics, and network telemetry in the background.
This guide explains Apple's privacy taxonomy, outlines how to conduct an AI-assisted project audit, clarifies the critical boundary between on-device Privacy Manifests and App Store Connect disclosures, and provides complete manual instructions for completing the questionnaire in the web portal.
The Core Privacy Concepts: Collection, Linking, and Tracking
Apple structures the App Privacy questionnaire around three fundamental concepts:
┌───────────────────────────────────────────────┐
│ Apple App Privacy Taxonomy (3 Pillars) │
└───────────────────────┬───────────────────────┘
│
┌──────────────────────────────────────────┼──────────────────────────────────────────┐
▼ ▼ ▼
┌─────────────────────────────────┐ ┌─────────────────────────────────┐ ┌─────────────────────────────────┐
│ 1. Data Collection │ │ 2. Data Linked to User │ │ 3. Data Used to Track │
├─────────────────────────────────┤ ├─────────────────────────────────┤ ├─────────────────────────────────┤
│ Gathering data from the device │ │ Associating collected data with │ │ Linking data collected from │
│ and transmitting it off-device │ │ the user's personal identity │ │ your app with third-party apps │
│ for longer than transient, │ │ (e.g., account ID, email, │ │ or websites for targeted ads or │
│ real-time memory processing. │ │ phone number, or device UUID). │ │ data broker monetization. │
└─────────────────────────────────┘ └─────────────────────────────────┘ └─────────────────────────────────┘
1. What Counts as "Data Collection"?
Data collection occurs whenever data is gathered from the device and sent to an external server (yours or a third party's) where it is stored for longer than needed to service the immediate request in real time.
Exceptions: If data is processed purely on-device (e.g., local CoreML inference, device sensors used live, or local SQLite storage) and never transmitted over the internet, Apple does not consider it collected.
2. What Counts as "Linked to You"?
Data is "Linked to the User" if it can be tied to a specific user's identity—such as through an account login (User ID), email address, phone number, name, or a persistent device identifier.
If data is strictly anonymized and aggregated such that it cannot be tied back to an individual user account, physical hardware identifier, or session ID, it may be classified as "Not Linked to You."
3. What Counts as "Tracking"?
Under Apple's App Tracking Transparency (ATT) framework, tracking specifically means linking user or device data collected from your app with user or device data collected from other companies' apps, websites, or offline properties for targeted advertising or advertising measurement.
Tracking also includes sharing user data with data brokers. If your app performs tracking, you must implement the AppTrackingTransparency framework and request user authorization via the ATT system prompt.
Apple's 14 Data Categories
Apple groups all potential user data into 14 distinct categories:
| Data Category | Typical Data Examples | Common Collecting SDKs / Services |
|---|---|---|
| Contact Info | Name, email address, phone number, physical address | Auth0, Supabase Auth, Firebase Auth, Clerk |
| Identifiers | User ID, Device ID, Vendor Identifier (IDFV) |
Firebase Analytics, Mixpanel, RevenueCat, AppsFlyer |
| Usage Data | Product interaction, launch count, button taps, session time | PostHog, Amplitude, Google Analytics for Firebase |
| Diagnostics | Crash logs, launch performance data, stack traces | Sentry, Crashlytics, Datadog |
| Purchases | In-app purchase history, subscription state | StoreKit 2, RevenueCat, Adapty |
| User Content | Emails, text messages, photos, audio, AI chat prompts | OpenAI API, Anthropic API, Gemini API, Cloud Storage |
| Location | Coarse or precise GPS location, IP geolocation | CoreLocation, MapKit, IP lookup services |
| Search History | Search queries typed into in-app search fields | Algolia, Typesense, internal backend search |
| Browsing History | Content viewed inside in-app web views | In-app browser analytics |
| Financial Info | Payment info, credit card details, bank details | Stripe SDK, Plaid SDK |
| Health & Fitness | Heart rate, workouts, medical data | HealthKit |
| Contacts | Address book contacts from the user's phone | CNContactStore (Contacts framework) |
| Sensitive Info | Racial or ethnic data, religious beliefs, sexual orientation | Custom profile forms |
| Other Data | Any collected data not fitting the above categories | Custom proprietary telemetry |
Evaluating Generative AI and LLM APIs: A 10-Point Factual Assessment
For apps that integrate external generative AI APIs (such as OpenAI, Anthropic Claude, Google Gemini, or custom hosted models), whether and how to disclose data collection depends on how your app transmits, processes, retains, and links that data.
Apple's official privacy rules do not apply a universal label to every AI feature. Instead, developers must conduct a factual assessment across ten critical dimensions:
- What Information Is Transmitted? Does your app transmit free-form user prompts, structured search parameters, pre-written templates, system context, user profile details, or attached media?
- Does It Include Free-Form Text or User Communications? If users can type arbitrary text, prompts may contain sensitive personal thoughts, messages, or inquiries. In Apple's taxonomy, free-form text entered by users is classified under User Content (Emails or Text Messages if communication-focused, or Other User Content for general generative prompts).
- Does It Include Personal Identifiers or Attached Media? If your app allows uploading voice memos, images, or documents for AI processing, those trigger additional categories: User Content (Audio Data, Photos or Videos). If user accounts or email addresses are attached to API requests, Contact Info or Identifiers (User ID) are also involved.
- Is Transmitted Data Processed Purely in Real Time or Retained? Apple defines data collection as transmitting data off the device and storing it in an accessible form for longer than what is necessary to service the request in real time. If an API processes requests purely in transient memory with zero server-side logging or persistence, and neither you nor the vendor retains the payload, it may fall outside Apple's definition of collected data. Conversely, if the vendor retains logs (even for 30-day abuse monitoring under standard commercial tiers), the data is considered collected.
- Does the Vendor Use Transmitted Data Beyond Servicing the Request? Does your commercial agreement permit the AI vendor to train models on user inputs, run internal evaluations, or perform telemetry? If an AI vendor uses customer data to improve products or train foundation models, you cannot claim the data is used solely for immediate app functionality.
- What Are the Applicable Purposes in Apple's Taxonomy? - App Functionality: Transmitting inputs to generate real-time answers, summaries, code, or images that the user requested. - Product Personalization: Using historical prompts or user style preferences to customize model behavior, system prompts, or recommendations over time. - Analytics: Measuring prompt lengths, response latencies, failure rates, or feature engagement.
- Is the Data Linked to the User's Personal Identity? If API calls pass authenticated user IDs, customer emails, device UUIDs, or persistent tokens that tie back to an individual account, declare Linked to User: Yes. If requests use stateless, ephemeral, rotating session tokens where neither your server nor the AI vendor can associate inputs with an individual, declare Linked to User: No.
- Is the Data Used for Tracking? Under Apple's App Tracking Transparency (ATT) rules, tracking occurs if prompt data or associated identifiers are shared with third-party advertising networks, data brokers, or combined across external companies' apps for targeted advertising. Most generative AI workflows do not constitute tracking (Tracking: No), unless monetization or ad networks are integrated.
- Do Contractual Agreements Match Code Configurations? Many AI providers offer different data retention policies depending on tier. For example, standard developer endpoints may retain prompt logs for abuse monitoring, while enterprise Zero-Data-Retention (ZDR) agreements contractually guarantee that prompts are never written to non-volatile storage. Developers must verify that their actual API keys and backend routing correspond to their declared privacy terms.
- Are There Applicable Optional Disclosure Exceptions? Apple allows omitting data from the nutrition label only if the data meets all four strict conditions: (a) not part of the app's primary functionality; (b) infrequent and optional; (c) collected in an interface where the collection is evident to the user; and (d) never used for tracking or advertising. For conversational AI assistants, generative search engines, or AI text editors, prompt processing is primary functionality, meaning the optional disclosure exception does not apply.
Data Collection in Common Mobile SDKs
Beyond AI APIs, third-party mobile dependencies frequently introduce data collection that must be declared:
1. Crash Reporting and Diagnostics
If your app includes Sentry, Firebase Crashlytics, or Datadog:
- Data Type: Diagnostics (Crash Data and Performance Data).
- Purpose: App Functionality and Analytics.
- Linked to User: Select No if using anonymous device UUIDs; select Yes if your code attaches user emails or database IDs to crash scopes.
2. Authentication and User Databases
If your app integrates Supabase Auth, Firebase Auth, AWS Cognito, or Clerk:
- Data Type: Contact Info (Email Address, Phone Number) and Identifiers (User ID).
- Purpose: App Functionality.
- Linked to User: Yes (tied directly to user identity).
3. In-App Purchases and Subscriptions
If your app uses RevenueCat, Adapty, Superwall, or StoreKit 2:
- Data Type: Purchases (Purchase History).
- Purpose: App Functionality (and Analytics if tracking lifetime value).
- Linked to User: Yes (if purchases are tied to authenticated user accounts).
AI Audit Prompt: Evidence-Based Privacy Inventory
Before opening App Store Connect, have your AI coding assistant perform an evidence-based audit of your local repository. The prompt instructs the assistant to separate confirmed source-code findings from unresolved vendor-policy questions that require human verification.
Evidence-Based Project Audit Prompt
Copy and paste this prompt into Cursor, Claude Code, or Antigravity:
Audit my iOS project and its third-party dependencies to build an evidence-based inventory for Apple's App Privacy questionnaire.
For every identified SDK, framework, network endpoint, and data transmission path:
1. Identify the specific data types handled (e.g., User Content, Identifiers, Diagnostics, Contact Info, Purchases).
2. Distinguish confirmed source-code observations from unresolved vendor-policy questions (such as server-side log retention, model training terms, or backend analytics).
3. Check whether identifiers (User ID, email, IDFV) are attached to outgoing telemetry or AI API calls.
4. Check for any usage of the AppTrackingTransparency framework or advertising SDKs.
5. Provide a list of specific third-party vendor agreements and backend logging practices I must manually verify before completing the questionnaire.
Do not make changes to source files. Output the findings as an evidence inventory.
What the AI Assistant Analyzes
Your assistant scans project configuration manifests (Swift Package Manager Package.swift, CocoaPods Podfile.lock, Flutter pubspec.yaml, React Native package.json), source imports, and API request schemas. It outputs a structured report separating verified facts from open questions:
Evidence-Based Privacy Inventory:
1. Confirmed Codebase Evidence:
• Sentry SDK: Captures stack traces, thread state, and OS version.
- Category: Diagnostics (Crash Data, Performance Data)
- Purpose: App Functionality & Analytics
- Linking: Anonymous installation UUID detected (Not linked to user unless configureScope attaches User ID).
• Supabase Auth: Collects user email address and mints JWT UUID.
- Category: Contact Info (Email Address), Identifiers (User ID)
- Purpose: App Functionality
- Linking: Linked to User (Yes).
• OpenAI API Client: Transmits chat text payloads to https://api.openai.com/v1/chat/completions.
- Category: User Content (Other User Content)
- Purpose: App Functionality
2. Unresolved Vendor Policies (Requires Manual Verification):
• OpenAI Data Retention: Confirm whether your API tier has Zero-Data-Retention (ZDR) active or retains 30-day abuse monitoring logs.
• Sentry User Scope: Verify whether AppDelegate.swift or main.dart calls Sentry.setUser() with real user emails.
• Analytics Opt-Out: Check whether crash reporting respects user telemetry opt-out preferences.
3. Tracking Assessment:
• AppTrackingTransparency (ATT): Not imported.
• Advertising Identifiers (IDFA): No calls to ASIdentifierManager.
• Tracking Declaration: "No, we do not track users."
Source Inspection Is Only One Part of Compliance:
An AI coding assistant can inspect what client source code and imported libraries do locally. It cannot read your commercial vendor contracts, inspect private backend server databases, or know your external data retention schedules. Developers remain legally responsible for ensuring that declarations in App Store Connect match their complete data ecosystem.
Privacy Manifests vs. App Store Privacy Nutrition Labels
Starting in iOS 17, Apple introduced a two-tier privacy system that addresses different compliance layers:
┌─────────────────────────────────────────────────────────────────────────────┐
│ Tier 1: On-Device Privacy Manifests │
│ (PrivacyInfo.xcprivacy) │
├─────────────────────────────────────────────────────────────────────────────┤
│ • An XML property list file embedded directly inside your compiled `.ipa`. │
│ • Declares "Required Reason APIs" used by your code or third-party SDKs │
│ (such as file timestamp APIs, system boot time, disk space, or user defaults).│
│ • Scanned automatically during upload processing (triggers ITMS-91053 if │
│ missing or undeclared). │
└─────────────────────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────────────────┐
│ Tier 2: App Store Connect Privacy Nutrition Labels │
│ (Web Questionnaire) │
├─────────────────────────────────────────────────────────────────────────────┤
│ • Completed manually in the App Store Connect web portal. │
│ • Creates the public-facing "Nutrition Label" cards on the App Store. │
│ • MUST ACCURATELY ALIGN with the declarations in your Privacy Manifest. │
└─────────────────────────────────────────────────────────────────────────────┘
An upload processing error such as ITMS-91053: Missing API declaration cannot be resolved in App Store Connect web settings. It requires embedding a valid PrivacyInfo.xcprivacy manifest in your project bundle. Conversely, having an PrivacyInfo.xcprivacy file does not exempt you from answering Apple's App Store Connect web questionnaire. Both are mandatory.
Product Boundary: What FlightDeck Can and Cannot Automate
Understanding where automation ends and the Apple web portal begins is essential:
- No Public Questionnaire API: Apple does not provide a public REST API for programmatically answering or updating the App Privacy nutrition label questionnaire. The questionnaire must be completed inside the App Store Connect web portal.
- Automating Privacy Policy URLs: Apple requires a valid Privacy Policy URL for every app that collects data or offers subscriptions. FlightDeck can inspect and stage updates to your localized Privacy Policy URLs: - Assistant Prompt:
"Check our privacy policy URL across all locales in FlightDeck, and stage an update setting the en-US privacy policy URL to https://example.com/privacy."- FlightDeck stages the update for your review in the Approval Center and executes it viaapp_info_localization_update. - Pre-Submission Release Readiness Audit: FlightDeck can check your app's pre-flight release readiness: - Assistant Prompt:
"Using FlightDeck, run a release readiness audit for our app and confirm whether all required metadata, version information, and privacy prerequisites are complete."- FlightDeck invokesrelease_readiness_check. If privacy policy URLs or required store fields are missing, FlightDeck flags them before you submit to Apple App Review.
Step-by-Step Walkthrough: Completing the App Privacy Questionnaire
Follow these manual steps in your browser to complete Apple's questionnaire:
Step 1: Navigate to App Privacy in App Store Connect
- Sign in to App Store Connect with an Admin or App Manager account.
- Go to Apps and select your application.
- In the left sidebar under the General section, click App Privacy.
Step 2: Confirm Privacy Policy URL
Under the Privacy Policy section at the top:
- Confirm that your public Privacy Policy URL is entered and loads correctly in a browser.
- If your app offers auto-renewable subscriptions, ensure your privacy policy clearly describes your subscription terms.
Step 3: Complete the Data Collection Questionnaire
- Click Get Started (or Edit if previously completed).
- Answer the primary question: - Do you or your third-party partners collect data from this app? - If your app uses Firebase, Sentry, Supabase, or external AI APIs, select Yes, we collect data from this app. Selecting "No" when third-party telemetry SDKs are present will lead to App Review rejection under Guideline 2.3.10.
- Check each relevant data type identified during your audit: - Contact Info: Check Email Address (if using email login). - Identifiers: Check User ID (if using user accounts) and Device ID (if using analytics SDKs). - Diagnostics: Check Crash Data and Performance Data (if using Sentry or Crashlytics). - User Content: Check Emails or Text Messages or Other User Content (if sending prompts to LLM APIs). - Usage Data: Check Product Interaction (if tracking screen views or feature usage).
- For each selected data type, answer Apple's three secondary questions: - Data Use: Select the purpose (e.g., App Functionality, Analytics). - Linked to User: Select Yes if data is tied to user accounts or device identifiers, or No if strictly anonymous. - Tracking: Select No unless you share data with ad networks, data brokers, or cross-app tracking platforms.
- Review your answers on the summary screen.
- Click Publish in the top-right corner to make your Privacy Nutrition Labels active.
Top 4 Privacy Pitfalls in AI-Generated Apps
| Pitfall | Why It Causes App Review Rejection | How to Fix It |
|---|---|---|
| Claiming "No Data Collected" with SDKs Installed | App Review scans the binary for known SDK signatures (e.g., Firebase, Mixpanel). If detected without disclosures, review fails under Guideline 2.3.10. | Declare the analytics, identifier, and diagnostic data collected by your dependencies. |
| Missing Privacy Policy URL | Guideline 5.1.1 requires a functioning privacy policy link for all apps collecting data. | Host a live privacy policy and configure privacyPolicyUrl via FlightDeck or App Store Connect. |
| Undisclosed AI Prompt Processing | Sending user input to third-party LLMs without disclosing User Content collection. | Disclose User Content for App Functionality in the App Privacy questionnaire. |
| Mismatch with Privacy Manifest | Disclosures in App Store Connect contradict reasons declared in PrivacyInfo.xcprivacy. |
Ensure your Xcode privacy manifest and App Store Connect answers declare consistent data types. |