Back to Explore
GuideVibeFix editorialUpdated Nov 16, 2026

Don't Guess the Three Small Lines on Your Store Page: Privacy Labels and Data Safety for Your Vibe App

Apple's three App Privacy tiers and Google's five-step Data safety questionnaire, and the inventory that comes before either form: 240 collection points, cut what you cannot explain, and line the label up with the policy word for word. All example projects, people, and numbers in this guide are fictional.

A real App Store screenshot: an iPhone on the App Privacy page, with 'Data Used to Track You' and 'Data Linked to You' sections listing Contact Info, Location, and Identifiers.

A-Yuan's "PocketLedger" is a one-person expense tracker: snap a photo of a receipt, log the spend, get a pie chart at the end of the month. That is all it does. The app is listed on both the Apple App Store and Google Play, with a few hundred people opening it every day. In early November, A-Yuan prepared a new version with a "budget reminder" feature: when the month's budget is nearly spent, the phone buzzes once. The feature itself took half a day to build. What actually blocked him was the form that popped up at submission time: Apple wanted him to fill in "App Privacy", and Google wanted a "Data safety" declaration. He recognized every word on the forms, but put together he had no idea how to answer them — what data does the app actually collect? What does that analytics SDK collect? Do crash logs carry users' email addresses? He stared at the forms for a whole evening, filled in a version by gut feeling, and hit submit. Three days later, Apple's review reply arrived: the label did not match the app's actual data collection, please fix it and resubmit. Google was no better: the Data safety form said "No" to "does the app collect location", while the app clearly had a "spending places nearby" feature that reads location. Only then did A-Yuan realize that a privacy label is not a box-ticking formality. It is something users can see on the store page at a glance. Fill it in wrong, and at best the review bounces back; at worst, users spot the lie on the spot and uninstall.

Two things first. First, the walkthrough app "PocketLedger", its maker "A-Yuan", and all dates and numbers in this guide are fictional, invented only to make the process concrete. Second, the boundary: this site has already published a privacy compliance guide about how to write a privacy policy and the broad compliance picture, an app-store rejection appeal guide about what to do after a rejection, and a mobile app store launch guide about the whole launch process. This guide repeats none of those three. It covers only one thing — how to actually fill in Apple's App Privacy label and Google's Data safety declaration, what to inventory before you fill them in, and how to line them up with your privacy policy word for word. The compliance guide handles the policy text, the launch guide handles the whole process, the appeal guide handles what happens after a rejection; this guide handles the few small lines on the store page and how to get them right.

A real App Store screenshot: an iPhone on the App Privacy page, with 'Data Used to Track You' and 'Data Linked to You' sections listing Contact Info, Location, and Identifiers.

1. What a privacy label is: the ingredients list on the store page

Let us get the concept straight first. A privacy label is not a shortened privacy policy, and it is not a slogan like "we take your privacy seriously". It is an ingredients list on the store page: before downloading, users can see what data the app will take, what it will be used for, whether it goes to third parties, and whether it is used to track them. Apple has required every listed app to fill in App Privacy since late 2020, and Google has rolled out the Data safety declaration on its store. The two stores use different names and different questionnaires, but they do the same job: they move the question "what do you collect" out of the small print on page 8 of the privacy policy and next to the download button, where everyone can see it.

This matters especially for vibe projects, for a very practical reason. First, most vibe projects are built by one person or a team of two or three, the code is written with AI help, third-party libraries are installed on a whim, and many makers genuinely do not know what their app collects. Second, store reviewers now actually check: Apple reviewers compare the label against the app's real network requests and permission calls, and bounce the submission if they do not match. Third, users have learned to read this label: if an expense tracker is marked as "tracks you", many people will hesitate no matter how good the features are. The more honestly a label is filled in, the steadier the download conversion; the vaguer it is, the higher the chance of a bounce or an uninstall. A-Yuan's first label was the cautionary tale: to save effort, he leaned everything toward "does not collect", and the reviewer spotted at a glance that the "spending places nearby" feature reads location, and sent it straight back.

Apple's label has three tiers, and this structure is the backbone of the whole guide, so remember it. The first tier is "Data Used to Track You": data used to track users across apps and websites owned by other companies, or handed to data brokers. Anything touching the word "tracking" belongs here, with no room for negotiation. The second tier is "Data Linked to You": data that is collected and can be linked to a user's account, device, or identity — for example, expense records uploaded after sign-in, or a bound email address. The third tier is "Data Not Linked to You": data that is collected but cannot be linked to any specific person, such as crash statistics with device identifiers stripped out. Beginners often confuse the second and third tiers, so remember one test: if the data can be followed back to a specific person, it is "linked"; only if it cannot is it "not linked". A-Yuan's expense records are uploaded after sign-in and can be linked to an account, so they belong in the second tier; his crash statistics only qualify for the third tier if the device identifiers have been removed.

Google's Data safety is a different questionnaire, built around five steps: first an overview, then three key questions — "does your app collect or share data", "is all user data encrypted in transit", "can users request deletion of their data" — then a step-by-step selection of data types and purposes, and finally a preview of what the store page will show. Google's questions are wordier than Apple's, but the upside is that each step comes with explanations, so it is hard to skip something by accident. The two stores' labels do not recognize each other: once Apple is done, Google still has to be filled in from scratch, and vice versa. A-Yuan's lesson was exactly this: he filled in Apple's form carefully and dashed off Google's with a few "No" answers, and the two stores ended up showing two different versions of the "truth".

2. Before filling in the form: inventory what your app actually collects

Before opening the store console, open your own code and dependency list. When A-Yuan filled in the form the second time, he reversed the order: one afternoon of inventory first, then the form. The method was clumsy but it worked: he made a table with "where the data comes from" on the left and "do we collect it, what for, who gets it" on the right. Data comes from four sources. The first is what users type in themselves: expense amounts, categories, notes, email addresses, avatars. The second is what the device and the system provide: device model, system version, language, time zone. The third is what features generate while running: location (the "spending places nearby" feature), photos taken in the app, push notification device tokens. The fourth is the easiest to miss, and it is what third-party libraries collect on their own: the analytics SDK, the crash reporting SDK, the push notification provider. Adding the four sources together, A-Yuan counted 240 candidate collection points — not 240 data types, but 240 "data plus purpose" combinations counted one by one, like "email for sign-in", "email for password recovery", and "location for nearby merchants".

Of the 240, the first pass cut 44. The reason for cutting was always the same: the purpose could not be explained. The classic case was "device identifier for personalized recommendations": A-Yuan searched the whole codebase and could not find anywhere that actually used the device identifier for recommendations — it was a default collection item of the analytics SDK, and the app itself never used it. The test is a single sentence: if you cannot explain what data is for, do not collect it. Not "collect it now, it might be useful later", but "if it is not used now, switch it off". Switching it off is straightforward too: turn that collection item off in the analytics SDK's dashboard, or switch to a version that does not collect it. After cutting the 44, the remaining 196 could each explain "do we collect it, and what for". Those 196 are the second step of the funnel.

Of the 196, another 38 got stuck on "the third party cannot explain it". The crash reporting SDK's documentation, for example, said it "may collect device information for diagnostics" — and that word "may" made it impossible to fill in the form. The fix is to check three things: the SDK's official privacy statement, the SDK's actual network requests, and whether the item can be switched off in configuration. A-Yuan checked 3 third-party libraries: the analytics SDK had the clearest documentation, listing its collection item by item; the crash SDK's documentation was vague, so he captured one actual crash report request and looked at what fields the request body really carried; the push provider's documentation was the most hassle-free, stating plainly that it only transmits the device token and no user content. Of the 38, 30 could be written up clearly after reading the documentation, and 8 turned out to be unused at all, so they were switched off. 196 minus 8, with the remaining purposes written out, gives 158 — every one of which can explain "what it is used for, and whether it goes to a third party". That is the third step of the funnel.

The 158 still had one gate to pass: matching the privacy policy word for word. A-Yuan took the 158 and compared them with the privacy policy, and found 37 mismatches: the policy said "we do not collect location information", but the "spending places nearby" feature plainly reads location; the policy said "we do not share data with third parties", but the analytics SDK plainly sends anonymized usage data to its own servers. There are only two ways to fix the 37: either change the product (turn "spending places nearby" into a manual location entry that no longer reads system location), or change the policy (change "we do not collect location" to "when you use the nearby merchants feature, we collect location, only to show nearby merchants, not for tracking"). A-Yuan changed the product for 12 of them and changed the policy for 25, and in the end all 121 could show the label, the privacy policy, and the actual behavior agreeing in three places. That is the fourth step of the funnel, and the only step that counts as submittable.

Privacy-label funnel (fictional sample): 240 collection points, 196 where the team could say whether to collect, 158 where purpose and third parties were clear, 121 where label, policy and behaviour all matched

3. Funnel review: how 240 collection points became 121 submittable answers

Draw the process as a funnel and the numbers look like this: 240 candidate collection points, 196 that can explain whether they are collected, 158 that can explain their purpose and third parties, 121 where all three places agree. As conversion rates: the first step is 196 divided by 240, which is 82%; the second step is 158 divided by 196, which is 81%; the third step is 121 divided by 158, which is 77%; and the whole journey is 121 divided by 240, which is 50%. In other words, only about half of what the inventory starts with survives to submission unchanged. That ratio is nothing to worry about — it is exactly the proof that the inventory worked. Of the 119 the funnel filtered out, 44 were "if the purpose cannot be explained, do not collect it", 38 were "if the third party cannot explain it, go and check, and switch off whatever turns out to be unused", and 37 were "if the policy and the reality do not match, change one side". Nothing filtered out at any layer is a "loss"; it is something that never belonged on the label in the first place.

The funnel has a second use: it tells you where to patch next time. If the first step converts worst, the biggest problem is "purposes that cannot be explained" at inventory time, so next time start by switching off unused collection. If the third step converts worst, the policy text is the stalest part, so next time rewrite the privacy policy first. In A-Yuan's funnel the third step was 77%, the lowest of the three, so the first thing he did in the second round was rewrite the two paragraphs of the privacy policy about location and third-party sharing, rather than counting collection points again. A funnel is not there to look nice; it exists so that "what do I do next" has a number behind it.

30 days after launch, A-Yuan did a review: the new version passed both Apple and Google review on the first try, the label on the store page matched the privacy policy word for word, and in 30 days there were 0 user complaints about "label mismatch", and no takedown over a label issue. The more concrete change was that download-page conversion did not drop: he had worried that honestly writing "collects location, linked to your account" would scare users away, but in practice users care more about whether "what it is collected for" is written clearly. An app that plainly writes "location is only used to show nearby merchants, not for tracking" earns more trust than one that vaguely writes "may collect location". Honesty is not the cost; vagueness is.

4. How to split Apple's three tiers: three mistakes that are easy to make

The first mistake is judging the "tracking" tier. Many people think "tracking" just means "an ads SDK is installed", but Apple's definition is wider: if data is used to identify users across apps and websites, or handed to a data broker, it counts as tracking. A-Yuan's analytics SDK had an "attribution across apps" item, used to tell whether a user arrived from an ad — and that item has to go in the "Data Used to Track You" tier. In his first version he put it under "Data Linked to You", and the reviewer bounced it. The test is to look for the words "ad attribution", "across apps", or "data broker" in the SDK's purpose statement. If they are there, it goes in the first tier; if you are unsure, put it in the first tier anyway — over-disclosing is safer than under-disclosing.

The second mistake is the boundary between "linked" and "not linked". Crash logs are the most common disputed case: a crash log carries no name, but if it carries a device identifier, and that identifier can be matched to an account, it is still "linked". A-Yuan's crash SDK carried the device identifier by default; he switched that item off in configuration, and only then did the crash logs qualify for "Data Not Linked to You". The test is the one already given: can the data be followed back to a specific person? There is also a practical question to ask yourself — if a user asks you to "delete all data about me", can you find this data item and delete it? If you can find it, it is linked; only if you cannot is it not linked.

The third mistake is data from "optional features". A-Yuan's "spending places nearby" is an optional feature; users can ignore it. But as long as the feature exists and tapping it collects location, the label has to say so. Many people think "most users never touch this feature, so I will leave it out" — that is wrong. The label asks "does the app collect it", not "do most users trigger it". The right approach is to state the trigger condition: "when you use the nearby merchants feature, we collect location information, only to show nearby merchants". Stating the condition clearly is far safer than pretending the feature does not exist. What reviewers dislike most is "it is in the feature, but not on the label".

5. Google's five-step questionnaire: follow the steps, do not skip any

Google's Data safety questionnaire has five steps: Overview, Data collection and security, Data types, Data usage and handling, and Preview. The second step's three key questions are the easiest to get wrong. The first asks "does your app collect or share any of the required user data types". A-Yuan first answered "No", because he thought "collect" meant "sell data". Google's definition is: if data leaves the user's device, that is collection; if data goes to a third party, that is sharing. Expense data uploaded to his own server is collection; analytics data sent to the analytics SDK's servers is sharing. The second asks "is all of the user data collected by your app encrypted in transit", and the answer depends on whether your APIs are all HTTPS and whether the third-party SDKs transmit over encryption. A-Yuan checked the whole chain, confirmed his own APIs and all 3 SDKs use HTTPS, and only then dared to answer "Yes". The third asks "do you provide a way for users to request that their data is deleted", which maps directly to the "delete account and data" entry inside the app — if there is no such entry, do not answer "Yes".

The fourth step, "Data usage and handling", is the most time-consuming, because every data type has to be ticked one by one for "collected or shared", what it is used for, and whether it is "processed ephemerally". "Ephemeral processing" means the data is used in memory for a moment and then discarded, never stored. When A-Yuan's app uploads a receipt photo, it compresses the image in memory first — that compression is ephemeral processing — but the photo itself is still stored on the server, so the photo item cannot be ticked as ephemeral-only. Tick every item according to the real flow; do not tick the same answers everywhere to save time. The final Preview step deserves a careful look: what the preview shows is exactly what users will see on the store page. In the preview, A-Yuan found "location information" described as "for personalization", while his real purpose was "show nearby merchants"; he went back, fixed the purpose description, and only then did the preview look right. The preview is not a formality; the preview is the final draft.

6. Three real failure scenes: SDKs, updates, and children's apps

The first failure scene is an SDK upgrade quietly adding a collection item. A-Yuan once upgraded the analytics SDK from 2.4 to 3.0, and one small line in the changelog said "new device fingerprinting for fraud prevention". He did not notice, and the label was not updated; two months later, Apple's review of the new version spotted it and bounced the submission. The lesson is: every time you upgrade an SDK, read the changelog for newly added collection items first, and only then decide whether the label needs changing. Build a habit: add one line to every upgrade note — "does this upgrade change data collection" — and if it does, update the label and the privacy policy together; if it does not, write "no change".

The second failure scene is forgetting to update the label after a release. When A-Yuan added the "budget reminder" feature, he used the push provider's device token, but the label had no "device token" item. The right fix is to put "label update" on the release checklist: every release, one checklist item reads "did this release add any new data collection — if yes, update the label before submitting". It is a single line, but it blocks almost all "I forgot" problems. Both Apple and Google require the label to be updated when data collection behavior changes, at the time the new version is submitted — not "whenever I get around to it".

The third failure scene is children's apps. A-Yuan's expense tracker has no child users, but if your app targets children under 13, both stores' requirements get much stricter: Apple requires children's apps not to collect data for tracking, and Google requires extra declarations in the Data safety form for children's apps. The test is whether your app is positioned as "primarily for children" — if it is, read both stores' dedicated children's-app rules before filling in the form. Do not copy the adult-app answers onto a children's app.

7. The submission checklist: tick every item before you submit

Before submitting, go through this checklist item by item, and submit only when everything is ticked. First, does the inventory table cover "for every item, can we explain whether it is collected, what it is for, and whether it goes to a third party", and have the items that cannot be explained been cut or switched off? Second, are Apple's three tiers split by the test "can this be traced back to a specific person", and has anything with ad attribution gone into the "tracking" tier? Third, has Google's five-step questionnaire been walked through step by step, and does the purpose description shown in the preview mean the same thing as the real purpose? Fourth, have the privacy policy and the label been compared word for word, and is it true that where the policy says "we do not collect", the app really does not collect? Fifth, have the collection lists of all 3 third-party libraries been checked one by one, and does every upgrade note carry a "data collection change" line? Sixth, does the release checklist carry a "label updated in sync" item? Seventh, for a children's app, have the dedicated rules been read? Eighth, is there a review 30 days later: did the review pass on the first try, did any user complain about the label, did conversion move oddly? Of the eight, the fourth matters most, because it is the point reviewers catch most often, and the point users spot most easily.

One last blunt but honest point: how well a privacy label is filled in is, at its core, not a form-filling skill but a question of whether you are honest with yourself about your own app. A one-person vibe project has no legal team and no compliance department. What it can do is write down clearly "what we collect and what it is for", not collect what it cannot explain, and write down plainly whatever it does collect. Users are not afraid that you collect data; they are afraid that you collect it while saying "we do not collect". A-Yuan's first label was bounced not because he was dishonest, but because he filled it in by gut feeling; his second version passed on the first try not because he hired a lawyer, but because he spent one afternoon counting his 240 collection points properly. Inventory before filling in the form, and honesty before inventory — do not reverse that order.

One reminder: this guide is a hands-on product guide, not legal advice. Different countries and regions have different data disclosure requirements, and children's apps, health apps, and finance apps have additional dedicated rules. If your app handles sensitive data, or you are entering multiple jurisdictions, having a professional lawyer or compliance adviser look it over once is far cheaper than guessing on your own.

Views 9Comments 0

Comments (0)

ME
0/1000
Loading comments...
Browse projectsPublish your project

Related articles

Real photo: a laptop, tablet, and phone on a wooden table showing the same green-leaf website, the phone view with navigation in one column, a potted plant nearby
Guide
Squashed on phones? Responsive layout and media queries, explained plainly

One page should stand up on a phone, a tablet, and a computer. This guide follows MDN's Responsive web design, Using media queries, and the viewport meta reference: fluid HTML, the viewport tag, fluid grids and max-width, images that never burst their container, media query syntax, breakpoints set by content, mobile first, responsive type, testing capability instead of width, and a pre-launch checklist.

Mobile AppsDesign ExperienceFrontend Engineering
Real photo: a bank of light gray metal filing cabinets with one drawer pulled open, revealing rows of hanging file folders inside
Guide
A real database inside the browser: IndexedDB local storage, explained plainly

localStorage only stores strings and has no indexes. IndexedDB is a real database in the browser: structured data, far larger capacity, indexes, and transactions. Following MDN's Using IndexedDB, IDBDatabase, and IDBObjectStore: the five-step routine, opening, upgrades, keys, indexes, transactions, add versus put, get, cursors, ranges, and multi-tab upgrades.

Tool TipsData WorkflowsFrontend Engineering
Real photo: hands typing on a MacBook Pro showing JavaScript code that changes HTML content, with a white mouse, papers, and binders on a white desk
Guide
Functions without the confusion: JavaScript functions, parameters, and arrow functions, explained plainly

You already call functions every day; this guide names what you have been doing. Following MDN's Functions, Arrow function expressions, and Rest parameters: declarations versus expressions, parameters and arguments, default values, rest parameters, spread syntax, return values, arrow function shorthand, the this trap, callbacks, and scope.

Tool TipsFrontend EngineeringAI Coding