← Back to blog
AiAbout 7 min read

A community spreadsheet of taps that turns off the ads inside Chinese Android builds

Published Nov 15, 2026
A community spreadsheet of taps that turns off the ads inside Chinese Android builds

The advertising sits in the settings app

Anyone who has used an Android phone bought in mainland China recognises the shape of this before they have words for it. Open the app store and the home screen is a carousel of promoted apps above a row labelled as essentials. Pull down on the launcher and the search page opens on a trending list. Swipe right from the first home screen and the cards are news. The stock weather app asks whether you want "content and marketing" notifications, and the settings app has a personalisation toggle that arrived switched on.

None of that is a third-party SDK buried in a free game. It is the system UI, shipped by the company that sold the phone, and it is a revenue line. Hardware margins in that market are thin. The software surface is where the margin gets rebuilt, so every screen that can hold a placement holds one, and each placement comes with a preference that can turn it off.

The README of 肉呆去广告 (Android Ad Cleaner, rddmwmddr/android-ad-cleaner, 8,115 stars on 15 November 2026) counts those preferences per handset. Its marketing notification column, which counts only the channels a model reserves for promotions, runs from single digits into the sixties.

PhonePromotional notification channels turned off
OPPO Find X9 Pro60
Xiaomi 14 Pro60
Huawei (ALN-AL00)48
Redmi K80 Pro48
OPPO Find N542
OPPO Find N630
Xiaomi REDMI K90 Pro Max26
Honor (PNM-AN10)26

Two of those numbers need a footnote before anyone quotes them. Huawei's 48 came out of a manual review of 79 candidates, 31 of which were already blocked at the app level, so it records what one reviewer switched off in one sitting rather than what a script achieves in a pass. Xiaomi's 60 on the 14 Pro started as 64 candidates with four service channels removed by hand. A channel count records how many places the vendor reserved the right to send you something, and the difference matters when you are deciding whether a phone is tolerable.

A blocker cannot reach any of it

A DNS filter or a hosts list stops a request. It cannot flip a preference. The promoted row in the app store is drawn by the app store itself, from the store's own backend, over the same connection that serves the listings you want, so blocking the ad domain risks breaking the update check too. The switch that says "show recommendations" lives three levels into a settings page whose label changes between OS releases.

So the fix has to touch the UI. That is the whole design of this project, and it explains why the thing ships as a desktop client with a USB cable rather than as a background service.

One file per phone model

The core artefact is a 机型脚本, a model script: one file for each handset, recording which app to open, which words to tap to reach the settings page, which switches to turn off and which to leave alone. The PC client drives the phone over USB debugging and finishes in a few minutes. There are builds for macOS on both Apple silicon and Intel, signed and notarised, plus Windows 10 and 11 in installer and portable form, plus an Android 11 and up APK for people without a computer. Downloads are mirrored on Gitee because GitHub is not reliably reachable from inside the country.

A blank black glass slab shaped like a large phone, one brass toggle switch at its lower edge flipped down.

Why one file per model rather than one clever heuristic? Because the settings pages differ by model and by OS sub-version. The same brand will move an entry, rename a page, or split a section between two HyperOS builds. The engine checks the detected version against the script, warns when they disagree, and skips a step whose entry it cannot find instead of tapping at coordinates and hoping.

This is the least glamorous engineering choice available. A root module or a hooking framework would be one generic solution covering every phone at once, and it would break on every OTA in ways the user cannot see. The volunteers picked the other trade: dozens of small solutions, each verified on one device, each needing a human to walk through a new ROM with a screen recorder and a text editor. Maintenance burden instead of privilege escalation.

The failure modes are in the same document, which is the part I trust. Models are marked working, partial or failing with the date of the last real test. Two OPPO handsets sit at failing. Contributor-submitted scripts are logged as written-but-not-rerun, so a Redmi K90 entry notes it ran eight of nine sections on a borrowed script from a different model. Nothing here is dressed up as universal.

What it refuses to do, and what that costs

No root, no bootloader unlock, no flashing, and system updates keep working. It only flips switches the system and the apps already expose. Suspending an ad package, with com.opos.ads and com.miui.systemAdSolution given as examples, needs the owner's explicit consent and one command puts it back. Search, the negative-one-screen cards, app store installs and updates, weather alerts and parcel notifications are kept, which is the detail that separates this from the scripts people trade in forum threads and forget about.

Two things in the risk column deserve naming. USB debugging is a real capability to hand a program, even briefly. Package suspension is a change to what runs on the device, not a cosmetic one. Both are reversible, and the project publishes a restore path with a command for the suspended packages and a command for the notification channels, which is more than most debloat scripts bother to write.

Version 0.4.4, released on 13 November 2026, fixed a long-list page that only turned off its final section, fixed a vivo lock-screen section that had never taken effect, merged duplicated section bodies and navigation in the engine, and added scripts for the OnePlus 16 and a GT5 Pro from the same ColorOS family. In-app update checking has existed since 0.4.2, so earlier installs had to be replaced by hand. Since that same release the Mac and Windows clients can also connect to HarmonyOS NEXT phones, a path the maintainer says has not been retested on real hardware.

The spreadsheet is the product

Here is the part worth sitting with. This tool exists because the default state of a phone sold in that market is worse than the default state of a phone sold in yours, and the correction is a community-maintained list of tap sequences with dates next to them. Somebody sat with an OPPO Find X9 Pro and counted sixty promotional notification channels, one at a time, and wrote down where each switch lives. Then somebody else did it for a Huawei, and again for a Honor, and the table grew to more than forty rows because ROMs keep moving.

That is a strange thing to have to build. It is also a reasonable response to a product decision: vendors added a switch because a switch is cheap and a lawsuit is not, and then made the switch hard enough to find that most buyers never use it. The script does not defeat that decision. It collects on it.

The newest date in the table is 15 November, on the OnePlus 16 row, two days after the release that added that phone. By then the maintainer was already re-checking two sections, because the video app had renamed its personalisation entry in a routine update. That single line tells you how this will go. The scripts work as a maintenance rota, published in public, and their useful life is measured against whatever the vendors ship next quarter.

Related articles