Why deep links need their own test
Most campaign QA checks that a link works by tapping it once, on the tester's own phone, with the app installed and the tester signed in. That is the one case least likely to fail. The cases that fail are the ones a tester rarely sees: a person who has never installed the app, a person who is signed out, an item that sold out last week, a screen that was renamed in the latest release, an older version that never had the screen, or a tap inside a social app's own browser. That last case decides most creator campaigns, which is why the guide to influencers for app installs has every creator link tested before the post goes live.
Each failure is silent. Universal links on iOS and App Links on Android are ordinary web addresses that open the app when the operating system can confirm the app is allowed to, and the web page when it cannot. That fallback is a feature for people without the app and a hidden fault for people with it, because a verification problem looks like a working link that happens to open the browser. The deep links guide also notes that some in-app browsers do not honour universal links, that custom URL schemes have no web fallback at all, and that deferred links depend on a link service or measurement partner whose matching can be limited by platform privacy rules.
The cost lands in the numbers that judge the campaign. People who reach the wrong screen activate less, so a routing bug reads as a weak audience or poor creative. The measurement plan for app installs puts deep-link tests in the release QA list for exactly that reason, and the growth loops page for app installs treats the link as the joint of every invite loop, with one named owner. This tutorial is the test that owner runs.
Before you start
Gather four things. First, the list of links the campaign will send, from the brief or the UTM link request that produced them, with the channel each one runs in. Second, the app's route table, the list that maps each link path to a screen, from whoever maintains routing; if there is no written table, the first step creates one. Third, devices: at least one of each platform, with the current release, a way to install an older live version, and a device or account on which the app has never been installed. Fourth, access to the analytics debug view, or the link service's test tools, so you can see what the app recorded on each open.
Agree in advance what counts as a pass. For an installed, signed-in user it is the intended screen with the right item loaded. For a person without the app it is a real, useful web page at the same address, or the store and then the intended screen if the link is meant to carry them through an install. For edge cases it is the designed fallback, written down before testing so that nobody argues afterwards about whether a login screen was acceptable. Where the route tableThe app's map from each deep link path to the screen it opens, including what happens for logged-out users, missing items and renamed screens. already records the fallback for a signed-out person or a missing item, those rows become the pass conditions for the edge-case tests.
The steps
List every link the campaign will send
Write one row per distinct link: the channel (ad, email, push, SMS, social post, printed code), the destination the message promises, and the full address including campaign parameters. Links that differ only in parameters can share a row for routing but need their own row for the parameter check later. Include links that people will forward, such as invite and share links, because they travel into contexts the campaign never planned for.
Write the expected screen and web fallback for each
Against each link, record the screen it should open in the app, the item or filter it should load, and the web page a person without the app should see. If the web page would be a dead end, fix that before testing anything else: the deep links guide is clear that the page behind a universal link or App Link should be a real page a person can use. This column is the pass condition for every later step.
Build the test matrix
Cross each link with the platforms, the app states and the contexts in which people will tap it. Not every cell needs a test for every link, but every link needs the cells its channel will actually produce: an email link needs email clients, a social ad needs the social app's in-app browser, a printed code needs the camera. The table after these steps shows a starting grid.
Check the domain association first
Universal links and App Links depend on the app having verified the domain, through an association file on iOS and a digital asset links file on Android, as the deep links guide describes. If verification is broken, every installed-user test will open the browser and look like a routing problem. Confirm that a plain link on the domain opens the app from the system browser on each platform before working through the matrix. If it does not, stop and hand it to the developer who owns the files.
Test installed users on the current version
Signed in, current release, each context in the matrix. Record whether the app opened, which screen it showed, and whether the right item loaded. Where a context opens the web page instead, note it: the deep links guide explains that some in-app browsers behave differently from the system browser, so the result may be expected rather than a fault, and the fix may be on the web page rather than in the app.
Test logged-out, missing-item and renamed-screen cases
These are the failures the push notifications and in-app messaging guide lists. Sign out and tap: the app should return the person to the intended screen after sign-in, not leave them on the home screen. Point a link at an item that has been removed or sold out: the app should show a sensible alternative. Use a link built for a screen that was renamed or moved: the route table should still resolve it.
Test older live versions
People do not all update at once. Install the oldest version still in meaningful use, as the analytics shows it, and repeat the installed-user tests for any link that points at a screen added or changed recently. A link that works only on the newest version needs a fallback for older ones, or the campaign needs to wait.
Test new installs through the deferred link
On a device or account that has never had the app, tap the link, follow it to the store, install, and open the app for the first time. The intended screen should open, and any context the link carried, such as who invited the person or which offer they tapped, should arrive with it. The deep links guide notes that matching can be limited by platform rules, so also check the designed default for the case where the destination is not recovered. Log whether context arrived on the first-open event, as the growth loops page recommends, so the failure rate can be read in production as well as in the test.
Confirm campaign parameters reach analytics
Tagged links only help if the tags survive the jump into the app. Google's documentation on URL builders describes the campaign parameters a tagged link carries; the deep links guide describes passing them through the link and recording them on the session or the first event as event parametersA named value sent with an analytics event that describes it, such as a page location, form name or item price.. Open the analytics debug view, tap each tagged link and check that source, medium and campaign appear with the names the naming conventions guide sets. For paid installs the source comes from the measurement partnerAn independent service that attributes app installs and in-app events across ad networks. instead, so check that its test install was attributed to the test link.
Record results and retest every release
Keep the matrix with a result in every tested cell, the device and version used, the date and the person who ran it. File each failure with the link, the cell and the expected result. Note fixes and routing changes in the measurement change log, so a later dip in activation can be checked against it. Before the next campaign, or after the next release, repeat the cells that could have changed and a short pass through the rest.
A starting grid for the matrix. Add rows for the contexts your campaigns use and drop the ones they never touch.
| Context | Installed, signed in | Installed, signed out | Older live version | Not installed |
|---|---|---|---|---|
| System browser | Intended screen, item loaded | Sign-in, then intended screen | Intended screen or designed fallback | Real web page at the same address |
| Email client | Same as above | Same as above | Same as above | Web page, or store then screen if deferred |
| Social app's in-app browser | App or web page; note which | Same | Same | Web page |
| SMS or messaging app | Intended screen | Sign-in, then intended screen | Fallback | Web page |
| Camera scanning a printed code | Intended screen | Sign-in, then intended screen | Fallback | Web page, or store then screen if deferred |
| Push notification or in-app message | Intended screen | Sign-in, then intended screen | Fallback | Not applicable |
A bookshop app's pre-campaign link test (illustrative)
Tamblequin Books tests its gift-guide links
Tamblequin Books, an invented bookshop with an app, plans a gift-guide campaign with 6 distinct links: 2 paid social ads, an email, a push, an SMS for customers who opted in and a code printed on in-store bookmarks. Its matrix has 31 cells once irrelevant combinations are dropped, and two people work through it in an afternoon. The domain association passes on both platforms. Installed, signed-in users land on the right screens in every context except one social app's in-app browser, which opens the web page; the team accepts that because the web page carries a clear store link. Three failures matter. Signed-out users who tap the email link are sent to sign-in and then dropped on the home screen, losing the gift guide. Version 4.2, still used by about 9% of active users, has no gift-card screen, so one link fails there. And on a fresh Android device the deferred link opens the gift guide but loses the offer code. The developer fixes the sign-in return and the offer code within the week, the team swaps the gift-card link for the guide page on older versions, and the retest passes all 31 cells before the campaign starts.
The checks here sit alongside the store, measurement and payback checks in the app install campaign launch checklist, which runs before any paid install campaign goes live. How each link type works, and what the events that record a tap should look like, is the subject of the deep links and app analytics guide, and the mobile measurement hub shows where links sit beside attribution.
Common mistakes
- Testing once, on your own phone. The installed, signed-in tester is the case least likely to fail. Work through the matrix.
- Treating a browser fallback as a pass. For someone with the app, a link that opens the web page usually means the domain association or the tap context failed. Check the association first.
- Skipping signed-out users. A link that drops people on the home screen after sign-in throws away the reason they tapped.
- Testing only the newest version. Links to new or renamed screens fail on the older versions many people still run.
- Never testing a fresh install. Deferred links are the path every new user takes and the one testers almost never see.
- Losing campaign parameters at the app boundary. Behaviour cannot be split by source if the tags do not arrive. Check them in the debug view.
- No written pass condition. Without the expected screen and fallback recorded first, every result becomes an argument.
- Keeping no record. Without the last run's matrix, every retest starts from nothing and nobody can tell a new failure from an old one.
