pixellint

A GA4F Agent Skill That Ships screen_view Is Not Conversion Tracking for Google Ads

Check your payload in the playground →

On November 18, 2026, Google added a release note to What's new in Google Analytics stating that app developers who use the Google Analytics for Firebase SDK can now use Firebase agent skills to empower coding agents to make changes to behavioral and conversion tracking for iOS and Android apps. The same entry limits what works today: initial integration and screen view logging are supported, with support for additional development tasks coming soon. Eight days earlier, on November 10, Firebase product managers Don Kahn and Charlotte Liang published a blog post that introduced agent skills for Google Analytics for Firebase and never used the word conversion. PPC Land reported on November 21 that the marketing sentence and the shipped skill describe different products.

The industry argument is not whether AI writes boilerplate. Publishers and media brands increasingly ship SwiftUI, Jetpack Compose, and Flutter builds where automatic screen reporting breaks because declarative frameworks collapse many screens into one activity or view controller. Firebase documentation has required manual screen_view events in those stacks for years. Agent skills package the setup prompts and linker-flag checks so Cursor, Claude Code, Codex, and Antigravity can add the SDK and log firebase_screen parameters without a human scrolling Gradle and Info.plist docs. That is real engineering relief. It is not the same as wiring the purchase and in_app_purchase events a linked Google Ads account trains on when UA teams import GA4 key events.

The November 11 release on the same help page is a useful neighbor: Google launched a diagnostic in Google Analytics for user-provided data issues, alerting properties that fail to receive UPD or hit quality problems in Admin and data stream panels. Server-side and on-device collection are under separate scrutiny this month. One path promises agents will soon edit conversion tracking; the other path already flags broken user-provided fields. Teams that treat the November 18 headline as shipped conversion instrumentation will debug Ads learning in Events Manager while the app only gained screen_view.

What the release note measures versus what the skill emits

Google's release note uses the abbreviation GA4F and groups behavioral and conversion tracking under one agent-skills capability. Behavioral, in Firebase's own November 10 post, means screenview instrumentation: logging which screens users open so engagement reports populate. Conversion tracking, in Google Analytics for Firebase and linked Google Ads, normally means named events such as purchase, in_app_purchase, sign_up, or custom key events you mark for bidding and import.

Firebase agent skills documentation, last updated November 18, lists thirteen skills. Only firebase-analytics assists with getting started and measuring screenviews for Google Analytics in iOS and Android apps. The table does not list purchase logging, Firebase Analytics ecommerce helpers, or Google Ads app conversion linker tasks. Google's release note explicitly defers those to additional development tasks coming soon without a date.

PPC Land compared the two Google documents side by side. The release note promises behavioral and conversion tracking edits through agents. The blog post promises getting started and adding screenview instrumentation. What works now, in both, is initial integration and screen view logging. Conversion tracking appears only in the release note vocabulary.

For a publisher, the measurement population that changes this week is apps whose engineering teams adopt agent-written screen_view calls on Compose, SwiftUI, or Flutter. The population that does not change is checkout completion, subscription renewals, and ad-supported unlock flows that still need explicit logEvent code or ecommerce APIs. Google's Firebase support release log for November 3, 2026 also notes the firebase-analytics skill availability for getting started and measuring screenviews, aligning the platform release calendar with the Analytics help page.

Why declarative UI broke automatic screens in the first place

The Firebase SDK logs screen views automatically by default through UIViewController and Activity class names, including method swizzling on Apple platforms. Declarative UI frameworks draw many logical screens inside one container, so automatic logging often records a single class while the user navigates across destinations inside it. Firebase screen view documentation therefore requires manual AnalyticsEventScreenView or FirebaseAnalytics.Event.SCREEN_VIEW with firebase_screen and firebase_screen_class parameters.

The November 10 post names Jetpack Compose, SwiftUI, Flutter, and React Native as reasons manual logging returned. Games and single-activity apps were already in that bucket. The agent skill's second suggested prompt, help me instrument my app to log screenviews using Google Analytics for Firebase, targets that gap. It does not generate purchase payloads with value and currency parameters.

Screen views feed engagement and funnel exploration in GA4. They are not interchangeable with Google Ads conversion actions unless you deliberately map them, and most performance teams do not bid on raw screen_view. When automatic reporting was wrong, dashboards under-counted engagement. When agents fix screen_view but nobody touches purchase, dashboards look healthier while Ads optimization still starves.

The Firebase post also mentions finicky setup: the -ObjC linker flag on iOS, BoM coordinates on Android, multiple build variants, and walking developers through Google Analytics account creation when the agent cannot complete console steps. Those are legitimate failure modes agents reduce. They are setup failures, not post-setup conversion gaps.

Manual screen_view on declarative stacks

Agent skills documented in November 2026 focus on the first block. Google Ads key events and linked import still depend on explicit purchase or in_app_purchase instrumentation the release note labels conversion tracking but does not ship in the skill table.

// SwiftUI pattern per Firebase screen view documentation
Analytics.logEvent(AnalyticsEventScreenView, parameters: [
  AnalyticsParameterScreenName: "Article",
  AnalyticsParameterScreenClass: "ArticleView"
])

// Purchase still requires a separate logEvent, not covered by firebase-analytics skill today
Analytics.logEvent(AnalyticsEventPurchase, parameters: [
  AnalyticsParameterCurrency: "USD",
  AnalyticsParameterValue: 4.99
])

A coding agent that stops after the screen_view block has done exactly what Google documented. The production failure appears later, when media operations sees key event volume flat in the linked property while product analytics shows rising screen depth. The advertiser blames the buy; the publisher blames the network; neither side diffed the agent's pull request against the Ads conversion action list.

Firebase screen view documentation also lists switches to disable automatic screen reporting when you take over manually. Mixed builds that leave automatic reporting on and add manual events can duplicate or mislabel screens. Agents that only read the skill markdown may not reconcile Info.plist and AndroidManifest.xml flags with your existing analytics module.

Linked Google Ads accounts import GA4 events you mark as conversions. If purchase never fires, enhanced conversions for app streams and offline imports have nothing to match against click ids. The November 11 user-provided data diagnostic is a parallel warning: even when events exist, UPD quality problems surface in Admin. Empty purchase streams fail earlier and quieter.

What a solutions engineer at a publisher can see in the sandbox

A solutions engineer at a publisher sits between the brand's app release train and the advertiser's conversion definitions. You can run the vendor sample in a sandbox Firebase project, load the firebase-analytics skill in Cursor or Codex, and watch the agent add dependencies, initialization, and screen_view hooks on a staging build. You do not own the advertiser's production dataset or their Google Ads conversion action IDs.

The publisher owns the binary where tags must fire and the consent surfaces that gate analytics on iOS and Android. The advertiser owns which GA4 events count as conversions and whether a linked Ads account optimizes on purchase, first_open, or a custom key event. When an engineering lead pastes the November 18 release note into sprint planning as conversion tracking done, your job is to prove which events actually reach the linked property in DebugView versus which fields the agent sample never populated.

Sandbox versus production diverges on app_instance_id, stream linkage, and build variants. Skills documentation warns about multiple flavors and directory layout; production may ship a flavor without the agent-instrumented target. Advertiser testers often validate on staging streams that are not linked to the live Ads account. Green DebugView on staging screen_view is not proof the production purchase stream trains bidding.

Agency trading desks will ask for Screens per session while performance leads ask for cost per in_app_purchase. The same Firebase project can satisfy both only when both event types exist and key events are marked consistently. Agent-assisted screen_view fixes the first question while leaving the second unanswered until someone writes purchase code or enables recommended ecommerce events.

This week’s warehouse value scoring story on Databricks CustomerLake Conversions is the server-side mirror: platforms optimize on the event schema they receive, not on the narrative in a launch note. App agents that instrument engagement without conversion events recreate that gap inside GA4F before data ever leaves the device.

Checklist: sandbox sample versus production conversion

Why the gap persists after agents

Google benefits when more apps install GA4F cleanly because linked Ads and BigQuery exports grow. Agent skills lower the setup tax for screen reporting on modern UI stacks. Conversion instrumentation remains sensitive: value parameters, refund handling, subscription renewal semantics, and SKAdNetwork or Privacy Sandbox adjacency on iOS all need business rules agents cannot infer from a single prompt.

Trade press this week framed the release as agents wiring analytics into mobile apps. That headline is directionally right for SDK setup and screen_view. It is wrong for performance teams who read conversion tracking literally. Until additional development tasks ship, the honest capability table is two rows: integration and screen views.

Pixellint is independent of Google and Firebase. Linting a Conversions API JSON does not validate Firebase logEvent payloads, but the same discipline applies: the collector optimizes on the events that arrive with valid names, clocks, and value fields. Screen_view depth without purchase is engagement reporting, not conversion proof.

When Google expands firebase-analytics to conversion tasks, diff the new skill version against your production tag map before you tell advertisers the agent closed the loop. Until then, treat the November 18 sentence as intent, not shipment.

Sources

Contract pages

The dated argument is above. These pages are the field lists.

Read event volume and sampling Docs