Skip to content
All field notes
Measurement
Published
Nov 1, 2026
Reading time
11 min read

AdMob SSV Callbacks vs Impressions: Reconcile Reward Events

Investigate AdMob rewarded SSV callbacks that differ from impressions: verify signatures, deduplicate reward events and align tests, sources and report dates.

AP
AdProof Research
Independent measurement desk
Separate phone-playback, signature-checked reward callback and impression-report lanes, with duplicate callback envelopes grouped into one reward event.
A received callback, a verified reward event and a reported impression are different records. Conceptual illustration.
In this guide (9 sections)

Counting requests to an AdMob rewarded-ad verification endpoint does not give you an impression counter. Start by separating raw HTTP requests, authenticated distinct reward events, app reward fulfillment and the selected AdMob report. A difference between those records is an investigation queue, not an automatic fraud diagnosis.

The useful first question is: which boundary has a demonstrated mismatch? Preserve the original callback evidence, verify it, group repeated reward transactions and match the app, ad unit, source, environment and reporting period before subtracting totals. Keep anything you cannot correlate explicitly unresolved.

The real question: can server callbacks replace impression reporting?#

An older public AdMob question asks whether finishing a rewarded video counts as one impression. The poster compares calls to their server callback with a lower reported impression count and asks whether this indicates a reward or fraud problem. Replies offer conflicting explanations, including speculation about incomplete views and invalid traffic. The opened thread does not establish the cause in that app.

A separate older Stack Overflow question asks how to verify rewarded ads on a server. Its discussion spans the introduction of SSV, illustrating why an old answer about feature availability needs a current documentation check. These are evergreen implementation questions, not evidence of a new AdMob incident.

This guide checks the diagnostic task against official references available on 1 November 2026. It does not audit either poster's account or identify the cause of their discrepancy.

Related guide

Keep reward callbacks and impressions separate

Record the callback, reward and report boundaries as you investigate the difference.

Use the tracking QA checklist

Opens a practical guide. No email required.

1. Name the records before comparing them#

Use a short measurement contract that both the backend and monetization teams can read. “Views yesterday” is too ambiguous to reproduce.

RecordWhat to countWhat it cannot establish alone
Endpoint trafficIndividual received HTTP requestsDistinct rewards or reported impressions
Verified reward eventsSignature-validated, distinct reward transactions in scopeFinal earnings or independently verified human attention
Fulfilled rewardsDurable app-ledger credits tied to a known eventWhether AdMob counted the corresponding impression
Client observationsImpression and reward callbacks with integration contextSuccessful delivery to your backend
AdMob reportNamed metric, selected dimensions, filters and intervalWhich raw endpoint requests caused the total

Google's User Activity report definitions distinguish impressions, rewarded events and rewarded users. They also define rewarded-event percentage against rewarded impressions. These are different populations; a server access-log count is not another name for any of them.

For example, an HTTP retry can add a request without adding a reward transaction. A reward transaction can be recorded without your app successfully committing its corresponding credit. Conversely, a client may show a reward while its server validation remains pending. Those are hypotheses to test against records, not conclusions drawn from a dashboard gap.

2. Freeze the scope and both clocks#

Save the report export before changing SDK code or backend handling. Record its retrieval time, app, platform, rewarded format, ad unit, ad source, selected metric, timezone and exact start and end boundaries. Save an unmodified copy of the query or report settings.

For backend evidence, retain both the event timestamp and your receipt timestamp. A callback arriving just after midnight need not describe an event that occurred in the same calendar day. Use event time for the intended cohort and receipt time to investigate transport delay; never silently exchange the two.

Google's timezone guidance separates app-related reporting settings from campaign settings and the billing timezone. Read the actual surface you exported. A server's UTC calendar day and an account's local calendar day require an explicit conversion, including daylight-saving transitions where applicable.

Keep the first export as a snapshot and take a later, comparable export if the interval is still developing. Record what changed rather than promising that waiting will reconcile everything. A stable difference still needs diagnosis; a moving difference does not prove that every missing event will eventually appear.

3. Verify callbacks before counting distinct rewards#

The current Android SSV guide describes signed reward notifications and a verification helper. It defines transaction_id per reward-grant event. It also warns against altering the signed content or parameter order, limits key caching to 24 hours and documents delivery retries when the expected HTTP 200 response is absent.

Preserve the original request safely before framework parsing or normalization changes it. Validate through the supported verification procedure for your implementation. Separate verification failure, unknown-key handling, malformed input and a successfully verified transaction in your diagnostics. A signature failure is not permission to grant a reward merely because a request contains a plausible transaction identifier.

Then design the fulfillment operation to be idempotent: repeated handling of the same verified transaction must not create repeated credits. The deduplication record and credit operation should survive a worker retry or crash together. This is a backend design recommendation, not a claim that Google provides your application ledger.

Keep three counters: attempts received, distinct verified transactions, and credits actually committed. Preserve repeat-attempt evidence instead of deleting it from every log. A duplicate-received counter can diagnose transport behavior while a distinct-transaction counter protects the reward ledger.

Do not report every failed verification as a fraudulent user. Encoding changes, verifier configuration and missing key refresh can also require investigation. A valid signature authenticates the notification under the documented scheme; it does not certify a paid impression, identity, attention or final revenue.

4. Check app callbacks without merging their meanings#

The Android rewarded-ad integration guide separately exposes an impression callback and an earned-reward listener. It also scopes callback-order behavior differently for Google-served ads and mediation. Record the installed SDK family and adapter versions; do not combine Legacy and Next-Gen sample APIs by name alone.

For an authorized diagnostic attempt, record the ad instance or local attempt identifier, impression observation, earned-reward observation, dismissal and any show error. Associate those observations with backend evidence only through identifiers your integration actually provides. An invented join on “same user within a few seconds” can conceal ambiguity.

A reward callback is not an instruction to synthesize an extra impression. Dismissal is not a substitute for reward eligibility. A missing client log is also not proof that the underlying event never happened: first establish whether the instrumentation was active, retained and delivered for that build.

If a mediated source behaves differently, keep that source separate. Confirm its callback and SSV configuration before pooling its records with another source. The goal is a reproducible comparison, not a universal ratio across all adapters and rewarded experiences.

5. Separate setup verification, test playback and production reports#

Google's SSV setup instructions provide a callback-URL verification step. The optional user and custom-data fields in that setup interface are for testing; live values are supplied through the SDK. The page also exposes an option for applying SSV to third-party mediation networks.

A successful URL test demonstrates the path that was exercised. It does not establish that a live app populated the intended values, that every source uses the same setup or that a production reward was credited once. Label setup-generated requests separately from app-generated diagnostic events.

Use the official test-ad guidance for development. Google states that labeled test-ad requests, impressions and clicks do not appear in account reports. Mediation requires source-specific test setup and does not reliably supply that same visible label. Do not generate live ad interactions to make a dashboard total move.

Build a small acceptance matrix: callback-URL verification, supported test-ad playback, repeated processing of a controlled test transaction, failed validation and a simulated credit-write interruption. Record which boundary each test covers. Do not claim that the UI verification tool exercises the full SDK-to-reward journey or that every demo ad produces callbacks for your configured production endpoint.

Worked example: reconcile requests without inventing lost revenue#

The following numbers are hypothetical. One backend receives 1,080 endpoint requests during an investigation window. Its raw request count is compared with 900 rewarded impressions in a saved AdMob report.

Backend classificationRequests or distinct events
Setup-test requests, explicitly identified40 requests
Requests failing verification20 requests
Verified requests remaining1,020 requests
Repeat attempts for already-seen verified transactions120 requests
Distinct verified reward transactions900 events

The arithmetic is 1,080 − 40 − 20 = 1,020 verified requests, then 1,020 − 120 = 900 distinct transactions. The 120 repeats are extra attempts, not 120 extra rewards.

Event-time alignment then identifies 30 of those distinct transactions as outside the selected report period. That leaves 870 in-period verified reward events. The reward ledger shows 860 committed credits and ten pending transactions within this same cohort.

The ten pending transactions are an actionable fulfillment queue. The difference between 870 reward events and 900 reported impressions is 30, but the aggregate difference does not identify 30 individual abandoned ads, excluded impressions or fraudulent users. The data sets have different meanings, and this example provides no event-level AdMob report join.

For this defined cohort, committed-credit coverage is 860 / 870, approximately 98.9%. Label its denominator as in-period verified reward transactions. Calling it impression validity, completion rate or payment accuracy would misstate the calculation. Even equality between reward and impression totals would not prove record-level correspondence.

Choose the next action from the demonstrated boundary#

ObservationNext useful action
Repeated valid transaction is credited twiceRepair idempotent fulfillment and reconcile the affected ledger
Verification fails after request normalizationCompare preserved input with the verifier contract
Callback verifies but credit remains pendingTrace the durable queue and credit-write result
Tests appear only in backend evidenceConfirm environment labels before comparing account reports
Difference concentrates in one source or date boundaryRecreate that exact filtered cohort
Only aggregate totals are availableRetain the unresolved remainder; request a scoped provider investigation

Choose reward timing deliberately. Google's SSV guide recommends immediate client-side reward handling with subsequent server validation for user experience, while describing delayed verified-server fulfillment as an alternative where validity is critical and delay is acceptable. Make that choice explicit in the product and backend contract; do not switch users into an unexplained waiting state during a counting investigation.

For monetization questions, continue with AdMob estimated versus finalized earnings. An estimate, reward credit and finalized payment are separate records. If the comparison involves Analytics collection, the AdMob-to-Firebase revenue guide addresses that collection path instead of treating SSV as a revenue feed.

The anonymized streaming-publisher verification case illustrates bounded measurement reconciliation, not an AdMob reward integration. For campaigns measured through AdProof, request a scoped review using the available event export, exact date boundaries and counting policy. Keep AdMob reporting and your reward ledger alongside that evidence; do not assume AdProof supplies SSV verification or app reward fulfillment.

Frequently asked questions#

Should SSV endpoint calls equal reported impressions?

No equality follows from those raw totals. First remove identified setup traffic, verify callbacks, group repeated transactions and align the report scope. Compare distinct reward events with impressions as different measurements, preserving any unresolved gap.

Does a verified callback prove the impression will be paid?

No. It authenticates a reward notification under the verification scheme. It does not establish final earnings or payment. Use the relevant revenue and payment records for that separate question.

Can a retry give the user another reward?

Repeated processing must not produce another credit for the same verified transaction. Make fulfillment idempotent and durable. Keep repeat attempts for transport diagnosis while counting distinct transactions separately.

Does the callback URL test verify the entire app integration?

It verifies the setup path it exercises. It does not prove live SDK values, source coverage, app playback, production reporting or once-only reward fulfillment. Keep those acceptance checks separate and use supported test traffic.

Your next step

Keep reward callbacks and impressions separate

Record the callback, reward and report boundaries as you investigate the difference.

Use the tracking QA checklist

Opens a practical guide. No email required.

Back to all field notes