Push Ad Campaign: Creative, Frequency and Source Playbook

Official boundaries for notification permission, prompt timing and frequency reporting

The WHATWG Notifications standard defines permission states and notification behavior while leaving display details to the user agent and platform. It is a technical specification, not an advertising network, marketing-consent rule or delivery guarantee. Chrome's notification guidance recommends avoiding permission prompts on page load and asking after a user chooses a notification type; that is Chrome product and usability guidance, not universal browser behavior or legal advice. Google Ads explains reach and frequency through its own measurement definitions. Those terms can illustrate why exposure metrics need a named identity scope, but Google advertising figures do not define a push provider's subscription, device or impression counts. Confirm the live provider, browser, operating system, market and consent process. Use platform documentation for the named product only and keep notification permission distinct from any additional permission required for marketing or data processing.

• WHATWG Notifications API - technical permission states and notification behavior, not an ad delivery or consent guarantee
• Chrome notification permission prompt guidance - Chrome usability guidance, not universal browser behavior or legal approval
• Google Ads reach and frequency - Google-specific exposure definitions used only to frame identity-scope questions

Write the push operating workflow

Define the campaign job, eligible audience, permission route, provider, destination, accepted outcome, observation period and largest next exposure. Assign owners for creative approval, technical delivery, suppression, data review and commercial acceptance before funds are released.

Keep subscription, attempted delivery, reported delivery, display, interaction, usable visit and approved action as different events. Document which system observes each transition and what remains unknown. A dashboard total should never silently stand in for a later business result.

Audit the subscription source

Record where and how a person encountered the notification request, what notice accompanied it, which browser or application stored the choice and how the subscription entered the provider. Separate owned subscriptions from third-party inventory.

Do not assume every permission was collected for the present advertiser, message or market. Request source and recency evidence available from the provider. Exclude unclear pools or keep them in a capped observation cell until customer impact and downstream quality are interpretable.

Time permission requests around intent

Ask only after the person can understand the type of notification being offered. Avoid interrupting the first page render with a prompt that lacks context. Store the page, trigger, wording and browser response for the owned flow.

A browser permission answers whether notifications may be shown through that channel. It does not automatically establish permission for every marketing purpose, data combination or recipient. Map additional duties to the actual campaign and jurisdiction under responsible review.

Separate permission states from availability

Treat granted, denied, default and unavailable states according to the current user agent and integration. Test revoked permission, expired endpoints, replaced devices and unsupported browsers. Do not count a stored subscription as permanently reachable inventory.

Report the eligible endpoint population with its retrieval date and definition. Deduplicate only within a documented identity scope. A single person can hold several subscriptions, and one endpoint can become inactive without a visible unsubscribe action in the provider report.

Define the message job

Choose one purpose for each notification: announce a qualified update, restore a known journey, explain a time-bound offer or invite a relevant next step. State the evidence that makes the message useful for the selected cell.

Avoid generic urgency and unsupported personalization. The notification should remain truthful without revealing a private inference on a shared screen. Keep the sender, proposition, material condition and action understandable within the supported notification fields.

Build for platform-dependent display

Document title, body, icon, image, badge, action and destination fields used by the provider, then test them on representative browsers and operating systems. Unsupported fields can disappear or render differently.

Put essential meaning in fields most likely to remain visible and verify truncation. Do not rely on a late action button or image to supply a qualification. Preserve screenshots or device records for the exact creative version released.

Protect destination continuity

Test the path from notification interaction through browser or app opening, redirects, consent state, authentication, landing content, forms and confirmation. Record fallback behavior when an application is absent or a session has expired.

Align message, eligibility, price basis and timing with the destination. Count broken deep links, blank states and unexpected sign-in gates as campaign loss. Provider-reported interaction does not excuse an unusable handoff.

Schedule by service reality

Set delivery windows around the message's usefulness, destination availability, support coverage and market conditions. Document timezone handling, delayed queues and daylight-saving behavior where relevant. Test scheduled and immediate sends separately.

Do not interpret a weak hour as an audience trait when the offer, support or destination was unavailable. Stop expired messages and verify queued notifications cannot appear after the condition ends. Archive the schedule with the result.

Control exposure within identity limits

Find out whether frequency is counted per endpoint, browser, device, account or modeled group. Set caps supported by the provider and review the distribution, not only an average. Preserve repeated-delivery and complaint evidence.

Avoid describing endpoint frequency as exact person-level exposure. A user can have several devices or browsers, while shared devices complicate interpretation. When identity scope changes, begin a new comparison period rather than merging incompatible counts.

Manage suppression as a first-class control

Maintain unsubscribe, revoked permission, customer exclusion, complaint, conversion and internal-test suppression according to the approved purpose. Define refresh timing and which system wins when lists disagree.

Test suppression before launch and after every audience update. Record late-arriving or failed removals. A provider upload success does not prove that every affected endpoint was excluded from already queued delivery.

Track attempted delivery without overclaiming

Record the provider's exact labels for eligible, sent, failed, expired and delivered events. Ask what each label can observe across browser and operating-system boundaries. Keep retries and endpoint cleanup visible.

Do not translate a provider delivery status into screen display, attention or comprehension unless the product supplies supported evidence. Compare rates only when denominators and failure handling are stable across cells.

Treat clicks as navigation evidence

Reconcile reported interactions with destination requests, usable pages and accepted actions. Inspect duplicates, rapid returns, unsupported handoffs and accidental activation. Document the provider's invalid-activity treatment where available.

A high click rate can coexist with poor continuity or qualification. Optimize the first material loss in the path rather than the earliest metric. Keep notification action clicks separate when different buttons lead to different destinations.

Map campaign data responsibilities

List subscription fields, identifiers, message history, interaction records, destination events, recipients, purposes, retention and deletion. Identify which party controls each stage and where data crosses systems or borders.

Use the least data needed for the approved job. Test non-consenting and deleted states. Escalate sensitive data, profiling, children, precise location or cross-context matching instead of assuming browser permission resolves those questions.

Audit source and placement quality

Request the inventory or subscription-source fields the provider can supply. Compare cells by source, browser, device, market, schedule and creative while respecting aggregation limits. Preserve unknown categories.

Use several signals to investigate fraud, coercive permission or unsuitable context. One rapid visit does not prove invalidity, and one clean sample does not certify a source. Apply narrow caps, exclusions or pauses supported by the evidence.

Run a controlled creative test

Hold source pool, destination, schedule, cap and accepted outcome stable while changing one message element. Give variants an allocation rule that supports comparison. Predefine the primary result and technical stop conditions.

Report delivered composition for every cell. If browser or source mix differs materially, narrow the conclusion or rerun. Mark exploratory breakdowns and require reproduction before creating a permanent push-copy rule.

Reconcile complete campaign cost

Include media, creative, subscription acquisition where applicable, integration, landing work, data review, verification, support, refunds and staff monitoring. Keep currency, billing period and outcome maturity aligned.

Compare cost with accepted value rather than the cheapest send or click. A low delivery price can coexist with inactive endpoints, complaints, destination loss or manual qualification. Preserve delayed adjustments and rejected outcomes.

Set customer-impact stop rules

Pause for misleading urgency, hidden conditions, broken destinations, uncontrolled repeat delivery, failed suppression or a material complaint pattern. Name who can stop the campaign and which records must be preserved.

Use a narrow pause when one source or creative is implicated, and a broader stop when permission or data handling is uncertain across the pool. Do not continue spending while the evidence required to protect recipients is missing.

Define restart evidence

Before resuming, verify the corrected creative, destination, cap, suppression and event path on representative devices. Close the incident with owner, timestamps, affected scope, customer handling and residual risk.

A configuration change is not enough. Run a limited restart cell and compare the stage that failed. If the defect recurs or moves downstream, stop again rather than increasing budget to collect more ambiguous evidence.

Read results within the tested cell

Report provider, source pool, permission route, device conditions, market, schedule, message, destination, measurement rules, accepted outcomes and complete cost. Separate observation, inference and decision.

Limit the conclusion to the recorded period and inventory. Do not claim that push works for an industry or that subscribers prefer a message from one campaign. State missing display, identity or attribution evidence plainly.

Close with a push campaign record

Store campaign and creative IDs, settings, eligible and delivered evidence, frequency scope, suppressions, destination versions, incidents, accepted value and unresolved gaps. Link every material claim to current product documentation or internal records.

Choose continue, revise, retest or stop. Name the next exposure, decision owner and rollback condition. A push programme improves through traceable operational corrections, not through a generic promise of instant reach.

Separate owned and network push evidence

Owned notifications begin from a permission flow controlled by the site or app, while network inventory can involve third-party subscription sources and provider contracts. Keep their acquisition history, control, data roles and economics in distinct records.

Do not transfer an owned-subscriber result to a network pool or the reverse. If both are used, compare matched campaign jobs and destinations while preserving source differences. A combined total can hide that one route created reach and another produced accepted value.

Verify icon and sender recognition

Test whether the displayed sender name, domain, application identity, icon and badge let a recipient recognize the source before acting. Rendering can vary by platform, so preserve evidence from the actual device conditions used in the campaign cell.

Do not imitate system warnings, security alerts or another organization. If a provider controls part of the visible identity, document that limitation. Pause combinations that make the advertiser or commercial nature unclear, even when the notification technically delivers.

Manage endpoint health

Track expiration, invalid endpoints, repeated failures and inactive subscriptions using the provider's definitions. Define when endpoints leave eligibility and how the system avoids repeated attempts that add cost or distort rates.

Keep endpoint cleanup separate from marketing suppression. A technically invalid subscription and a valid endpoint that must not receive the campaign require different records. Verify that cleanup jobs and audience refreshes do not restore excluded entries.

Review localization on locked screens

Check how language, sender, dates, price conditions and action text appear when the notification is shown outside the destination context. Use qualified review and consider that another person may see a shared or unlocked screen.

Avoid sensitive personalization and unsupported local urgency. A translated notification must preserve the same qualification and expiry as its destination. Record locale and fallback behavior for devices whose language differs from the campaign setting.

Test notification action branches

Where multiple actions are supported, map each label to a distinct destination and accepted event. Test dismissal, primary action, secondary action and notification-body interaction separately because platforms can expose different behavior.

Do not merge action totals when the buttons imply different intentions. Preserve which fields disappeared on unsupported systems. If a branch cannot retain meaning or measurement, remove it from that device cell instead of guessing its contribution.

Audit queue cancellation

Create a test notification, queue it, then withdraw or expire it using the provider's available controls. Record whether queued, retried and offline endpoints honor the cancellation and how long uncertainty remains.

Use the result to set safe offer cutoffs and emergency procedures. Do not promise that a stopped campaign prevents every already-dispatched notification from appearing. Explain residual delivery risk to the owner before time-sensitive sends.

Push campaign operating matrix

Each decision keeps permission, delivery, display limits, destination evidence and accepted value distinct.

Retained push campaign resources

Existing push, measurement and campaign links and graphics remain below in their established order. They do not verify a current subscription source, permission, device support, delivery, display, frequency, price, user identity or business outcome.

Push ad campaign questions