Google’s new Chrome spam filter for Web Push: what’s changing and how to stay deliverable

Modified on Fri, 29 Aug at 2:40 AM

As Google Chrome’s new spam filter for web push notifications launches on Android, the marketing landscape is shifting dramatically. At NotifyVisitors, we want to empower you with the strategies needed to protect engagement rates and ensure every message counts in this new era.

What is changing?

Chrome’s on-device ML-based spam filter now pre-screens web push notifications for signs of spam, misleading, or low-value content. Notifications flagged by the filter don’t show up right away: instead, users see a warning prompt with two options:

  • Unsubscribe (quit your subscriber list with a tap)
  • See notification (user has to take an extra action to reveal it)

When users choose to view the notification, they’re shown further options:

  • Always allow (whitelist your site, so your future notifications are not blocked)
  • Unsubscribe (stop all future notification until user resubscribe)

For marketers, this means every blocked notification risks immediate subscriber loss and rising invisibility.

The notification content is not sent to Google and happens on the device, using the title, body, and action button text. Currently, the feature applies to Android only.

Chrome also keeps other protections in place, like revoking notification permissions for abusive sites via Safe Browsing and one-tap unsubscribe on Android.

What it means for your KPIs

  • CTR & visibility: Warning friction lowers opens and clicks on Android if your notification triggers the filter.
  • Churn risk: The warning UI puts Unsubscribe front-and-center, so irrelevant or clickbaity messages can accelerate list loss.
  • Attribution gaps: There’s no “flagged” callback; impressions may still count even when a warning hides the message. Diagnose via CTR drops and unsubscribe spikes.

What tends to trigger the filter

From field guidance and vendor testing, the model is sensitive to three buckets—content, history, and repetition:

  1. Content quality
    Clickbait (“FREE,” “WIN,” “ACT NOW”), excessive urgency, emoji stuffing, and vague/misleading claims are risky. Avoid notifications that resemble system warnings.
  2. User history with your site
    If a user ignores/dismisses your pushes repeatedly, your future messages to that user are more likely to be warned or hidden. The system adapts per user.
  3. Repetition
    Re-sending identical titles/body to the same users increases the chance of warnings—rotate copy and tailor by segment.

How to stay ahead (your playbook)

  1. Craft clear, honest content

  • Lead with value in the first 6–8 words.
  • Be specific: “Order #7842 out for delivery—track it” beats “Hurry! Big update now!”.
  • Keep emojis minimal; avoid FREE/URGENT/ACT NOW unless truly necessary and contextual.
  1. Don’t impersonate system alerts

Never mimic OS/browser/security language (e.g., “Warning! Your phone is at risk”). Chrome explicitly penalizes this pattern.

  1. Personalize and vary messaging

  • Segment by intent and lifecycle (viewed product, added to cart, category electronics, etc.).
  • Rotate titles and avoid identical repeats to the same user.
  1. Test rigorously on Android

  • Send every new template to real Android test devices first.
  • If a warning appears, rewrite and retest (removing a single bait word like “FREE” has flipped outcomes in A/B tests).
  1. Monitor analytics

  • Use NotifyVisitors analytics to watch unsubscribe rates and click-through declines. A spike can be a sign your campaigns are being flagged.
  • Act fast: analyze which content gets flagged and adjust strategy promptly.
  • Since there’s no real-time “blocked” signal and impressions can still log, use trend diagnostics to catch issues quickly.

NotifyVisitors implementation tips

Targeting & frequency hygiene

  • Segments that matter: last viewed category, price-sensitivity, cart status, churn risk.
  • Frequency caps: e.g., ≤2/day, ≤5/week per user; add quiet hours (DND).

Copy guardrails (use as a pre-send lint)

  • No system-like phrasing.
  • Minimize ALL-CAPS/bait words; use one meaningful emoji max.
  • Put the concrete benefit or context up front.

Good vs risky examples

ScenarioSafer copyRiskier copy
Cart reminder“Your cart’s ready—extra 10% auto-applied”“URGENT! LAST CHANCE! BUY NOW!!!”
Content alert“New episode S3E4 is live—watch”“System alert: Update required to view”
Shipping“Order #7842 out for delivery—track it”“FREE gift inside! Click to claim”

(Examples aligned with Chrome’s focus on deceptive/spammy patterns and vendor guidance.)

Experimentation

  • A/B test titles, body content, and action buttons on a small Android user base before a full rollout.
  • Also, even during the full rollout, send in small batches, check what happens, improve, then keep sending.

Governance

Keep domains clean (HTTPS, no deceptive landers). Chrome/Safe Browsing can revoke permissions for abusive behavior.

FAQs (Frequently Asked Questions)

  Which devices are affected?  
For now, just Chrome on Android (since May 8, 2025). Expansion to other platforms may be evaluated later.
  Can we detect blocked/flagged sends?  
There’s no live callback—the decision happens on the user’s device. Watch CTR and unsubscribe trends for clues. Note: impressions can still be counted even if a warning hid the message.
  How are unsubscribes tracked from the warning?  
You won’t see it right away. The unsubscribe is picked up on the user’s next visit to your site.
  Does a flag hurt the whole site/domain?  
No, it’s per user. If someone keeps ignoring or unsubscribing, it gets harder to reach that person in the future, not everyone.

Was this article helpful?

That’s Great!

Thank you for your feedback

Sorry! We couldn't be helpful

Thank you for your feedback

Let us know how can we improve this article!

Select at least one of the reasons
CAPTCHA verification is required.

Feedback sent

We appreciate your effort and will try to fix the article