Push Traffic with AI: Push Notification Ads Guide (2026)

2026 Update:

Classic web push is not the 2018 gold rush. Chrome is actively starving spammy notification streams.

In October 2025 Chrome started automatically removing notification permission from sites you have not used, when those sites send a lot of alerts. Google said less than 1% of notifications get any interaction (Chrome blog). In January 2026 they added Push API rate limits for sites that spam without engagement (Chrome Developers). Google later said Android protections blocked over 7 billion unwanted notifications a day in Q1 (Google Security).

What is left: tighter classic push, in-page push that never needed a subscription, and operators who refresh creatives every week.

This is an affiliate / CPA channel. A dentist should not buy push. They should buy Google Ads and local SEO.

What push traffic is

A small icon, a 30-50 character title, a short description. It looks like a system alert. On classic push, someone opted in on a website months ago and forgot. On in-page push, it is an overlay on a site they are browsing. No permission.

You interrupt. You have seconds. Simple offer or you lose.

Classic vs in-page

Classic (subscription) push

• Desktop + Android
• Real opt-in at some point
• iOS Safari is still a weak story for web push
• Inventory is shrinking because browsers revoke and throttle

In-page push

• Overlay. Works on iOS Safari
• No subscription
• Growing because classic is getting policed
• Feels spammy. Convert or it is just noise

Start: classic on Android if you can get volume. Add in-page for iOS and extra scale. Track them as separate formats.

Networks worth using

Pick one. Learn the source report. Then expand.

PropellerAds - volume. Push, in-page, pop, interstitial. CPA Goal if you post back conversions. You will blacklist.

Tokens (confirm in their current docs): ${zoneid}, ${device}, ${os}, ${browser}, ${geo}, ${creative}

RichAds - cleaner, less scale. AdScore. Tiers. Better if you already hate remnant traffic.

Adsterra - multi-format, direct publishers plus junk. Fine as a second source.

Admaven - cheaper tests. Volume comes in spurts.

ZeroPark - push plus domain traffic. Handy if you already live in Voluum.

I am not listing every CIS-market Telegram network. If you cannot verify they pay and they filter, skip them.

Verticals that fit the format

Works: sweepstakes, utilities (VPN, cleaner), finance offers that are allowed in-geo, flash ecom, game installs, dating trials.

Does not work: B2B, anything that needs a 20-minute explanation, local services, high-ticket with no intermediate step.

Push is interruption. If the user has to “think about it,” use search or native.

Local businesses: a push that says “dentists near you” is how you buy angry clicks. Use Maps and Search.

Creatives: this is the whole job

Push is a volume game. Agencies used to win by shipping 50 headlines a week. You can do that with Claude in an hour. The market still decides.

Headline prompt:

Write 20 push titles for [OFFER].
Audience: [WHO]. Benefit: [WHAT]. Angle: [urgency | curiosity | social proof].
30-50 characters. No fake "you won." No medical scare claims.
Format: 1. [title] - [count] - [angle]

Run it three times (urgency, curiosity, proof). Keep 20. That is a week of tests.

Description prompt: pair your top 5 titles with 40-90 character supporting lines. Do not repeat the title. Soft CTA.

Icon: 192x192. Face, simple object, product. No tiny text. It has to read at the size of a fingernail.

Match the landing page. If the push says “your results are ready,” the page better open on results, not a generic home page. Disconnect kills CR.

For pages: landing page guide.

Patterns that still click

• Personalization tokens when the network has them ({city})
• Time (“expires tonight”) only if the page is actually timed
• Curiosity without a lie
• Notification-style copy that is not impersonating a bank or OS alert (that is how you get banned)

Do not impersonate WhatsApp, Google, or a carrier. Networks will cut you. So will browsers.

Campaign setup (two weeks, not a quarter)

Week 1

1. Ask your AM which push offers convert in your geo
2. Spy (Adplexity) + the research prompt
3. 50 titles, 10 icons, mix them
4. Tracker: Voluum, RedTrack, or BeMob. Capture source, creative, device, OS, browser, country. Tracking guide.
5. Landing page that repeats the push promise

Week 2

1. One network. Android + your geo. $50-100/day. Frequency cap 1-3 / 24h.
2. Hands off for 3-4 days
3. First AI pass: blacklist, note winning titles, do not rebuild the whole campaign
4. Then the weekly loop: new titles from winners, kill dead sources, optional whitelist campaign at a higher bid

Postbacks: network -> tracker (clicks), affiliate network -> tracker (conversions), tracker -> push network (so CPA Goal has something to learn). Skip that last hop and “smart bidding” is theater.

What Chrome’s rules change in your buying

Classic push inventory is a rented list of forgotten opt-ins. When Chrome revokes permission on sites with high send volume and low engagement, those lists shrink. Rate limits then cap how fast a domain can blast the remaining users (Chrome Developers).

Practical fallout:

• Do not build a business that assumes last year’s subscriber counts
• In-page push is not “worse classic.” It is a different animal (overlay, no permission, works on iOS)
• Frequency caps matter more, not less. Hitting a shrinking pool harder just trains Chrome to treat the publisher as abusive
• If CTR is fine and volume falls off a cliff, it may be the browser, not your creative. Check format split in the tracker before you rebuild 50 icons

I still run classic on Android because the click is often cheaper and the user at least clicked “Allow” once. I do not pretend that list is an asset I own.

Bid notes

CPA Goal / target CPA is fine after you have conversions feeding back. Before that, it optimizes on noise.

Manual CPC on a whitelist of zones is how I extract a winner. Prospecting stays on a lower bid or on the network’s auto bid so it can find new zones.

If you raise bids on the prospecting campaign, you often buy worse remnant, not more of the good stuff. Split the campaigns.

Optimization prompts that earn their keep

Sources (Monday):

Target CPA $[X]. Data: source, imps, clicks, conv, spend, revenue.
1. Blacklist (spend > 3x CPA, 0 conv)
2. Whitelist (profit, 10+ conv)
3. Need more data
4. Shared traits of winners

Creatives (after 4-5 days):

Headline, description, icon, CTR, CR, CPA.
What patterns win? Give me 10 new titles in that pattern.

Fatigue: when CTR and CR slide for a week, ask whether it is fatigue vs a bad source mix. Fresh titles beat “raise the bid.”

Fraud: high CTR, zero conversions, instant “converts.” Kill those sources. Do not wait for the network to care.

Scaling without blowing the winner

• +20-30% budget at a time
• Whitelist campaign for proven zone IDs, bid up 20-30%
• Clone geos (US -> UK/CA/AU, DE -> AT/CH). Translate, do not run English in Brazil
• Same winners on a second network. Different sources, different results
• New titles every week. Monthly angle change. Fatigue is the tax you pay for cheap attention

Common ways to lose

1. Ten titles, not fifty
2. Ignoring the source report
3. 10x budget overnight
4. A complicated offer
5. No frequency cap
6. Optimizing CTR instead of CPA
7. Buying push for a local service brand
8. Sending the same English title to BR, DE, and US
9. No postback, then complaining “smart bid doesn’t work”

Icon and copy details people skip

Icons at 192x192 cannot hold a sentence. If you need text, it does not belong in the icon.

Title truncates on some Android skins. Put the hook in the first 30 characters, not the clever twist at character 48.

If the network supports large image / banner, test it as a separate creative ID. Do not assume it lifts CR. Sometimes it just looks more like an ad and CTR drops.

Landing page language equals push language. Always. A Portuguese push onto an English page is how you buy accidental clicks.

iOS is a separate campaign

Classic web push on iOS is still a mess. If you need iPhone users, build an in-page campaign with iOS-only targeting and copy that does not assume they “subscribed.” The overlay is the ad. The page has to load fast on Safari.

Mixing iOS in-page with Android classic in one campaign poisons the report. You will blacklist good Android zones because the blended CPA looks ugly.

Creative calendar (minimum)

• Mon: source blacklist from last 7 days
• Tue: 10 new titles off the winning pattern
• Thu: kill titles with spend and no conversions
• Whenever volume dies: check classic vs in-page split before you blame the offer

That is two hours a week if the tracker is clean. If you cannot spare two hours, do not buy push.

Is push right for you?

Yes if you want cheap, controllable traffic, you will work the source report weekly, and the offer is one tap.

No if you want set-and-forget, precise job-title targeting, or booked dental appointments.

The classroom value is real. Source hygiene and creative volume show up on every other channel.

Next step

This week: PropellerAds or RichAds. One offer. 50 titles. Tracker on.

Next week: $50-100/day. Do not “optimize” for four days.

Week 3: blacklist, clone winners, new titles.

Local business: Google Ads and local SEO.

Related: paid traffic map | pop traffic | native ads | tracking

Have questions? Contact me.