Buying push ads without inheriting stale opt-in lists

What push ads are, and why the format outlived three rounds of browser intervention aimed at ending it

A push ad rides on the channel a news site uses to tell you a story broke. The site registered a service worker in your browser, you clicked Allow on a prompt, and from then on a server can hand the browser a message whenever it likes, tab open or not. Networks lease that channel from publishers. When a publisher has nothing worth sending, the empty slots are filled with push ads.

Three interventions were supposed to end this. Firefox 72, January 2020, stopped showing the permission dialog until the visitor clicked something, and Chrome 80 followed in February with what Google called the quieter permission UI, which hides prompts from people who habitually decline them and on sites where almost nobody accepts. Neither killed the format. I went looking for a plain account of what those two changes did to opt-in volumes and found one on push-ads.io, which is where the rest of this page started.

Quiet prompt, loud numbers

Google's engineers published their reasoning in 2021. Across more than 40 million Chrome users, 74 percent of all permission prompts were notification requests, and they were also the ones least likely to be granted. The quiet UI cut unnecessary prompt actions by up to 30 percent. Grant rates fell by less than 5 percent, which is the figure to read twice, because it means the people who wanted notifications kept accepting them and the lists kept growing. Quietly.

Chrome 84 then auto-enrolled sites with deceptive permission patterns into the quiet UI. Chrome 86, October 2020, went after content: Google's crawler now subscribes to sites, evaluates what arrives using Safe Browsing, warns registered owners by email 30 days ahead, and blocks the prompt on origins with a history of abusive messages if nothing changes. That machinery still runs. Networks that survived it did so by policing creatives before delivery rather than after complaints, the first thing to ask any of them about.

Where push ads subscribers come from

Nobody subscribes to receive advertising. People subscribe to a weather page, a torrent tracker, a live score site, and the network buys the quiet hours in between. So the quality of push ads is inherited from whatever site collected the opt-in, and from how long ago that happened. Last week's subscriber remembers clicking Allow. Vividly.

The decay is not gentle. Browsers drop endpoints when a profile is reset, a subscription whose deliveries keep failing eventually dies at the vendor, and since Android 13 a browser update can trigger a fresh prompt that the same person now declines. What remains on an aged list is the residue of people who never clean anything up.

Reading a subscriber age report

Most networks expose subscriber age as a targeting dimension, in days or in bands. Fresh traffic under seven days old costs the most per click and holds attention best on offers that ask for a real decision, while anything past ninety days is bought for volume plays where the tap alone earns the money. The same arithmetic applies to push notification ads sold per delivery, where an old list inflates the denominator without adding anyone who will act. Age first, price second. Always.

How push ads are priced, bid on and capped, and which of the three the panel tends to hide

Two models dominate push ads pricing. Cost per click makes the network carry the risk of a weak creative, since an ignored message costs nothing, and cost per thousand deliveries moves that risk to the advertiser, and is cheaper per message. Some feeds hide a third model behind an automated bidder that lowers the bid on sources beating target. Convenient. Until you notice it also raises the bid where you were already winning.

Reading the bid rules matters more here than with almost any other format. The clearest set I have found is on the page where push ads are sold per click, with the cap logic in a paragraph rather than a help centre. Read it before the deposit, then again after the first invoice.

Frequency caps are the other half of the price. A subscriber hit every two hours by four networks stops reading, and the cost of reaching that person collapses along with the value. I keep a note on this here on The Albany Notebook, because a figure nobody dated does the same damage to a subscriber list as to a stadium. The useful question is whether the network applies it per advertiser or across everyone bidding on that person, because a cap that only counts your own messages protects your budget while leaving the subscriber just as exhausted, and just as unlikely to tap.

Delivered against displayed

A message can be delivered to an endpoint and never rendered, because the phone sits in a Focus mode, the notification shade is full, or the desktop browser is closed. CPM feeds bill on delivery. If a panel shows only impressions, ask what the word means to them, in writing, before spending anything.

Creative limits that decide whether push ads get tapped

The canvas is tiny. A title, a line of body text, an icon, and sometimes a larger image, all of which the operating system truncates without asking anyone. Android shows more of a long title than desktop Windows does, and a locked iPhone shows the least. The creative that works is the one that survives truncation and still reads as a message rather than an advert, and that constraint, not clever copy, is what separates push ads that earn a tap from ones that get swiped away.

Icons carry the weight. A system-style bell or a shield reads as an alert and gets opened, the pattern Chrome 86 was built to punish, so a sensible network rejects it at vetting. Product photography at icon size becomes a blur. Unreadable. What survives vetting and still gets tapped is a flat, single-colour icon with one recognisable shape, next to a title that states an outcome in plain words.

Landing speed finishes the job. The tap happens on a phone, often on mobile data, the page behind it is judged in the first second, and anyone buying push notification ads for a slow page is paying for clicks that end before it renders, though the report will show every one. Test the landing page on a throttled connection before touching the bid. Most of a campaign's taps can vanish before the first paint.

In-page push ads after iOS 16.4

For years iPhones were invisible here. Safari on iOS had no Web Push, so a subscriber list was Android and desktop by definition. Apple changed that in March 2023 with iOS and iPadOS 16.4, which let web apps added to the Home Screen request notification permission after a tap, and the catch is the Home Screen step, which very few people take. So iOS remains a thin slice of push ads inventory even now.

The industry's answer was to stop depending on the permission dialog. In-page push renders a notification-shaped element inside the web page while the visitor is on it, with no service worker, no opt-in, no browser policy to fall foul of, and it reaches iPhones, which browser push still barely does. It also reaches only people currently on a page, which is a different audience from someone glancing at a lock screen at seven in the morning.

Two placements, two audiences

Treat them as separate line items. Classic browser push gives you a subscriber list you can reach again, with the decay described above, while in-page push gives you fresh eyes on every session and no list at all. When a push ad network reports both under one figure, split them yourself before deciding what worked. The same creative behaves differently on a lock screen and inside a page someone chose to open. Different audiences, different bids.

None of this makes the format complicated. Legible. That is the better word, because the browser rules are public, the subscriber age is reported, the pricing model is a checkbox, and the creative canvas is smaller than a business card. Most money wasted on push ads is wasted by treating the format as a cheaper banner, when it is closer to buying a place in someone's inbox, one they agreed to and have forgotten.