Invalid traffic in push ads.

A push ad is not shown on a page the visitor chose to read. It is delivered to a browser that, at some point, accepted a notification prompt on some website, and it arrives later, on whatever that person, or that device, is doing at the time. Everything about the quality of the traffic is decided upstream, when the subscription was collected.

That makes push unusual in two ways. First, the buyer never sees the site where the subscriber was acquired; the network, or the partner feeding the network, did the collecting. Second, one subscription is sold many times: a subscriber who accepted one prompt can receive notifications from many advertisers for as long as the subscription lasts. A list built carelessly keeps producing clicks long after it stopped producing interest.

So the claim that push is all human is half right. Every click does come from a subscribed browser. Whether a person behind it wanted anything, or whether a person was there at all, depends on how that browser became a subscriber.

• Subscriptions collected by deception. Prompts disguised as a video play button, a captcha or a download step get accepted by people who had no idea they were subscribing. The browsers are real and so are the people, but their clicks come from confusion or reflex, not interest.
• Automated subscriptions. Browsers subscribed in bulk by software, or devices kept running to collect subscriptions and generate taps. The notification is delivered and clicked with no person involved. This is the part that is invalid by any definition.
• Mixed supply behind one source. Many push networks buy part of their inventory from partner feeds. A single source or feed ID can carry subscribers from dozens of sites of very different quality, which is why one feed can look fine on average while hiding a small, bad slice.
• Accidental taps. Notifications appear on phone screens where they are easy to hit by mistake. Accidental clicks are invalid traffic in the measurement sense even though nobody meant any harm.
• Traffic sold as push that isn’t. On any network that aggregates supply, traffic from other sources can end up labelled with the format you bought. The only way to know what you actually received is to look at the visits themselves, per source.

The common thread is that the problem sits in specific zones, sites and feeds, not in the network as a whole. That is good news: specific sources can be cut.

Push campaigns fail for two different reasons, and the remedies are opposite.

A real subscriber who taps a notification out of mild curiosity and leaves after a few seconds is a valid visit that did not convert. That is a targeting and offer problem: tighten the audience, change the creative, lower the bid, or accept the format’s economics. Blocking the source would throw away real people.

A tap produced by automation, or by a browser that never had a person behind it, is an invalid visit. No creative or bid change fixes it; the only remedy is to stop buying from that source.

This is why judging push by engagement alone goes wrong. Short, shallow visits are normal for the format, so a model that condemned them would condemn the whole channel and still not tell you which zones to cut. ValidVisit checks every visit against 100+ independent data points and judges it Valid or Invalid, with its evidence, so a zone of uninterested humans and a zone of automation look different even when both have poor conversion rates.

Pass the network’s own identifiers through your tracking URL so every visit carries its source:

• Zone or site ID: the most precise unit most push networks let you block or bid on.
• Source or feed ID: the upstream supplier, where the network exposes it. Useful for spotting a whole feed that is weak.
• Campaign and creative IDs: to rule out the creative before blaming the source.
• Click ID: to tie each visit to the network’s click record and to a conversion postback.

The exact macro names differ by network; the tracking tokens reference lists them. Some networks also let you target by subscriber age or activity; if yours does, compare quality by that setting too, because older, over-mailed lists tend to behave differently from recent ones.

Rank sources by the share of their visits judged invalid, not by click-through or conversion rate, and act from the worst down:

1. Block the zones or sites whose visits are mostly invalid. This is the precise cut and keeps the rest of the feed.
2. Drop a whole feed only when the weakness is spread across it rather than concentrated in a few zones.
3. Lower bids or whitelist where a source is merely weak rather than invalid.
4. Re-check after the cut. Push supply rotates; zones that were clean can degrade as lists age.

Where each network lets you do this, and the limits on list sizes, is in blocking by network. The terms used here are defined in the glossary. ValidVisit reports which sources to cut; the exclusion itself is made in the network.