OneSignal is the name ChatGPT and Gemini reached for most: it appeared in 5 of the 6 answers we measured on 2026-08-03, an 83% hit rate and more than any other push notification service in the set. We put six buyer-style questions to the two assistants, gave each one a single live web search, and recorded every product named and every page cited. This is the whole result, counted.
Across those 6 answers the assistants named 18 different product labels, but the field thinned quickly. Only three products were named in half the answers or more: OneSignal at 5 of 6, PushEngage at 4 of 6 (67%), and MagicBell at 3 of 6 (50%). Everything else landed at 2 of 6 or 1 of 6. So the "who gets recommended" question has a very short real answer, and a long tail behind it.
The names that repeated
OneSignal was named in 5 of 6 answers (83%), and the interesting part is where those mentions came from: onesignal.com was cited in only 3 of the 6 answers, so at least two of OneSignal's mentions came from pages OneSignal does not own. Being named on someone else's page is the strongest position in this whole dataset.
PushEngage was named in 4 of 6 answers (67%), and pushengage.com was itself cited in 4 of 6 answers. That is a tight one-to-one line between publishing content the assistants pull and being named in the result. PushEngage shows up almost exactly as often as its own site is read.
MagicBell was named in 3 of 6 answers (50%), with magicbell.com cited in 3 of 6 as well. Below the top three sits a five-way tie at 2 of 6 (33%): Airship, Braze, Brevo, Firebase, and Knock. Then ten more products were each named in exactly 1 of 6 answers (17%): CleverTap, Expo, FCM, Firebase Cloud Messaging, MoEngage, PropellerAds, PubNub, Pusher Beams, PushOwl, and Pushwoosh.
How the shortlist changed with the question
This is the part worth reading twice. The six questions did not return the same list with the order shuffled: they returned different casts of products, and the wording decided who showed up.
The bare "best push notification services" question returned 8 names, and four of them appear nowhere else in the whole measurement: Expo, Pusher Beams, PubNub, and FCM. These are developer and infrastructure tools, and the moment we added a price or team word to the question, they vanished.
"Best free push notification services" returned a tight 4 names: MagicBell, Knock, PushEngage, and OneSignal. "Most affordable push notification services" returned 7: OneSignal, PushEngage, MagicBell, Brevo, PushOwl, Knock, and PropellerAds. The free list is a strict subset of the affordable list: every product the assistants called "free" they also called "affordable," but "affordable" added three more, and two of those (PushOwl and PropellerAds) appear in only that one question. Free and cheap are not the same shortlist to an AI.
"Best push notification services for small teams" pulled in an almost separate universe: OneSignal, PushEngage, Airship, Braze, CleverTap, MoEngage, Pushwoosh, and Firebase Cloud Messaging. Only OneSignal and PushEngage carried over from the price questions. The other six are heavier enterprise and infrastructure names, and four of them (CleverTap, MoEngage, Pushwoosh, Firebase Cloud Messaging) were named in this question and no other.
The two vaguest phrasings collapsed to almost nothing. "What push notification services should I use" returned 3 names (OneSignal, Airship, Firebase). "Push notification services recommendations" returned just 2 (Braze, Firebase). The precise commercial questions produced 7 and 8 names; the open-ended ones produced 2 and 3, and both leaned on Firebase, the only product named in both vague answers. If you sell in this category, the lesson is blunt: you are competing for a different, shorter list depending on whether the buyer types "free," "for small teams," or just "recommendations."
The source map
Every one of the 6 answers cited a Reddit thread: reddit.com appeared in 6 of 6 answers (100%). Quora and Business of Apps each fed 4 of 6 (67%). Community discussion and independent roundups, not directories, are doing the heavy lifting here.
The next tier is the giveaway. Four of the most-cited domains are vendor sites whose product was also named: pushengage.com (4 of 6) with PushEngage named 4 times, magicbell.com (3 of 6) with MagicBell named 3 times, onesignal.com (3 of 6) with OneSignal named 5 times, and brevo.com (3 of 6) with Brevo named twice. When a vendor's own page is retrievable and on-topic, that vendor tends to appear in the answer. That is the key insight of this whole page.
Knock is the instructive exception. knock.app was cited in 4 of 6 answers, tied for the second-most-cited domain, yet Knock the product was named only 2 times. Its content gets pulled heavily, but that content is comparison writing that names competitors, so it lifts other products as much as its own. Being cited as a source is not the same as being named as a pick.
One absence matters: no G2, no Capterra, no TrustRadius, no Gartner appeared in any of the 6 answers. The directories that dominate category SEO were nowhere in what actually fed these AI recommendations. The pages that mattered were reddit.com, quora.com, businessofapps.com, wisepops.com (2 of 6), and a scatter of single-cite sources like aws.amazon.com (3 of 6, as reference docs), eugeneyan.com, globaldev.tech, sashido.io, viasocket.com, and wpbeginner.com.
What a push vendor would do to get named
Nothing in this data rewards market share directly. What gets you named is being present, in your own words and in other people's, on the exact pages these assistants retrieve. That means a Reddit and Quora presence (cited in 6 and 4 of the answers), a spot in the Business of Apps roundup (4 of 6), and your own domain publishing on-topic comparison pages the way pushengage.com and magicbell.com do. OneSignal's 5 of 6, with only 3 of those from its own site, is the target: named on pages you do not control.
The full list, counted
*Every product the two assistants named across the 6 answers, top of the list. Firebase is listed under all three labels the assistants actually used.*
| Product | Named in | Share |
|---|---|---|
| OneSignal | 5 of 6 | 83% |
| PushEngage | 4 of 6 | 67% |
| MagicBell | 3 of 6 | 50% |
| Airship | 2 of 6 | 33% |
| Braze | 2 of 6 | 33% |
| Brevo | 2 of 6 | 33% |
| Firebase | 2 of 6 | 33% |
| Knock | 2 of 6 | 33% |
| FCM | 1 of 6 | 17% |
| Firebase Cloud Messaging | 1 of 6 | 17% |
Firebase is the double-labelling case: the raw measurement lists it three ways, "Firebase" (2 of 6), "FCM" (1 of 6), and "Firebase Cloud Messaging" (1 of 6). We keep them as separate rows because that is how ChatGPT and Gemini said them, and summing them would invent a combined figure the measurement never produced. Eight more products were each named once and sit just below this table: CleverTap, Expo, MoEngage, PropellerAds, PubNub, Pusher Beams, PushOwl, and Pushwoosh.
Where we fit
Glotier does not sell push notifications, so we are correctly absent from all 6 of these answers, and we would be suspicious of a visibility tool that somehow named itself here. But this page is the exact thing we run for a customer: the same six-question fan-out, the same count of who gets named and which pages fed the answer, pointed at your category instead of this one. If you want to see where your own product lands, the check is free, needs no account, and takes about a minute.