What in-app messages are for
The push guide draws the line clearly. A push is an interruption delivered through the operating system to someone doing something else, and its job is to earn a return visit. An in-app message is shown inside a session the person chose to start, and its job is to help them get more out of the visit they are already making. In the AARRR model, that puts in-app messages mostly at activation, getting a new user to the moment the app makes sense, and at the points where an existing user could get more value from something they have not yet found.
Most in-app messages fall into a few jobs: guidance on the screen in use, the next step in onboarding, a feature the person has not discovered, a change they need to know about, an offer, and a record of messages that can wait. The push guide lists onboarding and feature discovery as in-app types and promotional messages as push used sparingly with an in-app inbox. What none of those jobs includes is talking at someone in the middle of a task. An in-app message is part of the interface, and it should be designed with the same care as any screen.
Firebase's documentation describes In-App Messaging as a channel for in-session campaigns with event-based triggers, as the push guide states, and most engagement tools work the same way: a message is defined once and shown when a person does something or reaches a screen. This guide makes no claim about any tool's templates, limits or targeting options; the design choices below apply whichever tool sends the message.
Choosing the format for the job
The push guide names the formats: modal, banner, tooltip, card and inbox. Each interrupts to a different degree, and the rule is to use the least interruptive format that will do the job.
| Format | Interrupts | Suits | Avoid for |
|---|---|---|---|
| Tooltip | Lightly; points at one element on the screen in use | Explaining a control or a feature where it lives | Anything not about the element it points to |
| Banner | Lightly; sits at the top or bottom without blocking | A short note after an action, a status, a gentle next step | Long explanations or several actions |
| Card | Not at all; sits in the content flow | Suggestions, the next onboarding step, an offer that can wait | Anything urgent |
| Inbox | Not at all; the person opens it when they choose | A record of offers, news and updates; messages that can wait | Anything the person must see now |
| Modal | Fully; blocks the screen until dismissed | Something the person must know before continuing | Tips, offers, tours, or anything on every open |
The table hides one asymmetry. A modal is the format teams reach for first, because it is the one that is certain to be seen, and it is the one people learn to dismiss fastest. The push guide's warning about full-screen modals on every open describes a familiar decline: a message shown to everyone becomes noise, people tap the close button by reflex, and the next modal that matters is dismissed with the rest. Keep modals rare enough that people still read them.
Triggers and placement
An in-app message should be caused by something the person did or a screen they reached, as the push guide's table of triggers puts it, not by the app opening. Three placements work well as general practice. At a natural pause: after a task is complete, such as a checkout, a saved item or a finished lesson, when attention is free. At the point of need: on the screen where a feature lives, when the person is about to use it or has just struggled with it. On return: when someone comes back to an unfinished task, a card that picks up where they left off.
Placements to avoid follow from the same logic. Mid-task, a message pulls attention away from the thing the person is trying to do. On every open, a message stops meaning anything. Immediately after the app launches, before the person has done anything, a message has no behaviour to be relevant to and stands between them and the reason they opened the app.
Give each message an exit condition as well as an entry one. A tip about a feature stops once the person has used the feature; an onboarding card disappears once the step is done; an offer ends when it is redeemed or expires. A message that keeps appearing after its job is done is the in-app equivalent of a reminder about a purchase already made. Where a message links to another screen, it needs the same care as any link into the app: the deep links and app analytics guide explains how routes and campaign parameters let the analytics tell which message produced the next step.
Copy, calls to action and accessible design
The words follow the habits in the writing for the web guide: the point first, plain language, specific nouns. A heading that says what is new or what to do, a sentence that says why it matters to this person, and one call to actionThe prompt that tells a reader the one action to take next, such as a button or link. whose label names the result ("Set up delivery slots", not "Learn more"). Where a second option is needed, make it the way out ("Not now"), not a competing action. If a message needs more than that, it is probably a screen, not a message.
Accessible design is part of the same job. The Web Content Accessibility Guidelines, as the accessibility guide summarises them, ask that content be perceivable, operable, understandable and robust, and the site's accessibility and writing guides draw out the requirements that matter most for a message. Images that carry meaning need alternative textThe text alternative to an image, read by screen readers and used by search engines., and anything an image says should also be said in text. Colour is never the only way information is conveyed, so a red banner still names the problem in words. Nothing should flash or move unexpectedly; an animated message should be calm and short. Link and button text should make sense on its own when read aloud. To those, add one rule specific to messages that cover part of the screen: a close control that is easy to find, easy to reach and labelled for screen readersSoftware that announces a page's text, headings, labels and messages to people who cannot see the screen, and reports what the page provides., and a message that can be dismissed without a precise tap.
Test messages the way people use them: with a screen reader on, with larger text, in dark mode, and on the smallest supported screen. A message that pushes its button off screen at a larger text size, or that a screen reader announces as an unlabelled image, has failed for some of the people it was meant to help.
Announcing features to existing users
A new feature is the most common reason a team wants an in-app message, and the most common reason messages pile up. The launch planning and tiers guide gives in-product messages a place in its lower tiers: a feature launch typically includes an in-product message alongside email to relevant customers, and a minor release an in-product note where useful. The tier decides whether a message is justified at all.
When it is, send one message to the people who would use the feature, not to everyone. A new filter for repeat orders belongs in front of people who reorder; a feature for teams belongs in front of people who have invited someone. Place it where the feature lives, as a tooltip or a card, at the moment the person is on that screen, and link it to the feature itself rather than to an announcement page. Keep a record in the inbox for people who want to read what has changed, and let the message expire once the person has used the feature or seen it a set number of times. The note that record points to is written with the tutorial on writing release notes and a changelog, so the message and the changelog describe the change in the same words.
Resist the queue of announcements. Each team's release looks like it deserves a message; together they bury the session. The cap and priority rules in the next section apply to announcements as much as to offers.
Caps, priority and measurement
In-app messages need a frequencyAverage times a person saw an ad. cap even though they need no permission. The push guide sets one cap per person per period across every campaign and every team, with a priority order for when several messages qualify, and in-app messages belong under the same cap, because the person experiences push and in-app as one app talking to them. As general practice, set a small limit on how many in-app messages a person can see in a session, show service messages, such as a problem with an order, ahead of everything else, and put promotions last. Where people can choose which kinds of message they receive, the tutorial on notification preferences by topic sets up the topics and defaults that inbox content should respect too.
Measure each message against a holdout groupA randomly chosen part of an audience kept from exposure to measure what exposure caused., as the push guide sets out: keep a random share of the people who qualify from seeing it, and compare the behaviour the message exists to change between those who saw it and those who did not. A message that is seen and tapped by many people who would have done the same thing anyway has done nothing. Content tests, an A/B testControlled comparison of two variants. of format, wording or placement, run inside that structure: the A/B test says which version is better, the holdout says whether showing it is better than showing nothing. Count the costs as well: dismissals, sessions cut short after the message, and complaints.
A grocery app redesigns its in-app messages (illustrative)
Pudgeling Grocery replaces its open-screen modal
Pudgeling Grocery, an invented grocery delivery app, shows a full-screen modal on every app open with the week's offers. About 91% of views are dismissed within 2 seconds, and a 4-week holdout of 10% of users shows no difference in orders: 1.42 orders per active user a month with the modal and 1.43 without. The team replaces it with 2 messages. A banner after checkout says "Book next week's slot now while Thursday evening is free" and opens the slot picker; an inbox card holds the week's offers for anyone who wants them. Both carry text alternatives for their images, a labelled close control and no animation. A new 4-week test with a 10% holdout shows the banner lifting advance slot bookings from 18% of checkouts to 24%, while orders per active user are 1.44 against 1.42 in the holdout, a difference the team treats as too small to claim. It keeps the banner, keeps the inbox card, and retires the modal for good.
The wider post-purchase programme these messages sit inside is set out in the lifecycle marketing guide, and the mobile messaging hub places in-app messages beside push and SMS.
Common mistakes
- A modal on every open. It blocks the task the person came for and trains them to dismiss without reading. Keep modals for things a person must know before continuing.
- Triggering on app open. A message shown before the person has done anything has nothing to be relevant to. Trigger from behaviour, at natural pauses.
- No exit condition. A tip that keeps appearing after the feature is used reads as the app not paying attention. End each message when its job is done.
- Two competing calls to action. One action per message, labelled with its result, plus a clear way out.
- Meaning only in an image or a colour. Some people cannot see either. Put the message in text, with alternative text for images that carry meaning.
- A close control that is hard to find. A message that is hard to dismiss feels like a trap. Make closing easy for every user, including screen reader users.
- Announcing every release to everyone. Send feature messages to the people who would use the feature, at the screen where it lives, under the shared cap.
- Judging on views and taps. Measure against a holdout, and count dismissals and abandoned sessions as costs.
