What exit intent actually detects
Exit intent sounds like mind-reading. It is not. The signal is mechanical and quite narrow: the visitor's mouse pointer leaves the top edge of the browser viewport, heading toward the tab bar, the address bar, the back button or the close box. That is the whole detection. Everything else — the idea that the visitor is about to leave — is an inference layered on top of one cursor movement.
Knowing that changes how you should use it. A pointer crossing the top edge is a reasonable proxy for leaving, but it also fires when someone reaches for a bookmark, switches tabs to check something you just told them, or moves toward a second monitor. Treat the signal as a decent guess rather than a certainty, and write the invitation so that a wrong guess costs nothing.
It does not exist on a phone
There is no cursor on a touch device, so there is no pointer to leave the viewport, so exit intent cannot fire. This is not a limitation to work around — the gesture genuinely does not exist. A visitor on a phone leaves by switching apps, pressing back or locking the screen, and none of that is visible to a page.
A well-built trigger engine treats the device check as a safety net rather than trusting you to remember. MyLiveChat gates exit intent on device automatically — the listener is not installed at all on a touch device, so adding a device is desktop condition to an exit-intent rule documents the behaviour rather than changing it (see what your chat triggers can actually measure). A rule you wrote on a laptop therefore does not quietly misbehave for the majority of visitors who arrive on a phone. It also means that if most of your traffic is mobile, exit intent is a small lever by definition — check your own split before you invest time in it. If phones dominate, your effort is better spent on the advice in supporting mobile visitors.
The pages where it earns its place
Exit intent is an interruption aimed at someone who has already decided to go. That makes it justified in a narrow set of places and obnoxious everywhere else.
- Checkout and cart. Leaving here has a measurable cost, and the reason is usually one unanswered question about shipping, returns or a fee.
- Pricing. Someone comparing plans who leaves without asking is often stuck on a single detail you would happily clarify.
- Quote or application forms. A half-finished form is a strong signal of friction — though see the warning below about interrupting mid-entry.
- High-consideration product pages. Long dwell followed by a move to leave is the shape of unresolved doubt.
Conspicuously absent: the blog, the homepage, the careers page and anything that reads as casual browsing. Someone leaving your blog post is behaving normally. Interrupting that is how a widget earns a reflexive close-click for the rest of the session.
Write it as an answer, not a plea
The most common exit-intent message is a variation on Wait! Before you go!, and it performs badly because it asks the visitor to do the work of explaining themselves. The better shape offers the specific thing that was probably missing.
- On checkout: Any questions about delivery times or returns before you order?
- On pricing: Not sure which plan fits? Tell me your team size and I will point you at the right one.
- On a product page: Happy to check whether this works with your setup — what are you running?
Each of those is answerable in one line, which is the test. If your invitation needs a paragraph of reply from the visitor before anyone can help, it is a survey, not an invitation. The same principles that govern a good opener apply here; see writing a live chat welcome message.
The mistakes that make it a nuisance
- Firing site-wide. The single biggest cause of exit-intent fatigue. Scope it to the pages above and nowhere else.
- Firing mid-form. An overlay over a half-completed payment form is the worst version of this feature. Invite before the form or after it visibly fails, never during entry.
- Asking twice. If the visitor dismissed it, that is an answer. A second attempt reads as pestering and is remembered.
- Firing when nobody can reply. An invitation that opens onto an empty queue is worse than no invitation, because it converts a quiet exit into a bad experience. Pair the rule with your real coverage, as in planning after-hours coverage.
- Stacking it with another invite. If a dwell-time rule already fired on the same page, the exit rule should stay quiet.
What a good engine does for you
Several of those mistakes are better prevented by the tool than by discipline. It is worth knowing which protections you get for free, because it changes how carefully you have to write the rule.
MyLiveChat's proactive invitation rules are evaluated in the visitor's browser, which is why an exit-intent rule can react the instant the pointer leaves rather than waiting for the next polling tick. At most one invitation fires per page view, so a dwell rule and an exit rule cannot stack. A visitor who is already connecting, chatting or has just finished a conversation is not interrupted at all. And a visitor who has already dismissed an invitation on that page will not be asked again.
What the tool cannot decide for you is scope and wording. Those are the two things that determine whether this feature reads as helpful or desperate.
How to tell whether it worked
Exit-intent invitations are unusually easy to fool yourself about, because every conversation they start looks like a rescue. Some of those visitors were not leaving at all, and some would have come back anyway.
Judge the rule per page rather than in aggregate, and give it enough time to accumulate a real sample. A rule on one checkout page with forty firings a week tells you more than a site-wide average. If the acceptance rate is very low and the dismissal rate is high, the message is wrong or the page is wrong — and it is usually the page.
What to measure
- Acceptance rate per page. The only number that tells you whether the interruption was welcome. Compare pages, never the site average.
- Dismissals. A rising dismissal rate is the earliest warning that scope has crept too wide.
- Conversations started that reached a resolution. A rescued visitor who still leaves unanswered was not rescued.
- Share of firings on mobile. Should be zero. If it is not, something is misconfigured.
- What people actually ask. Read the transcripts. Repeated questions on a checkout page are a page problem you can fix permanently, which is worth more than any invitation.