Proactive chat invites that don’t annoy: rules we actually use

Proactive chat has a bad reputation because of how it is usually configured: the widget pops on the homepage after three seconds with “Hi there! How can I help?” and the visitor closes it before reading. The same feature, pointed at the right page with the right delay and a sentence that proves you know where they are, behaves completely differently. This article shares the rules we run on our own site, the copy we use, and the comparison that tells you whether an invite is a conversion tool or a nuisance.

Why most invites fail

An invite is an interruption. For an interruption to be welcome it has to arrive when the visitor has a question they have not yet typed. On the homepage at three seconds, nobody has a question yet. On the pricing page after 45 seconds of scrolling between two plans, almost everyone does. The feature is the same; the moment is everything.

The second failure is sameness. “How can I help?” is what every site says, so visitors have learned to dismiss it without reading. An invite that mentions the page (“Comparing Pro and VIP? Ask about limits here”) is read because it is obviously written for this moment.

Three triggers, combined

TapCub invites combine three conditions. Page: a URL or pattern, like /pricing or /docs/*. Dwell time: seconds on the page, or scroll depth. Source: the channel or campaign the visitor arrived from, because someone from a paid ad has different intent than someone from your newsletter. Our pricing rule waits 45 seconds, only fires for visitors who did not arrive from the help center, and never fires on the first page of a visit.

Because chat and analytics share one snippet, the source condition uses the same channel definitions as the marketing analytics reports, and the resulting conversations carry the source into the agent’s context card. The rule editor below is the one on the proactive invites section of the chat page.

Copy that earns a reply

Good invite copy has three properties: it names the page or task, it asks a specific question, and it fits in one sentence. Write a different one for each rule. Here are the ones we actually run, with the replacement we made after watching the first week.

A useful test before you ship a sentence: read it as the visitor, on that page, at that second. If it could appear on any other page unchanged, it is too generic. If it asks a question the visitor cannot answer yet (“Ready to buy?” on a docs page), it is too early. The ones that work name the exact decision the visitor is stuck on.

Caps that keep it polite

Every rule gets the same four limits. Never invite a visitor who already has a conversation open. Never invite the same visitor twice in 24 hours, on any page. Never fire on the first page of a visit, because the visitor has not had time to form a question. And always include a dismiss control that is as easy to hit as the reply box. Rules that break these feel like pop-up ads, and the dismissal rate will tell you within a day.

Outside business hours, the invite switches to a different sentence that promises an email reply, and the AI assistant answers from the knowledge base first. The AI chat page describes the hand-off.

Measure conversion, not conversations

The wrong metric for an invite is the number of conversations it starts. The right one is whether visitors who chatted converted more than comparable visitors who did not, on the same page, in the same period. TapCub builds that comparison from the shared snippet: chat events are regular analytics events, so “chatted on pricing” is a segment like any other.

In our sample month the pricing page had 2,400 visitors who stayed long enough to be eligible. The gap in sign-up rate between the two groups is the invite’s value, and the dismissal rate is its cost. If the gap closes, the invite is not helping; change the copy or the trigger before you add more rules.

Sample month. Visitors who chat are already more engaged, so read the gap as an upper bound and watch the trend, not one number.

Start with one page

Do not start with a site-wide rule. Pick the one page where visitors are most likely to be stuck (pricing, a long docs page, the second step of checkout), write one sentence for it, and run it for two weeks. Read the dismissal rate daily and the conversion comparison weekly. Then add a second page. Our site runs four rules in total and has for a year; more than that and the rules start competing. Setup steps are in the widget guide.