The visitor tracker: see who needs help before they ask

By the Onsites AI team · Last updated August 28, 2026 · 5-minute read

Most support starts reactively: bubble in the corner, wait, respond. A visitor tracker inverts it — you can see who is on your site right now, which page they are reading and how long they have been stuck, and open the conversation at the moment help is worth the interruption. Used with restraint, it is the highest-leverage feature on the desk: the visitor gets a relevant answer in seconds, the business gets a conversation at the exact point of decision. Overused, it becomes the popup everyone closes on reflex. The difference is entirely in the rules you set.

What live visitor tracking actually shows

In your workspace, the visitor view shows the people on your site in real time: current page, the path they took, how long they have been on it, and — where the visitor identified themselves earlier — the context of who they are, attached to their customer record. That context is the whole point: "anonymous visitor on /pricing for 4 minutes" and "logged-in customer with three open orders staring at the shipping page" are different situations, and the difference is what you say when you open the conversation.

The tracker answers three questions support otherwise guesses at. Where do people stall? — the pages nobody leaves are fine; the page with an average dwell of eight minutes and no conversions is asking for a fix, chat-assisted or otherwise. Who arrived from where? — a pricing-page visitor and a help-documents visitor have opposite needs. When do they want talking to? — two minutes reading docs is research; two minutes on checkout with a half-filled form is a rescue.

When proactive chat helps (and when it reads as stalking)

The line is specificity. A generic "Hi! Need help?" that fires on any page after twenty seconds is the annoyance tax — visitors close it on muscle memory, and it dilutes every future genuine greeting. A message that proves someone looked at the session reads as service: "Hi — I see you're on the 40ft LCL rates page; want me to pull a quote with this week's numbers?" is a different species of interruption because it could not have been sent to everyone.

The situations where being first is usually welcome: high-intent pages (pricing, quotes, checkout, booking) where a hesitation has a meaning; long dwell on a complex page, which signals confusion worth offering to untangle; error or dead-end states, where a "looks like that search came up empty — what were you looking for?" saves the visit; and returning identified customers mid-task, where the account history lets you skip the small talk. The situations where it is not: first-touch blog readers, people mid-scroll on mobile, and anyone who has already declined chat this session — a cooldown rule fixes all three.

The trigger rules: defaults that pass the annoyance test

• Page rules, not timer rules. Trigger on specific pages (pricing, checkout, docs) rather than "anything after N seconds". Specificity is precisely what makes proactive chat feel personal.
• Dwell thresholds with meaning. A threshold near the page's average dwell flags actual confusion; a ten-second threshold flags everyone. Set it to something like 1.5× the page median.
• One proactive message per visitor per session. The declined chat must not escalate into a second popup. Persist across pages until the session ends.
• Cooldown across visits — a visitor who declined yesterday should not be greeted today the same way. Skip proactivity for returning decliners; the trigger design guide covers the escalation and cooldown patterns in detail.
• Staff it live. A proactive greeting that nobody answers for six minutes is worse than no greeting: it converts mild curiosity into active distrust. Open conversations only when someone is at the desk — the presence indicator in your working-hours settings is the switch.

From greeting to conversation to record

Proactive messages work when they hook into the machinery behind the desk. The conversation that opens becomes a normal thread in the shared queue — owned, taggable, measurable like any other. The visitor who chats is a contact on the CRM record from that point, with the transcript attached, which is how a chat that started as "what does insurance on a shipment cost" resurfaces later as the moment the quote was won. And because the widget is the same one from the installation guide — one snippet, before the closing body tag — turning visitor tracking on is a settings decision, not a development project.

Measure the outcome honestly: proactive opens that turn into conversations, and conversations that end in the tag you care about — "quote-request", "order-help", whatever your vocabulary says success looks like. If a trigger fires often and opens nothing, its threshold is wrong; if opens happen and conversations die immediately, the message is generic. Both are one-line fixes in settings.

Privacy: the part that determines whether this is creepy

Live tracking deserves its own paragraph of conscience. Track pages and behavior on your own site, for service, while the visitor is there — do not fingerprint, do not stitch anonymous identities across sessions with tricks, and do not display data you would not show the visitor to their face. When a visitor identifies themselves — by contacting you — the history attaches to their record, which is a feature; pretending you know more than you do about strangers is the failure mode that gets tools disabled by privacy-conscious teams. State in the privacy notice that the site uses chat with live visitor support, and honor a visitor's non-participation instead of re-triggering at them. The security overview covers what the platform itself guarantees; the etiquette rules above cover what your team guarantees. Together they are the reason "we saw you were looking at X and thought we could help" lands as service rather than surveillance.

Where the tracker fits in the wider desk

As a single tool, visitor tracking is a nice-to-have attached to the widget. Inside a working desk it is the front door: the visit becomes a conversation, the conversation a contact, the contact a quote, the quote an invoice — all in one system, all free to run on three seats, with the conversion playbook tuning what happens after "hi. " And if the data-residency question ever arrives from procurement, the entire loop — tracker, widget, queue, CRM — runs self-hosted inside your own network without losing a step. Start reactive if that is where your team is; turn the live view on when someone can watch it. Just do not let it open conversations nobody is there to finish.