By the Onsites AI team · Last updated · 5-minute read
Most live chat widgets are support desks doing a salesperson's job in a salesperson's clothes: a round bubble in the corner that says "Hi there!" to everyone, answers what it can, and converts approximately nothing. The teams where chat genuinely moves revenue treat it as a different instrument — chat is the only channel where the buyer is standing in the store while you talk to them, browsing the pricing page, hovering over the plan comparison, three clicks from the form — and they instrument it accordingly: placed where decisions happen, greeted with context rather than a greeting, answered with evidence, closed with a next step. This guide covers the placement, timing and copy that turn the widget into a channel, with the honest caveat first: chat converts only where a fast, human, honest answer is possible — the desk that can't meet the clock should place the contact form visibly instead, because an unanswered chat is worse than none.
Chat can't convert everywhere; it converts where hesitation lives. The product/pricing pages are the primary beats: a buyer on your pricing comparison or mid-scroll on a feature table is doing exactly the arithmetic chat is built to resolve ("can I use this with my shop?", "what's a seat if I'm 3.5 people?"). The checkout/start flow: the moment before a form, where abandonment concentrates — a small, quiet chat presence answers the last objection in-line. The high-consideration article: someone deep in a guide like this one is researching before deciding; the widget offers the human behind the research. Where it shouldn't be primary: the 404 page (frustration, not intent), every page on first visit from an unknown source with a pop-open (reads as ambush), and inside dashboards users already pay for — they'll find support when they need it. Placement isn't a settings page; it's a merchandising decision, revisited when pages change — the desk that greets the pricing page visitor differently from the documentation reader is already converting more than the one-size bubble does.
The generic "Hi there! 👋" is the most expensive words in chat — it spends the buyer's attention and purchases nothing. The converting greeting is contextual and late enough to be credible: fire after 15–45 seconds on the page (the tracking knows which page they're on and usually from where), and let the copy prove a human is behind it: "That page compares the two meter shapes — happy to run your numbers if useful" beats "Welcome to our site!" by every measure that matters. The rules that keep it welcome: once per page per visit, cooldown before any second trigger (the trigger discipline this desk's growth guide covers fully), never on top of a user's own scroll momentum, and never instant-on-load — instant chat reads the way a car lot's approach reads. The response clock is the real timing: a chat answered in two minutes converts on the answer; the same chat answered in fifteen announces nobody's home — which is why the honest desk sets the widget's working hours to match reality and shows an out-of-hours contact path the rest of the time, the channel-honesty that buyers reward with trust.
What converts inside a chat is not a pitch — it's an excellent answer to a buying question. The anatomy, observed a thousand times on good desks: lead with the answer ("Yes — the free tier covers all three of you, and seats beyond that are $15"), attach the evidence (the quote link, the document, the exact worked math for their case), one follow-up question maximum (the question that moves them forward: "worth emailing you the cost split?"), and a next step they can take now — "start here, takes two minutes" with the link, not "feel free to explore." The anti-patterns are equally observable: over-eager closers ("Let me get you a discount!!" as the third message), the wall of links in place of an answer, and the AI draft sent unconfirmed with the buyer's own name misspelled. The template discipline is the same one the desk already runs — replies drafted, human-sent — applied to buying conversations: consistent quality, a person's accountability, none of the script's staleness.
Two conversations decide whether the widget feels premium or pushy, and both deserve rehearsed lines. "Can you give me a discount?" — the buying question that kills deals only when fumbled; the converting answer holds the structure without inventing authority: "That's actually our lowest tier — the honest reason it can go lower is the meter: you pay for seats you use and no per-resolution line. If budget is the constraint, start free and let the first invoice price the upgrade." (The price-honesty register works in both directions; buyers can hear it.) The chat that goes quiet mid-conversation — the buyer who asked, got the answer, and walked; the desk's one permitted nudge is the next-day follow-up with new value, not a re-ask: "Yesterday's math was for 5 seats — if you've been running it at 10, the split under each meter is $x vs $y" (the new-value cadence, once, then silence). Both replies work because they respect the buyer's arithmetic — the chat widget's entire conversion advantage is being the moment where arithmetic is spoken aloud.
One small desk's experiment, because conversion claims deserve receipts. Setup: widget enabled only on pricing, comparisons and the start flow (placed where hesitation lives), one context greeting per page family, response target under two minutes during working hours, honest out-of-hours note otherwise. Week one: baseline — 61 chats initiated, 8 mentioning price hesitation, 2 ending with a start click; the greeting was a blank "Hi there!", response times 6–40 minutes, and two chats sat unanswered past working hours with no note. Week two: the fixes — contextual greetings that named the page, money-first triage on the queue, the boundary phrases pre-loaded for "can you go lower?" questions. Week four: 70 chats, 17 on price, 9 ending with a click through, plus three quotes requested mid-chat (the quote link dropped directly into the thread) and a monthly deal whose thread shows the whole arc: question, worked answer, warm start. The lift didn't come from magic copy — it came from the widget finally sitting where decisions happen, answered at the speed the channel promises, by people whose replies carried evidence. Chat converts when it's your best desk on its best day, made standard.
Does live chat actually convert, or is it a support cost?
When it's placed where decisions happen and answered fast, it converts measurably: buyers on pricing pages and checkout flows ask buying questions, and an answer with evidence plus a clear next step turns hesitation into starts and quotes. Treated as a corner bubble with a generic greeting, it stays a cost — placement and the response clock are what make it a channel.
Where should the chat widget go?
On product, pricing and comparison pages, the checkout/start flow, and deep guides readers use before deciding. Skip 404s and instant pop-open on first visit; don't force it inside paid dashboards. Placement is a merchandising decision, revisited when pages change.
How fast must first response be for chat?
Minutes, not more: two minutes is a strong working-hours target, and anything near fifteen reads as nobody's home. Set the widget's visible hours to match reality and show an out-of-hours contact path the rest of the time — honest slowness converts better than a live-looking widget with a dead clock.
What should the greeting say?
Contextual and human: fired 15–45 seconds in, naming what's on their screen, with an offer keyed to that decision ("that page compares the meters — happy to run your numbers"). Never instant-on-load, once per page, with cooldowns before any repeat trigger.
How does AI fit a converting chat without killing the tone?
As the draft layer: the copilot drafts from the thread and account context at published credit prices, and a human confirms before sending. Automated greetings and honest working-hours notes are fine; unattended selling isn't — the conversion happens because answers are excellent, not because nobody is writing them.