Explore the chat experience, compare widget styles, or preview MyLiveChat directly on your website.
Practical, product-honest guides: how to install chat anywhere, how small teams keep response times honest, and how AI and human agents share the same queue without visitors noticing the seam.
Getting started Conversation craft Team and operations Measurement and quality AI and deflection Sales and conversion Common support scenarios Channels and traffic sources By industry and business type Security, privacy and compliance
Install the widget, choose the software, and get the first conversations right.
Seven days for verification, thirty minutes for password recovery, five minutes between resends. The reasoning behind each number, and why your first email keeps working.
Four consoles, one account. What the browser, Windows, macOS and classic builds are each for, and why mixing them across a team is fine.
Five switches and a template picker decide where your chat opens and which theme renders it. The defaults are not what most people assume.
Three quick links and a dismissible announcement card. What each field accepts, and why editing the text brings it back for everyone.
The rollback screen is one-way, workspace-wide, and does not show your current version. What it actually changes when you save.
Install detected is a memory, not a live probe. Why testing on localhost never turns it green, and why it can never turn back.
The small set of JavaScript functions that set up a chat before it opens, when they can be called, and why a successful-looking call can still do nothing.
Topic buttons on the chat home screen, how the rules parse, the twelve-row ceiling, and the four buttons you never wrote.
What the chatbox behaviour settings control, why some switches ship on, and the two rules that silently rewrite what you saved.
Why the install snippet carries defer, data-cfasync and referrerpolicy, which attributes are safe to remove, and how to verify parallel loading.
Pre-chat, post-chat and offline forms each have an embedded version and a popup version with separate settings. Which one applies, and the handful of settings that are shared.
Page restrictions, vacation mode, and the pattern rules that decide where your widget shows up. Includes the anchoring trap that makes most first attempts silently miss.
The dry run, the duplicate rule, the fields that translate and the ones that quietly do not, plus the order that keeps a migration boring.
Signed-in visitors should not retype their name, email and order number. How to pass known details into a chat, and the security line that separates convenience from identity.
The theme settings cover most of it. When they stop short, custom CSS is the escape hatch, and it lands in the same cascade as your website.
Every visitor-facing string in the widget can be rewritten per language. Where the catalog lives, what it covers, and what it deliberately does not.
A short setup moves the chat window onto your address. Here is what it actually changes, and the one server rule that quietly undoes it.
A deployment with no chat button at all. What monitor mode changes, how visitors can still reach you, and the pages where hiding the launcher costs more than it saves.
Hide the widget from visitors you cannot help, by country, browser language or device -- and understand which way round the filter actually reads.
One account can cover several sites with the same chat code and one console. What gets shared, how to tell them apart, and when to split.
Custom fields decide what your agents know before they read a word. Which ones earn their place, where they appear, and what not to ask for.
When switching it off is the right decision, the three ways to do it, and what to sort out before you do.
Older accounts keep the widget generation they were created with. Here is how to tell which one you have and what moving involves.
Two rules, one visitor. How precedence, cooldowns and one-per-page-view decide which invitation fires, and how to order rules so the specific one wins.
What the exit signal really detects, why it does not exist on phones, and the few pages where catching someone on the way out is worth the interruption.
The install is correct and a slice of visitors still see nothing. Ad blockers, tracking protection, corporate networks and your own CSP.
Building your own chat looks like a two-week job and rarely is. What is in the box, what shows up after launch, and when building is genuinely right.
Installing chat through a tag manager keeps the change out of your codebase, and adds a layer that can silently stop the widget appearing. Here is what to check.
When less chat is the right answer, what to narrow before you remove, and how to withdraw a channel without punishing the people who used it.
A single-page app never reloads, which changes how a chat widget behaves: what survives a route change, what quietly stops firing, and the ten-minute test to run before you ship.
Eight assumptions that keep teams from using chat well, and what is actually true about staffing, speed, bots, small sites, and whether visitors find a chat widget annoying.
A chat widget is third-party code on every page, so the question is fair. Here is what it actually costs, why async loading matters, and the few things worth checking.
Concurrency, containment, deflection, first response time. What the vocabulary means and where each term gets used misleadingly.
Is chat worth it when few people visit? How to run it well without sitting and waiting.
What to export before you cancel, what has to be rebuilt by hand, and how to cut over without a gap in coverage.
The widget is live, now what. A day-by-day plan for week one, and which early numbers actually mean anything.
One founder cannot watch a chat window all day. Honest hours, an offline path that still captures leads, and when to switch it off.
Make the first line do a job: specific, low-effort to answer, matched to the page — and tested like ad copy.
Most chat happens on a phone. Dodge the thumb zone, keep messages short, cut form fields, and survive interruptions.
Keyboard, screen readers, contrast, motion — and why your widget must cooperate (not fight) with accessibility overlay tools.
Each field you add loses some chats — so every field must earn its place. What to ask, when to ask it, and when no form beats any form.
Bottom-right by convention, out of the thumb-path on mobile, labeled with words not just an icon — and never covering the thing people came to do.
Price it at your team size (not the teaser tier), read the free-plan fine print, ask how AI is metered, confirm you can leave with your data — then run the 30-minute hands-on test.
Put the snippet in the new layout on day one, remap page-based triggers to the new URLs, re-run the mobile placement audit — and read launch-week transcripts like telemetry.
The one-script-tag method that works everywhere, plugins where they exist, and the five settings worth configuring before launch.
Response-time targets, coverage honesty, canned responses, and the handful of habits that separate useful chat from an ignored widget.
The difference between a helpful invitation and a pop-up visitors resent comes down to page choice, timing, and one specific sentence.
Installing the widget takes minutes. This is everything to settle before you point it at real visitors.
Chat rarely fails because the software is bad. It fails in a predictable handful of ways, all of them fixable in an afternoon.
Most missed chats are missed because nobody was told. Getting alerts right without burning the team out.
Chat without an owner drifts into neglect. The three teams it usually lands with, what each does to it, and what ownership has to include.
What good agents actually type: openers, tone, de-escalation, and clear writing.
Role, plan, version, tenure, company size, industry, timezone, environment. How self-stated context is pinned to a card, and why the list empties on transfer.
Greeting, Discovery, Diagnosing, Resolving, Closing, Stalled. How the phase is scored, why Stalled is a sub-state, and why it cannot block your wrap-up.
Pinned and recent replies float to the top of the canned list. What is remembered, why a rename silently orphans a pin, and why the order is yours alone.
Emails, order numbers and tracking codes are pulled out of the transcript as copyable chips. What the patterns catch, what they get wrong, and when the panel empties itself.
The promise to come back in ten minutes is the one nothing tracks and everyone breaks. Here is how the follow-up card holds it, and the seven-day rule that drops it.
How the console spots a question, the short ones it keeps, the contractions it misses, and why a holding reply clears the entire checklist at once.
How the agent console decides that a sentence was an instruction, why a single confirmation only ever closes one step, and the phrasings it quietly misses.
Three sentence shapes get captured as a commitment, nine get refused, and one very common polite opener makes a real promise invisible.
How the visitor pulse pill turns a message into a number between minus ten and plus ten, and the neutral words that quietly drag it down.
A teal gauge counts words from both sides and calls the balance Listening, Visitor-led, Balanced, Agent-led or Monologue. What the thresholds mean in practice.
A running clarity label sits beside every chat you handle. This is the arithmetic behind Plain, Mixed and Jargon-heavy, and what the label is safe to use for.
A small pill above the compose area surfaces an unanswered question, a Caps Lock warning or a reply-length meter, one at a time and in that order.
The console can be driven without leaving the keyboard. Which shortcuts pay for themselves first, and how to get a team using them without a training session.
One ordinary conversation shown in full, with a note on what every turn is actually doing and where it could have gone wrong.
Sometimes the customer is right and the rule is wrong. How to stay honest without trashing your employer, and how a repeated complaint becomes evidence.
Real names, first names or a shared team identity, what each signals, and when a pseudonym is a safety decision rather than a marketing one.
The expert customer is not a threat. Drop the script, say what you do not know precisely, and capture what they just taught you.
The request is usually about progress, not hierarchy. What to say in the moment, how to hand the chat over without losing the thread, and what a rising escalation rate is telling you.
Warmth helps some conversations and irritates others. When small talk and humour earn their place, and how to sound human without making the visitor wait for it.
You told them the wrong thing. How to correct it cleanly, without hedging or over-apologising.
Most slow chats are slow because of the first question. Diagnosing fast without interrogating the customer.
The close is what customers remember. Confirming outcomes, handling unresolved issues, and avoiding the silent drift.
Most chat complaints are broken expectations, not bad service. Set them about wait times, scope and follow-up.
The first line sets the whole tone. How to open so visitors feel heard, and why the best opener is rarely scripted.
In a heated chat, exact words matter. Phrases that calm a frustrated customer, and the ones that make it worse.
Fast and good can pull in opposite directions. How to balance speed against quality and stop optimizing the wrong one.
A confusing reply creates more questions than it answers. Write chat messages that are clear at a glance and easy to act on.
Listening is harder in text. How to practice active listening so customers feel understood, not just processed.
Used interchangeably, but different tools. What each is, when to use which, and how to combine them cleanly.
Running several chats at once is a skill. Set a sane limit, keep every visitor attended to, and know when to stop taking new ones.
Five canned replies are easy; a library that stays accurate as you grow is the work. Structure, name, maintain, prune.
Warm, plain, brief — and consistent as you grow. Give agents guardrails and examples, not word-for-word scripts.
Openers for high-intent pages, honest objection handling, the discount question, and closes that move to a signup without the used-car energy.
Acknowledge before solving, never match the temperature, name what you CAN do — and know the two moments to leave chat entirely.
Chunk long answers, narrate every silence, mirror the customer's register, own the wait — the habits that read as professional in text.
Twenty copy-ready responses organized by moment — greeting, delay, escalation, refusal, closing — and the three rules that keep them from sounding robotic.
Walking someone through a process in a chat window is its own skill: pacing, describing what they will see, and knowing when to stop typing instructions.
A second meter grades the agent, not the visitor. This is how four kinds of warmth marker, a tight dismissive list and one short-reply rule turn a reply into a number.
Staffing, routing, coverage, and keeping a queue healthy as volume grows.
Settings screens fetch from the chat server over a call with a fifteen second budget, and a save is one way with no read-back. How to tell a slow hop from a lost setting.
Same account, two different sidebars. What removes an entry, why the menu is never the security control, and where the denied banner comes from.
Ten languages ship, but only part of the product is wired to them. How the language is chosen, why it follows the browser rather than the account, and which screens actually change.
Two record types, a two-character minimum, ten results each, and no transcripts. What the box actually reads, and where to look instead.
One note per visitor, shared across the team. What decides whether it survives until their next chat, and the three limits that never warn you.
Round-robin hands each new ticket to the next agent in your list. When it fires, what skills change, and the ways it quietly leaves a ticket unassigned instead.
Saved views turn a filter into a one-click chip. Shared versus personal, why the four starter views come back when you delete them, and the two limits that truncate quietly.
The automatic end-of-chat transcript email: what has to be true for it to arrive, why duplicates do not pile up, and where the ceiling sits.
Three lists, applied one year at a time. The UK preset deliberately omits Easter, and re-running a preset is safely a no-op.
Export writes eight fields per macro and import reads four. What survives the round trip, and where the duplicate check stops looking.
Related, duplicate of, blocks — three link types with different symmetry, and what each one looks like from the other ticket.
Four preset buttons, a custom field, and a ledger that never forgets. Why corrections are negative numbers and what the per-agent rollup can and cannot tell you.
What gets recorded when nobody answers, why the same visitor can appear twice, and how to work the list without deleting more than you meant to.
A per-site override sits on top of a system default, with a third state that clears it. What the flags control, why the defaults differ, and how to roll one out safely.
Four idle policies share one clock, only the mouse and keyboard reset it, and the lock setting is not a lock. How to set timeouts that fit real work.
Snooze is a promise with a date on it and merge is a link, not a move. What each one changes, and the traps in between.
Online agents, running chats and active visitors on one screen, plus the ability to message an agent or sign them out. What to use it for, and what not to.
Five templates fire automatically in your name. What each one is for, which two may never send at all, and how to edit them safely.
Mentions pull one person into a ticket without reassigning it. Which notes they are parsed from, which name they match, and why a mention is not the same as a notification.
The settings screen rather than the strategy: what enabling distribution changes, why auto accept is a commitment, and how to choose a reassign timeout.
An idle seat looks staffed and answers nobody. The four rules, what counts as activity, and why this is a queue control rather than a security one.
Route the moments that need a person into the tool your team already has open, without turning a useful channel into noise everyone mutes.
Joiners and leavers handled by your directory instead of a checklist nobody remembers to run.
The accessibility conversation usually stops at the widget. Your agents use software all day too, and shift design matters as much as the tooling.
Automatic assignment or manual pickup, what each one gets wrong, and why an unclaimed chat needs an owner by name.
Phone coverage extends your hours cheaply and degrades quality quietly. What it is genuinely good for, and the rules that stop it becoming always-on.
Whether to offer chat to free users, how to tier it without insulting anyone, and the wording that declines an out-of-scope request without losing the upgrade.
Contract and outsourced agents can extend your coverage well beyond what you could staff directly. What decides whether it works is the brief, the escalation boundary and the review loop.
You cannot read the queue on someone's face any more. The rituals, coverage rules and onboarding that replace being in a room.
Most escalations are permission problems. How to set decision limits agents can actually use.
What actually predicts a good chat agent, and how to test for it before you hire.
Distributed agents cannot lean over a desk to ask. Handovers, shared answers, and coverage across time zones.
Chat is relentless in a way phone and email are not. Concurrency limits, recovery time, and metrics that do not punish care.
Holidays bring more questions and fewer staff. Plan coverage, set honest expectations, and avoid burnout and silence.
One team, several brands. How to route, staff, and stay on-brand across sites without separate teams for each.
When chats outpace agents, the queue decides the experience. Set expectations, prioritize fairly, keep waiters from leaving.
Concurrent chats look efficient until quality drops. Find the point where more chats hurt, and balance load fairly.
Make a shift change invisible to the visitor: pass open chats with context, and never make them repeat themselves.
A playbook for spikes: routing, canned replies, concurrency caps, deflection, and honest wait times.
Size the team from your own arrival curve, set a concurrency you can sustain, and go honestly offline rather than staff a window thin.
Route only when it saves a re-explanation: real departments, skills for the hard questions, and intent before any dropdown.
Show offline honestly, let KB + AI carry the night, promise a reply time you can keep — and triage mornings by wait, not by arrival order.
Day 1-2 read transcripts, day 3 shadow, day 4-5 answer with a safety net — and the scoped-permissions setup that makes early mistakes cheap.
Pre-write the peak answers, retrain the AI on them, narrow invitations, borrow help you trained in advance — and plan the day AFTER the peak.
Localize the widget honestly, run translation-assisted chat with candor, translate the top articles first — and hire for a language only when the data says so.
Launch chat is telemetry: staff the spike window, pre-write the obvious answers, tag every conversation, and pipe what you hear straight into the fix list.
Theme each client's widget to their brand, decide who answers before launch, report with transcripts — and design the handover before you need it.
Why good chat agents leave, four directions to grow that are not management, and how to make progression concrete enough to believe.
The numbers worth tracking, and the review habits that keep quality honest.
The chats chart and the ratings chart apply different filters to the same week. What each one counts, and which day an overnight conversation lands on.
An offset cannot tell London from Reykjavik. Why timestamps drift by an hour for part of the year, and what setting a region actually changes.
Country centres rather than cities, an empty map when nothing upstream resolves a country, and the one way to get real coordinates.
The analytics screen reads a rolled-up table and rebuilds it while you wait. Knowing that explains the load time, the freshness, and one count that measures less than it seems.
The headline score, the response rate and the per-agent table are built from three different populations over two different clocks. What each one counts.
Nine topics scored from both sides of the conversation, with a Clear, Likely or Mixed label. Why one message is one vote, and where keyword matching fails.
A tab strip tells you which chats are open. It does not tell you which one is quietly going wrong. Here is how the attention ranking scores them, and where it stops.
What the agent stats widget counts, why the average reply time flatters you, and the three reasons the totals cannot be compared between people.
Yellow at thirty seconds, orange at sixty, red at two minutes. What the counter actually counts, and the one beep you can spend without hearing it.
The momentum card watches every other overlay once a second and scores the changes. Closing, Advancing, Steady, Drifting, Stuck, and the arithmetic underneath.
The live monitor shows everyone; the stored visitor list does not. Here is the engagement test that decides which visitors get a record you can segment and report on.
What each counter on the KB analytics screen is actually measuring, how queries get grouped, and the searches that never reach the log.
Two export types, two formats, and four limits that will stop you — plus the CSV quirks worth knowing before you open the file.
The full status vocabulary of the live visitor list, the five minute rule behind Browsing and Active, and the one label that reads differently in the agent console.
Two screens, four switches that start on, and one rule that withholds the survey entirely. What each does to your CSAT data.
Views, searches and helpful votes tell you which articles are quietly not working. How to read the four numbers, and which failure each one describes.
An SLA clock that runs overnight breaches tickets nobody could have answered. What the business calendar changes, which switch turns it on, and why the hours are in UTC.
Weekly and monthly digests, plus per-chat transcript forwarding. Where each is delivered, what silently stops it, and when email is the wrong channel entirely.
Who was signed in, when, and at what status. How the record is built, what its resolution really is, and the four things it cannot tell you.
A fall in chat volume is a symptom, not a cause. The triage order that finds the reason fastest, cheapest checks first.
What chat measures on its own, what you have to send yourself, and the double-counting traps that make chat look better than it is.
Phantom open chats, your own test conversations, inconsistent tags and a time zone nobody checked. The audit that takes an afternoon and settles the argument.
Chat numbers matter most beside everything else you measure. Grain, join keys, schedules, and the privacy promise that travels with the data.
Reviewing transcripts shows you how agents write. Only walking the whole thing as a visitor shows you what your chat is actually like to use.
Most chat changes are shipped on a hunch and judged on a feeling. A simple, honest test tells you whether the change helped, hurt, or did nothing at all.
The same number that informs a manager can distort a shift. How to show agents their own numbers without turning measurement into a threat.
Satisfaction tells you how a chat felt; effort tells you how hard it was. How to measure effort in chat, and the signals that reveal it without a survey.
Most chat experiments prove nothing. How to change one thing, run it long enough, and read the result honestly.
Why industry averages mislead, and how to build a baseline from your own data instead.
Tags turn transcripts into data. Design a small consistent tag set, tag without slowing agents, and turn the results into decisions.
Chat should move sales, not just satisfaction. Define a conversion, attribute it honestly, and use the number to prove chat's value.
Some visitors leave before anyone answers. Why it happens in that first window, what reduces it, how to recover misses.
Every transfer risks a dropped chat and a repeated story. Cut needless handoffs without trapping chats with the wrong agent.
Solve it in one chat, no follow-up needed. Why it beats raw speed, what raises it, and how to measure it honestly.
How much does chat prevent heavier tickets? Define deflection honestly — and remember it's only good if CSAT holds.
The metric visitors feel most: what counts as a first response, why it beats resolution time, and how to cut it.
An SLA you can't staff is a promise you'll break: realistic targets, honest coverage windows, and clean measurement.
Capture intent, context, resolution, tags and CSAT from every chat — and use it without breaking privacy.
Visitors leave for a few specific reasons — silent waits, form walls, dead ends. Read the pattern and close each leak.
Ask one tap at the right moment, read the comments, and connect each rating to the transcript that earned it.
Run QA as coaching, not surveillance: a small scorecard, a sampling rhythm, and one improvable habit per session.
Cut the wait visitors feel before you hire — answer the first message instantly, deflect the repeatable, and set expectations out loud.
First response time, missed chats, and resolution beat chat volume every time — and two popular numbers actively mislead.
Cost per resolved conversation on one side; assisted revenue, deflection value, and retention saves on the other — claimed honestly, with the caveats stated.
Five transcripts, four questions, one owner, thirty minutes — and every finding routed to a fix: a doc, a canned response, AI training, or a product note.
Which three numbers to hold steady, how to add the story the metrics cannot carry, and how to report a bad month without losing the room.
Where AI answers well, and how good content stops questions from being asked.
The separator rule, the three frontmatter keys that mean anything, where the title comes from when you do not give one, and the AI flag set on every imported article.
A monthly budget, a daily token limit and a per-minute call limit, what each counts, and the honest answer to whether reaching one stops anything.
A pack is a one-off purchase credited by two independent paths that cannot double-charge you. What buying one changes, and why the monthly cycle decides how much it is worth.
Four signals feed one ranked backlog: repeated questions, handoff reasons, empty help centre searches and stale URL sources. Deduplicated, capped at thirty, and read-only.
The catalog is a fixed reference card and the active badge follows two narrow rules. The usage table underneath is the part that cannot be wrong.
Closed tickets are the answers your team already wrote. How candidates get shortlisted, what the privacy label is actually measuring, and why nothing is promoted for you.
Nine possible sources, a six-chip ceiling, and a set of three defaults nobody configured. Why editing Conversation Starters sometimes changes nothing.
The console asks the server whether an AI session is live, caches it for ten seconds, and treats every failure as a no. Why that last part matters.
What the Suggest button reads, why it can never invent a tag, and the reporting quirk that makes reviewing suggestions worth doing rather than optional.
An opt-in setting hands the visitor to a human when the conversation turns. What it reads, when it refuses to fire, and the limitation to know about.
One timeline built from four separate sources, each capped and each able to fail silently. How to read it without drawing the wrong conclusion.
Fifty URL sources per click, oldest first, and a failed fetch still marks the source as checked. The mechanics behind the button.
The AI overview scores your setup out of four. What each step measures, and what the score cannot see.
Two ways to pay for AI chat: a managed plan with included replies, or your own provider key billed direct. What actually differs.
Load your site into the assistant fifty pages at a time, then keep it fresh — including the quiet failure that leaves a broken page serving stale content.
Let the assistant read recent conversations as examples — and understand the two switches, the six-conversation ceiling and the redaction that happens first.
What the ticket copilot reads, the two modes it offers, and why nothing it writes reaches a customer until you press a button.
Every time the AI handed a visitor to a human, and why. Including the counter that is not really an error count.
Where the AI reply name and picture come from, what the default really is, the photo limits, and how to name an assistant without misleading anyone.
Your assistant already knows where it fails. How four separate signals become one ranked backlog, and how to work it without writing pages nobody needed.
Asking the sandbox one question proves nothing about the next fifty. A graded scenario suite is how you find out whether a change helped or quietly broke something.
Rehearse the assistant against your own settings and knowledge sources, and learn the one branch the sandbox can never show you.
The first screen decides who opens the assistant and what they ask it. The name, the disclaimer, the starters, and the exit.
An AI that can look things up is a different product from one that only answers. Which actions to start with, and where to draw the line.
AI first, agents first, or only when nobody is online. The three modes behave differently and the default is the cautious one.
One edit can fix an answer and quietly break three others. Versioning, a fixed question set, and knowing how to roll back.
The instructions that shape every answer, plus the presentation settings around them: name, greeting, starters, disclaimer and answer length.
What really happens when a visitor writes in a language you do not staff, why your own content is the limit on answer quality, and how to pick a language policy you can keep.
People will test the bot on purpose. Design the scope so the worst outcome is a dull answer, not a promise you have to honour.
Containment is easy to fake and handover is not failure. How to judge AI chat honestly, including what the dashboard actually counts.
What actually drives AI chat spend, which choices quietly multiply it, and how to keep the bill predictable.
Overnight is where AI chat earns its place and where a bad setup does damage unseen. Scope, handoff, and morning review.
Ten people asking the same thing is a signal, not a nuisance. Answer fast, then fix the root cause so it stops.
What to feed the bot, how to test it before visitors do, and the one configuration rule that prevents invented answers.
Deflection done honestly: fix the causes, answer where questions arise, layer KB + AI, and never make support harder to reach.
Title with the customer's words, answer in the first sentence, one task per article — and audit against real transcripts monthly.
Your transcripts are a map of failed page copy: put the answer at the point of doubt, use the customer's words, and watch the question volume drop page by page.
Every answer discovered mid-chat gets written where the NEXT agent will find it — canned library, help center, or the AI's training set. Run the bus-factor audit quarterly.
A false choice, honestly resolved: what humans answer better, what AI answers better, and the layered setup that uses each where it wins.
AI is genuinely good at some conversations and genuinely bad at others. The honest list of moments a human should take over.
AI replies need QA too, but you are fixing content and coverage rather than coaching a person. A practical weekly routine.
The most important AI decision is what it may not answer. Drawing the boundary, writing it into content, and testing that it holds.
What happens when AI cannot answer is what visitors actually judge. Designing the handoff so nobody repeats themselves.
AI changes what agents do all day, not just how much. Introducing it without the fear or the quiet resistance.
Training the assistant is a day of work. Keeping it accurate as prices, policies and products change is the real job.
The assistant will eventually give a wrong answer. What the agent who inherits that conversation should say next.
Assisting the agent is a different job from answering the visitor. What drafting and summarising help with, where they hurt, and why the agent still owns the send.
Using chat to qualify, handle objections, and turn interest into revenue.
Readiness screens find your AI actions by matching words in the name and description. What the matcher looks for, and the asymmetry that makes a counted tool show zero use.
Seven rows, a score that stops at 86%, and tool detection that matches on names rather than behaviour. How to read the lead capture screen for what it actually measures.
Queued does not mean sent. The recipient cap, the re-fire dedup rule, the UTC scheduling trap, and who actually delivers the message.
Doubts, price complaints and blockers are lifted from the visitor's messages into a list you have to clear by hand. How the extraction works and why it never auto-resolves.
How the console scores purchase intent from the visitor's own words, why one category can only count three times, and what the label is safe to be used for.
Nine conditions, measured in the visitor's browser. The quirks in how each one is counted decide whether your rule fires at all.
Three lists, two views, and an importer that never deduplicates. What to check before you paste a CSV, and how to keep the list worth searching.
Segments turn visitor records into a reusable audience. The nine fields, the operator families, custom attributes, and the preview count that keeps you honest.
An agent picker suits relationship-led sales more than volume support. The four settings, list and float layouts, and the fallback when nobody is online.
Telling a partner from an end customer, protecting partner pricing, and routing deals properly.
Every claim invites the same question. What counts as evidence, what you can send in seconds, and the honest answer when the reference they want does not exist.
Formal evaluations often start with a quiet chat question. What to answer now, what to put in writing, and how to find out where you are in the process.
Free-plan users have no trial clock, so the urgency playbook does not apply. Deciding what free support includes, which conversations are worth having, and when to ask.
A price rise sends your existing customers straight to the chat widget. Decide the message, the limits and the exceptions before the first one arrives.
The person chatting is often researching on somebody else’s behalf. Your real job is to equip them for a meeting you will not be in.
Purchase orders, invoicing, tax forms and vendor onboarding arrive in chat once the buying decision is already made. Handle them well and the deal closes; handle them badly and it stalls for a month.
Reading the buying signal, asking for the decision in text without pressure, and handling not right now.
How to answer a discount request without eroding your pricing, and what to offer when the answer is no.
The internal handoff loses more leads than any pitch. What to pass along, and what to stop sending.
A single, useful follow-up beats a sequence. What to send, when, and how to read the silence.
A first-time visitor knows nothing about you yet. Make a warm first impression and answer the newcomer questions.
"Too expensive" is rarely about the number alone. Find the real concern, frame value, and know when to hold the line.
"How do you compare to X?" means they're close to deciding. Answer honestly, confidently, without trashing rivals.
Highest intent, sharpest doubt. Answer the objection that blocks the purchase and convert pricing-page visitors.
The best upsell is a helpful one. Spot real opportunities, recommend without pushing, and know when not to sell.
A trial is a race against the clock. Catch stuck users, prove value early, and turn evaluations into customers.
Not every good lead buys today. Nurture the not-yet-ready, recognize returning visitors, follow up with context.
Most carts leave over one small doubt. Catch the hesitation, answer the real objection, recover without discounting.
Chat is often step one toward a demo. Qualify, book the right meeting, and keep the deal moving before and after.
Answer first, qualify inside the conversation, read behavioral signals over form fields, and hand sales a transcript instead of a row.
The feedback is already in your transcripts — harvest verbatims, ask one in-context question at the right moment, and close the loop where customers can see it.
Praise already happens in your transcripts — capture it, ask permission at the thank-you moment, route delight to review platforms, and let the best stories grow into case studies.
A choice made in the middle of a long chat scrolls away like everything else. This is how a sentence becomes a pinned decision, and the deferral rule that keeps maybes off the board.
Scripts and judgment for the conversations that repeat every week.
A hard bounce suppresses an address permanently and tells nobody. The dashboard banner that catches it looks at two conditions and is deliberately biased against warning you.
Articles store the category as text, not as a link. That explains the rename cascade, the second question on delete, and the two ways a category goes invisible.
Behind your tickets sits a per-person record keyed on email then phone. What creates one, what it will never overwrite, and how it differs from contacts.
A returning visitor arrives under a brand-new session id. Here is how the history card recognises them anyway, what each saved row holds, and the one thing it cannot do.
Five conditions decide the readiness dot, the recap quietly drops one category of question, and absent data always reads as good news.
Every let me check gets pinned next to the question that caused it, and stays open until a human closes it. What that list is worth to the people writing your docs.
Why a legitimate visitor sometimes cannot send you a message. The field-by-field rules, the ceilings nobody documents, and the four errors you can reword.
A crawl of your own site, your AI training sources, or pasted markdown. What each importer creates, and what it silently skips.
The form asks for four things and the record holds seventeen — including the page URL that usually explains the message on its own.
Every save snapshots the article as it was, and the rollback itself is reversible. Here is what the history keeps and what it cannot give you back.
One link covers every ticket for an email address for thirty days. The other opens one ticket once. Choosing wrongly is a support ticket of its own.
Name, tagline, logos, accent colour and six layout switches. Plus the honest limit: a tenant help center is set to noindex, so treat it as a support asset rather than an SEO one.
The search box covers tickets and knowledge base articles, matches on substrings, and orders by recency rather than relevance. Knowing that changes how you type.
Your support address can feed the same queue your chats do. What the forwarding setup actually involves, and what each audit outcome is telling you.
Connections drop and tabs close mid-conversation. What happens to the chat, what the visitor keeps, and how to handle the reconnect.
Let visitors search your help articles without leaving the chat window, and read the searches that found nothing.
These chats start angry and end badly when handled with a script. What you can safely say, what chat cannot prove, and how to route it properly.
During a beta the product breaks, the docs lag, and every chat doubles as research. How to staff it, what to promise testers, and how to tell when the beta should end.
Plenty of chats end before the problem does. What deserves a follow-up, what belongs in it, and how to keep the thread out of one person's inbox.
Retiring a feature sends every affected customer to chat at once. What to prepare, what to say about the deadline, and how to handle the requests for an exception.
Every public chat widget attracts some junk. The goal is to spend as little time on it as possible without adding friction that costs you the genuine conversations.
Price rises and policy changes arrive in chat first. How to prepare the answers and hold the line without sounding like a script.
An out-of-stock item is a sale about to be lost. Offer real alternatives and capture restock interest to keep the customer.
A cancellation is a save opportunity in disguise. Handle it with grace, learn the real reason, and make leaving easy.
Gift buyers are anxious, time-pressed, and out of their comfort zone. Guide the choice and answer the deadline questions.
A bulk enquiry is a big-ticket lead. Qualify wholesale requests, handle custom pricing, and route serious buyers fast.
An unhappy customer is a chance to fix things before the bad review. Catch dissatisfaction early and resolve it.
A feature request is a gift of insight. Capture it without overpromising, route it to product, and close the loop.
The sale is the start, not the end. Use chat after purchase to reduce remorse, drive adoption, and earn repeat buyers.
Sometimes a chat is better finished by phone. Offer and schedule callbacks smoothly, and make sure they happen.
A quote request is a warm buying signal. Qualify it, set honest timing, and follow up so it turns into a sale.
A warranty claim is a trust moment when something broke. Clear policy, empathy, and a smooth path that keeps the customer.
Your best customers deserve recognition, not the back of the queue. Use chat to know them and earn their loyalty.
Confirming who you're talking to protects everyone. Verify identity in chat safely, and know where to draw the line.
A refund request is a trust test, not just a transaction. Handle it with clear policy, empathy, and an eye on the why.
Some conversations outgrow the chat box. Recognize when a call serves better, switch smoothly, don't escalate by habit.
'Where's my order?' is ecommerce's #1 question. Answer fast, verify safely, and cut the volume at its source.
A return request is a retention moment in disguise. Handle it with clear policy, zero friction, and an eye on the why.
Billing chats are high-stakes: money, trust, a frustrated customer. Handle upgrades, cancels, and refunds with zero surprises.
Reproduce, isolate, confirm the fix. Troubleshoot methodically, know when to move to a call, and document the hard ones.
Resolved is a judgment call, not a stopwatch: when to reopen, when to follow up, and why premature closes hurt CSAT.
Hand a chat to a ticket or email without a reset: carry the transcript, promise a reply time, and close the loop before you close the ticket.
There is a difference between an unhappy visitor and one mentioning their lawyer. This is what the risk radar looks for, why nothing clears itself, and how it holds up your wrap-up.
How chat pairs with your other channels, and what each traffic source needs.
Channel media is stored as the provider's link, not a file. The server fetches it against an exact host allow list and streams you bytes that can never run as active content.
One row per site, channel and month, bucketed by when the message landed rather than when it was sent. What gets counted, what does not, and why the number can undercount.
The visitor card shows the route someone took to reach you, the referrer that sent them, and a location reading that is a guess. Here is what each field is worth.
One POST per finished chat, with the transcript, and no retry if you miss it. The payload, the signature, and why the test button lies about the shape.
The retry schedule, the delivery log, and the reason the per-row Retry button will not touch a failed delivery.
Point a provider at one webhook URL, save two credentials, and inbound messages arrive as tickets. Here is what each switch really controls.
A text launcher lets an article, a footer or a nav item open chat. Two snippets for two states, tooltips that set expectations, and a choice between embedded and popup.
Outbound webhooks, honestly: the event list, signature verification, the retry schedule, and the receiver habits that keep it all boring.
Some places will never run your widget snippet. A plain link still gets people into a conversation, as long as you know what it cannot do.
No search engine rewards having a widget, and installing one badly can cost you. What actually connects chat to search, and how to use the questions you collect.
Webhooks before polling, a request budget you actually calculated, and backoff that reads the headers the API already sends you.
Audiences from a video or an episode arrive late, warm, and invisible to your referrer report. What that changes about how you greet and staff them.
A visitor on a 404 or a no-results page told you exactly what they wanted and got nothing. That is the cheapest rescue on your site.
Visitors sent by AI assistants arrive with an answer already in hand, sometimes an outdated one. How to spot them, correct it without friction, and hear what the machines are saying.
On-site chat and messaging apps solve different problems. Where each wins, what a new channel really costs, and how to choose without collecting all of them.
Someone heard about you from a colleague, a podcast or a community thread and typed your name in. They arrive with high trust and no referrer, and chat is often your only way to see them at all.
Print, packaging, events and signage send people with a question already formed. How to meet and measure that traffic.
Support inside the product: where to place it, what context to carry, and when interrupting a task does real harm.
Search traffic arrives mid-question and lands deep in the site. Using query intent and closing the mismatch gap.
Social traffic arrives curious, mobile, and easily distracted. Catch them before they bounce and convert casual interest.
You paid for every click, so a bounce is money lost. Answer objections fast and protect the spend behind each visit.
Blog readers are researching, not buying. Answer the deeper question and earn the lead without breaking trust.
A referred visitor arrives pre-trusted by someone they believe. Honor that borrowed trust and convert the warm arrival.
Onboarding emails point the way; chat answers when someone gets stuck. Pair them so new users reach the aha moment.
Docs answer the known questions; chat catches the rest. Fill the gaps, point to the right page, find what to write next.
A lead magnet captures interest; chat turns it into a conversation. Qualify downloaders and move leads forward.
The people in your chat are your most engaged customers. Recruit for interviews and gather research without a big program.
A chat is a warm moment to invite someone onto your list. Grow signups naturally by offering real value in return.
Chat and a forum solve different problems. How they compare on speed and reuse, and why many teams run both.
Text is the default; escalate to voice, video, or co-browsing only when the medium itself is the bottleneck — with consent and a clear exit.
Concurrency, records, and interruption cost — the real differences between the three channels, and the mix most small teams should run.
Sync the three events that change follow-up, attach transcripts by link not by blob, match identity by email with honest fallbacks — and build it in one afternoon, thinnest slice first.
The form asks visitors to wait; chat answers while intent is hot. Keep both — but let each do the job it is actually good at.
What changes when the business model changes.
Accessibility duties, plain language, records and no conversion metric. What public-sector chat has to get right that commercial chat does not.
Using chat as an internal helpdesk for IT, HR or facilities: where the widget goes, how to route it, and why it must never become an authentication surface.
Candidates and employers need opposite things. Split them at the door and qualify each side quickly.
MSPs juggle prospects and client incidents on one site. Triage into tickets instead of an unmanaged support channel.
Serving customers across time zones and languages is hard. Support international visitors with honest coverage and clarity.
Diners ask fast, practical questions: open now? can we book? delivery? Answer quickly and take pressure off the phone.
Prospects hesitate over schedules, pricing, and fit. Answer fast, book trials, and support members without tying up the desk.
Homeowners want a fast answer and a quote. Capture jobs, qualify urgency, and book visits before a competitor does.
Photography clients book on trust and timing. Answer availability fast, guide bookings, and handle gallery requests.
Insurance shoppers have anxious, detailed questions. Explain coverage, qualify leads, and guide quotes responsibly.
Travel buyers ask time-sensitive questions and plan in bursts. Answer fast, guide bookings, support before departure.
Buyers hesitate over the same doubts; students stall at the same lessons. Convert on the sales page, support learners inside.
Two customers, opposite needs. Support buyers and sellers, route each right, and handle disputes without picking sides badly.
No-shows cost real money. Confirm appointments, handle reschedules gracefully, and cut no-shows without nagging.
Local visitors ask practical questions: open now? in stock? help today? Answer fast and turn site visits into store visits.
Registrants ask time-sensitive questions and stall at checkout. Answer ticketing questions and cut registration drop-off.
Members expect faster, more personal help. Use chat to onboard, retain, and cut churn without drowning in tickets.
A hesitating donor often has one unanswered question. Build trust, remove giving friction, and answer with care.
Waitlist and pre-order buyers are your most eager and most anxious. Keep them informed and hold the excitement.
An event concentrates demand into a short window. Prep for the spike, answer in real time, and follow up on the leads.
One widget, two jobs. How support and sales use chat differently, and how to run both without confusing either.
Turn a chat into a booked slot: qualify, offer times, confirm, hand off — and know when a link beats live chat.
Turn trials into customers: in-product chat, help at the stuck moments, and a shorter path to the aha moment.
B2C plays a volume game (speed, deflection, AI-first); B2B plays a value game (depth, routing to competence, transcripts as deal notes). Know which you are — and steal from the other.
Answer the donation-page question at giving time, capture volunteer interest with routing, let the KB carry program FAQs — all workable on a genuinely free tier.
Local customers compare three tabs and pick whoever answers first. Cover the three deciding questions, capture after-hours requests with a promise, and let the phone stay for emergencies.
Handling personal data in a channel visitors treat as informal.
Thirteen event types are written, replies are not one of them, and four labels exist that nothing can ever produce. How to read the trail, and the gaps you should not treat as evidence.
The verification code appears after repeated failures from one address. What counts, what clears it, and why a shared office connection trips it for everybody.
Unsubscribe links carry a signature, unsigned ones are rate limited, and a single config flag ends the transition. The reasoning behind each half.
Support sometimes needs to see your settings. The alternative to knowing your password is a gated, time-limited, audited session that holds your rights and never an administrator's.
Ten failures in fifteen minutes, counted per identifier and shared across every sign-in door. It is not an account lock, it clears itself, and a correct password wipes it instantly.
Junk can be refused at submission, before a record exists. Why the filter rejects visibly rather than silently dropping, and why it fails open.
Two switches, one challenge box in the corner of the page, and a fail-closed rule. What each captcha setting protects, what it costs you in visitors, and which one to enable first.
Five stores, a preview step and one transaction. What the retention purge removes, what it deliberately leaves alone, and why it never runs on its own.
A single percentage built from nine mixed signals. Which rows reflect your configuration, which cannot, and why Ready sometimes means nothing has happened yet.
Four hosting regions, what moving between them actually does to your data, and the two conditions that decide whether the move can run at all.
Per-token usage history turns a scary revocation into a boring one, as long as you know which kind of job the chart cannot see.
The config test and the endpoint probe answer different questions, and one common result means we could not tell rather than everything is fine.
Run the dry run, read the per-category counts, and understand why an erasure keyed on a phone number alone leaves chat and AI records behind.
Size caps, extension lists, and the two rules that surprise admins: a blank list means different things in different places, and spreadsheets get through regardless.
Country, language and device rules decide who gets a widget at all. How allow and block mode really evaluate, and why adding a second rule widens the net instead of narrowing it.
What the rule box accepts, what a blocked visitor actually sees, and the two cases where an IP rule quietly stops applying.
Server region affects latency and jurisdiction. How to pick one, how to switch safely, and what a region choice does not settle on its own.
A fill-in template for the six decisions that matter, plus who signs it and what happens when it is broken.
Replay records the page, chat records the conversation. What each actually captures, and where the combination becomes a privacy problem.
You cannot verify age from a chat window. Collect less, give agents wording that accuses nobody, and know how to delete what you should not be holding.
What SSO fixes, what it leaves untouched, and how to roll it out without locking yourself out.
Read the availability commitment properly: the number in minutes, the exclusions that shrink it, and what a service credit is really worth.
Rivals read your site too. What is already public, what nobody should volunteer, and why being brisk with them costs more than answering.
AI answers mean conversation text reaches a model. Know what leaves, keep the wrong things out, and say so plainly in your privacy notice.
Someone mentions a lawyer, a regulator, or compensation. What to say, what to stop saying, and how to preserve the record before the conversation goes any further.
A growing list of states regulates the data a chat window collects. What notice, opt-out and deletion actually require of a support team, in plain terms.
Some sentences create obligations. Advice you are not qualified to give, guarantees nobody can keep, and speculation about causes, plus what to say instead.
What to record so you can reconstruct an incident, and how to review it without policing the team.
Integrations become the widest door into your chat data. Scoping, storage, webhooks, and rotation.
A public widget cannot prove who someone is. Where the boundary sits, and how to help without guessing.
Know what the widget collects, minimize pre-chat fields, set a retention window, and have an access/erasure process ready before the first request.
Per-agent accounts, least privilege, same-day offboarding, data hygiene in transcripts — the security that lives in your settings, not your vendor's brochure.
Some information should never reach a transcript. How to refuse it politely, verify identity without secrets, and redact what slips through.
Chat is staffed by people paid to be helpful and quick, which is what attackers count on. How to build agent instincts that resist pressure.
Who can read old transcripts, who can change settings, and what happens the day someone leaves the team.
Deciding how long to keep chat transcripts, and making sure deletion actually happens rather than staying a good intention.
What visitors deserve to know before they type, and how to say it without turning the chat window into a legal document.
The questions worth asking a chat vendor, the answers that should worry you, and which reassurances are mostly decorative.
Chat access is the one nobody remembers to revoke. A short routine for accounts, tokens, exports and the shared login you forgot about.
Screenshots resolve tickets fast, and every upload is an inbound file from a stranger. What to accept and how to handle it.
What to do in the first hour when an agent account is compromised or a transcript reaches the wrong person.
The console is only as protected as the machine it is open on. Practical device hygiene for support teams, including remote ones.
When a prospect asks how chat handles their data, the honest answer wins more deals than the impressive one.
One agent free forever, unlimited chats, and the widget live on your site in minutes.
Launch live chat, connect your knowledge base, and add AI answers when you are ready. No credit card, no trial clock.