From Website Chat to Co-Browsing: A Step-by-Step Flow for Support Agents
How an agent goes from chat to the customer's screen: live visitor radar, request, consent, pointing and guiding, masked fields and a 10-minute session.
In this article
The visitor has typed "I can't find where to add the second traveler" for the third time, and your agent has sent three screenshots that do not match what the visitor sees. This is the moment where website chat to co-browsing stops being a feature on a pricing page and becomes the fastest way to end a ticket. Instead of describing the screen, the agent looks at it with the customer and points.
Plenty has been written about what co-browsing is and why it is safer than screen sharing. This article is the operational side: the script an agent follows from the chat message to the closed session, minute by minute, with the phrases that work, the rookie mistakes that make customers decline, and the point at which a video call is the better tool. The customer never installs anything; everything happens inside the widget they already opened.
Minute 0: spotting the moment in the chat or on the radar
Co-browsing starts in one of two places.
From the conversation. Certain messages signal that words will not solve it: "the button doesn't do anything," "it says invalid but everything is filled in," "I don't see that option." Two exchanges like this and the agent should stop typing instructions.
From the live visitor radar. The radar shows who is on the site right now and on which page. A support lead at a US insurance broker sees a visitor sitting on the quote form for eight minutes, opens a chat from the radar, and offers help before the visitor gives up. The chat comes first; the request comes after they reply.
Either way, the agent introduces the idea before sending the request, so the consent prompt does not appear out of nowhere:
I can see your screen with you for a few minutes if that helps; you'll get a prompt to accept, I only see this browser tab, and anything like passwords or card numbers is hidden from me. Want to try?
Minute 1: request and consent
The agent sends the request from the conversation. The customer sees a prompt in the widget and accepts or declines; nothing starts until they accept, and there is no way for the agent to force it.
Three things to say or do while waiting:
- Tell them what to click ("a small window will appear at the top of the chat").
- Say how long it takes ("this usually needs two or three minutes").
- If they decline, do not ask again in the same chat. Offer a screenshot walkthrough or a video call instead.
The rookie mistake here is sending the request silently. An unexplained "an agent wants to view your screen" prompt tends to get a "no," and you have burned the option for that session.
Minutes 2-8: pointing and guiding, not driving
Once accepted, the agent sees only the tab with your site, not the customer's desktop, other tabs or notifications. They can highlight elements and guide clicks; the customer stays in control of their own browser.
The rhythm that works is narrate, point, wait:
I'm highlighting the "Add traveler" link, just below the first passenger's date of birth. Click that and a second block should appear.
Wait for the block to appear, confirm it, move on. Agents who race ahead with four instructions at once lose the customer, who is reading the chat and the page on the same screen.
What the agent cannot do is as important as what they can:
- Passwords and form fields are masked by default. The agent sees a blank or hidden box where the customer types their card number or password.
- The agent cannot type into masked fields. Even if the customer says "just enter it for me," the tool does not allow it; the agent says so and the customer types.
- The site owner can mask more elements (account balances, medical details, addresses) in the widget settings, so what is hidden matches your own policy.
For a deeper look at what the agent sees and does not see, read consent and masking in co-browsing. For the flow, the rule is simple: the agent points, the customer types.
| Step | Agent does | Agent says | Common mistake |
|---|---|---|---|
| Spot | Notices a visual problem or a stuck visitor on the radar | "I can see your screen with you if that helps" | Sending three more screenshots |
| Request | Sends the request from the chat | "You'll get a prompt to accept" | Requesting with no warning |
| Consent | Waits; respects a decline | "Take your time, it only shows this tab" | Asking twice |
| Guide | Highlights, guides one click at a time | "Click the link I'm pointing at" | Four instructions in one message |
| Masked field | Stops and lets the customer type | "That field is hidden from me, go ahead" | Trying to explain the masking mid-form |
| Close | Ends the session, summarizes in the chat | "Done, I've ended the session, here's what we did" | Leaving the session open while chatting |
Minute 9: watch the clock
A session has a 10-minute maximum. That is deliberate: co-browsing is for unblocking a specific step, not for a guided tour of your product. Two ways to use the limit well:
- Before starting, name the goal. "Let's get the second traveler added and the quote showing; then you can finish the rest on your own." A named goal keeps the session inside the window.
- If it runs out, close cleanly. Explain what was completed, what is left, and how to finish it. If needed, request a second session for a second, specific goal, rather than restarting the same one.
Minute 10: close and log
Ending the session is a click on the agent's side, but the closing message in the chat is what the customer remembers:
I've ended the screen session, so I can't see your page anymore. Recap: we added the second traveler and the quote is now showing at $412. The card details are yours to enter; once you pay you'll get an email confirmation. Anything else while I'm here?
Every session leaves a record on the contact's timeline in the shared inbox, alongside the chat. A colleague picking up the same customer tomorrow sees it happened; a supervisor can review how often sessions are needed on a given page (a product signal, not just a support one); and if a customer asks what was seen, there is an answer. The CSAT survey and the email transcript at close work as for any conversation.
When to escalate to a video call instead
Co-browsing shows the agent your website in the customer's browser. It does not show the customer's face, their physical setup or anything outside that tab. Switch to a video call when:
- The problem is not on your site (a printer, a router, a device the customer is holding).
- The customer needs to be shown something on the agent's screen, such as a demo or a document.
- The conversation is emotionally charged and a face would calm it down.
- The visitor speaks another language and would benefit from live translation.
The agent creates the meeting from the same conversation and the customer joins by link in the browser, no app, no account, with screen sharing and live translation available.
Frequently asked questions
Can the agent start co-browsing without the customer knowing?
No. The customer sees a prompt in the chat and must accept; if they decline or ignore it, nothing is shared. The session also ends automatically after 10 minutes.
What does the agent see during a co-browsing session?
Only the browser tab with your website. Passwords and form fields are masked by default, the agent cannot type into masked fields, and the site owner can hide additional elements.
Does the customer need to install anything?
Nothing. The session runs inside the chat widget the visitor already opened; the agent requests it from the conversation and the visitor taps accept.
Is co-browsing included in the website chat plans?
Website chat is included in every plan; co-browsing is a paid add-on with tiers by number of simultaneous sessions. Check the co-browsing page for current tiers.
A script your team can run tomorrow
Print the six-step table, put it next to every support agent, and run the next stuck-visitor chat through it: warn before requesting, respect the decline, point instead of driving, let the customer type into hidden fields, close inside 10 minutes and summarize in the chat.
Co-browsing is an add-on on top of the website chat, which is included in every plan; you can try the whole flow with your team during a 7-day free trial.
Ready to try it in your business?
Bring WhatsApp, Instagram, Messenger, email and calls into one inbox, and let AI handle the first reply. Set it up today.
7 days free. No credit card required.
Read more
-

WhatsApp Calling in the UAE: Providers and Availability
-

How to Give an AI Agent Personality and Tone
-

AI to Book Appointments From Chat: How to Automate It
-

The Best WhatsApp Business API CRM in South Africa (2026)
-

WhatsApp Business App Limitations Every Growing Company Hits
-

How to Generate Leads With Instagram for Your Business
-

How to Integrate WhatsApp with Shopify: Orders, Shipping, and Cart Recovery
-
Abandoned Cart Recovery for Canadian Online Stores
Comments (0)
No comments yet. Be the first to share your thoughts.
Leave a comment
Your email is never published. Comments are moderated before appearing.