YundaDeskThe website widget is where most of your customers speak up for the first time. She’s stuck at checkout, or torn between two sizes on a product page, and clicks the little window in the corner. How fast and how accurately you answer that first message often decides whether she checks out or closes the tab.
Widget conversations have a distinct shape: the questions repeat constantly, but the customer has zero patience. Where’s my order, does this size run true, why isn’t my discount code working — these get asked dozens of times a day, with barely any variation, which makes them a near-perfect fit for AI to answer first. When the AI can’t answer, or the customer explicitly asks for a human, it hands off. The scripts below are ready to paste into a chat window, and just as ready to drop into your knowledge base.
Why widget scripts deserve their own playbook
The widget behaves differently from email or social DMs. Customers stay for seconds, not minutes, and expect a near-instant reply — if the back-and-forth stretches past three exchanges, she may just leave. That’s why widget scripts need to be shorter, more direct, and lighter on preamble than scripts for any other channel.
The widget also sits at the shortest point in your conversion path — she’s still on your site, the tab is still open. Answer well here and you can push a sale through right then. Answer slowly or wrong, and the drop-off is silent; you’ll never know exactly why she left.
To wire these scripts into AI support systematically, start with how to build a knowledge base that actually feeds your AI. Widget FAQs are just one category of entry — they share the same knowledge base and the same customer profile as your email and social scripts.
Website Widget Support Scripts and FAQ Templates: put channel response times on one scale
Shipping: the three questions that come up constantly
Shipping questions dominate widget traffic, and customers usually arrive a little anxious — keep the tone steady and specific.
| Customer asks | Suggested reply |
|---|---|
| Where’s my package | Please share your order number or the email you used, and I’ll pull the latest shipping status right away. |
| When will it arrive | Delivery windows follow what’s listed on the page, and you’ll get a tracking link once it ships. If it’s shipped but tracking hasn’t updated, that’s usually customs or carrier handoff — delays there don’t typically affect final delivery. |
| Can you rush it | Orders move through our standard processing flow. If yours has a specific timing need, tell me more and I’ll see what can be coordinated. |
Don’t promise a specific arrival date in the widget unless your knowledge base has confirmed timing data to back it up. Cross-border shipping runs through too many hand-offs — “it’ll definitely arrive tomorrow” carries a lot more risk than “processed on the stated timeline.”
Sizing and product questions: turn “I’m not sure which one” into a decision
Sizing questions aren’t really lookups — they’re decisions the customer needs help making. In the widget, she’s usually one step away from checkout, so a clear answer can push the sale through right there.
- Height/weight sizing questions: if the knowledge base has a size chart, give the recommended size directly and link the chart. Without an exact match, offer a knowledge-base-backed range like “most customers in this height and weight range choose a size M” — don’t invent precise figures.
- Comparing two products: list the material, dimension, and color differences directly instead of sending her back to compare pages herself.
- Stock questions: answer directly if the knowledge base has the data; if not, say plainly “let me confirm that for you” rather than guessing.
Returns and refunds: acknowledge the request, then decide on escalation
Returns are the widget topic most likely to escalate. Routine questions can be answered directly; anything involving a specific dollar amount, fault determination, or an upset customer should move to a human quickly.
Asking about the return policy
We accept returns within the stated window after delivery — you can see the full policy here. If you’ve already received the item and want to start a return, share your order number and I’ll check whether it qualifies.
Reporting a defective item and asking for a refund
Sorry to hear that. Please send your order number and a photo of the item, and I’ll log the details now. Refunds need a human to confirm the order and item condition first, so I’m handing this to a teammate to follow up quickly.
Following up on a pending refund
I understand you want this resolved soon. Refund approval takes a bit of time to verify the order details — I’m passing your request and everything on file to the team handling it, and you’ll hear back shortly.
Any refund or compensation involving money always goes through human approval and an audit trail — the AI never commits to an amount or executes it automatically. That boundary isn’t overcaution; it protects both the customer and the team.
Discount codes and pricing: explain the reason, don’t force an answer
A failed discount code is one of the fastest ways to frustrate a widget customer — she’s usually already filled her cart and gotten stuck right at the finish line.
| Scenario | Suggested reply |
|---|---|
| Discount code shows invalid | Common causes include not meeting the minimum spend, certain items being excluded, an expired code, or the code not stacking with other discounts. Send me the code and a screenshot of your cart and I’ll check what’s going on. |
| Checkout total doesn’t match expectations | Please send the product link, cart screenshot, and code — I’ll verify the calculation. If there’s a genuine display error, I’ll pass this to the team for a closer look. |
| Requesting a discount be reissued | I hear you. Reissuing a discount needs approval — I’m logging your situation and passing it to the team, and you’ll hear back soon. |
Don’t let the widget verbally “reissue” a discount or adjust a price on the spot, even when the conversation seems reasonable. That kind of action needs to go through approval — AI handles the first response, a human makes the call.
Idle chats and common exit scenarios
Widget conversations have a scenario email doesn’t: the customer just stops replying, or closes the window mid-conversation. It’s not really a “question,” but it still deserves a prepared line.
- No reply after a while: after a reasonable wait, one gentle follow-up like “feel free to come back if you have any other questions” is enough — don’t keep pinging.
- “Let me think about it”: a simple “sounds good, reach out anytime if anything else comes up” lands better than pushing “why aren’t you checking out.”
- “Can I talk to a real person”: hand off immediately with something like “connecting you now” — don’t stall the handoff with more AI replies.
An explicit request for a human is one of the most basic guardrails in the system — the moment it comes up, the switch should be unconditional, not something the AI tries to talk its way past.
Turning these scripts into knowledge base entries
A one-line script in a document is just a reference. It only does real work once it’s in the knowledge base where the AI agent can pull it by scenario. When you add one, structure it like this:
- Customer phrasing (collect as many variants as you can, e.g. “code not working,” “discount won’t apply”)
- Standard reply (adapt the template above to your store’s actual policy)
- What the AI can answer directly (factual, policy-based questions)
- What must escalate to a human (anything involving money, complaints, high emotion)
- Update trigger (recheck when a promotion ends or a policy changes)
When agents add new phrasing or handle new scenarios in real conversations, they can flag it — “correct the AI” — and it becomes a suggested learning update. It only takes effect after a manager reviews and approves it, and every change stays traceable, testable, and reversible with one click.
Widget scripts don’t need to be clever — they need to be fast and accurate. Answer the high-frequency questions in one clean line, draw the escalation boundaries clearly, and let the knowledge base and its controlled learning loop fill in the rest. That’s what website widget support is actually supposed to look like.