Chapter 03

Spot the conversations that matter.

Your team cannot watch every AI-handled chat. It should not have to. Define the moments that deserve attention, then let Stand mark them while the conversation is still alive.

Get your StandRead the chapter

Stand Guidebook

Chapter 03 of 12

Field guide

What to learn in this chapter

Design a small label system that recognizes valuable conversation moments, applies clear rules, and separates live alerts from later learning.

Need an exact capability definition, plan requirement, or limitation? Browse the Feature Reference.

Signal design

01

Name the moment, not the department.

Weak labels sound like internal folders: Sales, Support, Product. Strong labels describe what is happening in the visitor conversation: high-intent buyer, confused evaluator, competitor comparison, migration blocker, unexpected use case.

A good label should make the next action obvious. If the action is "a rep may want to join now," notification makes sense. If the action is "review later," keep the label log-only.

Custom labels and their first-match notifications require Pro or Business. Open Stand-ins → the chosen Stand-in → Edit → Skills → Labeling. Enable the skill, add a name and matching prompt, turn on Active, choose whether to notify, and Save. The plan allowance shown beside Labeling limits the number of active labels.

Stand Labeling settings showing an active Design Partner Candidate label with a notification-on-match option.
Figure 3-1. A label is a judgment rule. The name, description, Active state, and Notify on first matched label setting define whether the signal is reviewed later or interrupts someone now.

Example

High-intent buyer

Situation
The visitor asks about pricing, timing, implementation, contract terms, or how to start.
Move
Notify the Stand-in owner so they can review context and offer to join.
Insight
Buying intent is often phrased as a practical constraint, not "I want to buy."

Example

Confused evaluator

Situation
The visitor asks what Stand replaces, compares it with support tools, or misunderstands the AI Stand-in model.
Move
Keep this label’s notification off and review it weekly. If a particular case needs live attention, create a separate, narrower label whose matching prompt describes that case.
Insight
Repeated confusion is page copy feedback, not merely a chat-routing problem.

Writing labels

02

Give the Stand-in examples and boundaries.

A label prompt should say when to apply the label and when not to. This matters because many conversations contain mixed signals. A pricing question from a student is not the same as pricing from a buyer with a rollout deadline.

Keep labels narrow. Three precise labels are more useful than ten vague ones, especially while you are still learning which conversations deserve interruption.

Label prompt

Competitor comparison

Apply when
The visitor compares Stand with Intercom, Tidio, Crisp, Chatbase, Drift, LiveChat, tawk.to, or a custom-built chat.
Look for
Questions about AI coverage, human takeover, pricing, free plans, setup effort, or whether Stand replaces a support suite.
Do not apply when
The visitor only asks a generic "what does this cost?" question without comparing alternatives.
Notify?
Enable notifications only after adding the live-project, existing-tool, budget, or migration condition to the matching prompt. Otherwise every first match qualifies for an alert.

Operations

03

Separate "notify me now" from "teach me later."

Notification fatigue is a design failure. Use notifications for moments where a real person might join while the visitor is still there. Use log-only labels for trends that belong in a digest or weekly review.

The notification option belongs to the label; there is no separate conditional notification rule. Include any urgency or page-context condition in the label prompt itself. A label’s first match in a conversation can notify its Stand-in owner; repeated matches do not send a fresh alert each time. Enable browser notifications in Chats and test both a qualifying and a nonqualifying conversation.

Notification-worthy labels

  • A human can add value inside the current conversation.
  • The visitor is still in decision or problem-solving mode.
  • The rep can understand the context from the label summary and transcript.
  • The label should not fire constantly on normal questions.

Practice

Try this next

  1. 01From the published website, run three controlled AI chats: one clear match, one similar question that should not match, and one genuine match phrased differently.
  2. 02Inspect the label and summary on the active conversation in Chats, then end each chat and find it in History. Try it out does not provide this saved-label review.
  3. 03With notifications enabled, confirm the matching case alerts the Stand-in owner and the near-miss does not. Repeated matches in the same chat should not create repeated first-match alerts.
  4. 04Tighten the matching prompt if needed, Save, and repeat in fresh conversations before relying on the signal.
SignalBuyer with timing
Notify nowYes, if a rep can help before the visitor leaves.
Review laterReview the transcript for objections and next-step language.
SignalRepeated confusion
Notify nowKeep off for the broad label; use a narrower rule if one case needs live attention.
Review laterUse the pattern to rewrite page copy or the Stand-in prompt.
SignalUnexpected use case
Notify nowUsually no, unless the visitor wants to talk now.
Review laterCollect examples for positioning and product discovery.
SignalCompetitor comparison
Notify nowYes, when the visitor has a live project or migration question.
Review laterTrack which alternatives appear before buying intent.

Questions

Common reader notes

Does a label automatically put a human into the chat?

No. A notification invites the rep to review the thread. Offer to join asks for the visitor’s agreement; Takeover now transfers the conversation directly.

Where do label matches show up?

Label matches can appear on active AI session cards, in history filters, and in AI digest summaries when enabled.

Continue the guide

Choose your next chapter.

Previous chapter 02

Train Your AI Stand-in

Give your Stand-in the right briefing and sources.

Next chapter 04

Join at the Right Moment

Take over directly or offer to join with consent.

Try the guide on one real page.

Create your account