How-to · Shopify chat best practices

12 best practices for running Shopify chat in 2026.

Practical playbook for getting chat right — greetings, response time, escalation balance, knowledge base hygiene, voice tuning, measurement. From what we’ve learned watching merchants run Clearly Agent on real storefronts.

Works for any Shopify chat tool, not just Clearly Agent. Principles, not product features.

The 12 practices

  1. 01

    Write greetings that ask a useful question

    Default "Hi! How can I help?" gets 5% engagement. "Looking for the right size?" on a product page gets 25%. "Questions about ingredients or shipping?" on beauty. Contextual + useful > generic. Test 3-5 greetings on different page types.

  2. 02

    Aim for sub-60-second response on escalations

    Customers in chat wait 1-2 minutes max. Above that, they leave. With mobile push, response should be 5-30 seconds for high-value escalations. If you can't meet this, configure clear out-of-hours messaging.

  3. 03

    Let the AI handle 70-85% — escalate only the hard ones

    Over-escalating creates inbox bloat. Under-escalating frustrates customers on hard cases. Right balance: AI handles product Q&A, sizing, shipping, order tracking, returns policy. Humans handle complaints, custom orders, B2B inquiries, high-value carts in hesitation mode.

  4. 04

    Keep replies short — chat isn't email

    Customers in chat expect 2-3 sentence replies, not paragraphs. Configure brand voice instructions: "keep replies under 3 sentences unless explaining a policy." Long replies feel like a lecture; short replies feel like a conversation.

  5. 05

    Knowledge base hygiene every 30 days

    Pricing changes. Shipping rates change. Product specs change. Bad KB info is worse than no KB info — agent confidently gives outdated answers. Block 30 min/month to audit KB content; flag stale entries.

  6. 06

    Treat the escalation queue as KB improvement signal

    Every escalation is a question your KB didn't answer. Review weekly. Add answers for repeating questions; tighten triggers for false-positives. The KB improves on its own loop if you watch this.

  7. 07

    Honest about AI vs human, on customer ask

    When customer asks "am I talking to a bot?", answer honestly. Customer-friendly disclosure beats AI pretending to be human. Most customers don't care once the answers are good; they ONLY ask when the AI is failing. Be transparent.

  8. 08

    Match brand voice to your category

    Luxury: serious, no emoji, restrained. Streetwear: casual, contractions, occasional emoji. B2B: formal, professional. Beauty: warm, knowledgeable. Pet: friendly, enthusiastic. Pick the voice that matches your product pages and apply consistently in chat.

  9. 09

    Surface free shipping thresholds proactively

    Customers at $55 with free shipping at $75: agent should mention it WITH 2-3 affordable add-ons. Don't leave it to the customer to figure out. Conversion on these prompts is 25-40%; AOV lift is real.

  10. 10

    Use customer memory across sessions

    Returning customer who said "allergic to nickel" 3 months ago shouldn't see nickel-containing jewelry. Memory persists per visitor; configure your agent to leverage it. Personal experience compounds across visits.

  11. 11

    Measure conversion-per-chat, not chat volume

    High chat volume isn't the goal — high revenue-per-chat-session is. Track in Pro dashboard: which conversation types convert, which UTM campaigns produce highest-AOV chats. Optimize for value, not volume.

  12. 12

    Run reviews of the agent's actual replies monthly

    Read 20-50 random chat transcripts from the past month. Note: where is the agent good? Where is it weak? Where are escalations getting handled poorly? Use as direct input to next month's KB + voice tuning.

FAQ

How long should an average chat conversation be?
2-5 turns total for transactional questions (sizing, shipping, tracking). 5-10 turns for product exploration (recommendation flows, comparisons). 10+ turns suggests the agent is failing — investigate KB gaps.
Should I have a human respond instantly or let the AI handle?
Let the AI handle first. Even at 1-second response, the AI is faster than a human typing. Step in when the AI hits its limits, not as a default. Customers don't need humans for "what size?" — they need humans for "this is broken."
How do I know if my chat is performing well?
Three KPIs: (1) engagement rate (% of widget opens that produce a customer message) — aim for 25%+; (2) conversion rate of engaged conversations — aim for 8%+ on commerce sites; (3) escalation rate — aim for 15-30% (lower = AI handling well, but too low means the AI is over-confident).
What if my brand voice doesn't match any tone preset?
Use the free-form brand voice instructions field. Examples: "Speak like a 1960s travel postcard: warm, evocative, never urgent." Or: "Casual Brooklyn coffee shop voice — lowercase, contractions, no exclamation marks." The agent applies whatever you write strictly.
Should I disable the AI when I'm online?
No. The AI handles the 70-85% of conversations that don't need you. You join the 15-30% that do. If you take over every conversation manually, you're losing the value of having AI installed. Trust the agent on routine questions; step in on hard ones.
Install Clearly Agent →Customize tutorial →