Can I customize how the AI responds?

Yes. Simple Chat lets you steer AI behavior with custom instructions in plain English—from global tone down to how individual tools should act. You can also define business-specific rules for orders, returns, shipping, and other topics your policies care about.

Think of it as teaching your support playbook to the automation layer, not editing code.

Layers of customization

Merchants typically configure several instruction surfaces:

• System prompt (overall behavior) — Voice, priorities, what the AI must never do, and when to escalate to humans.
• Product recommendations — How aggressively to suggest items, what to avoid recommending, and fit or compatibility language.
• FAQ handling — How to use your FAQ content, when to quote policy verbatim, and when to ask clarifying questions.
• Article references — How blog or help-center material should be cited or summarized.
• Per-tool instructions — Guidance for each integrated tool (order lookup, tracking, etc.) so automated actions match your SOPs.

Together these layers prevent one generic prompt from trying to cover every scenario poorly.

Business rules for sensitive topics

Beyond tone, you can set rules aligned to how your store actually operates:

• Orders — What the AI may confirm vs what requires staff (changes, cancellations, edits).
• Returns and refunds — Windows, exceptions, and language that must not over-promise.
• Shipping and fulfillment — Carrier expectations, international limits, preorder handling.
• Other recurring themes your team documents today in macros or internal wikis.

Write rules the way you would brief a new agent: specific, testable, and tied to outcomes (“if order is unfulfilled, do X”).

A practical rollout process

1. Draft your system prompt with brand voice and escalation triggers.
2. Add FAQ and article instructions so answers stay grounded in published content.
3. Tune product guidance before peak season so recommendations feel helpful, not pushy.
4. Configure each tool with limits (for example never issue refunds autonomously if that is your policy).
5. Run test conversations on staging or low-traffic hours; read threads in the dashboard and adjust wording.
6. Review Analytics validation metrics after a week—rejections often point to one instruction gap, not “bad AI” overall.

Customization is ongoing: when policies change, update instructions before marketing announces the change.

How customization interacts with staff

Instructions shape AI messages; staff takeover still wins when a human sends a customer-visible reply or turns AI off for a thread.

Use internal notes to flag when a human overrode policy so the next editor knows the conversation history includes an exception.

Well-tuned instructions reduce escalations; clear handoff rules keep exceptions safe when automation should stop.