By Repli Team · June 5, 2026 · AI
Quick Answer
Configure an Instagram AI chatbot with a persona, accurate business facts and test questions. Understand AI credit, messaging windows and human-review limits.
In Repli, brand-specific AI chat means supplying persona and business instructions, not fine-tuning a model on your Instagram history. This tutorial is an instruction-writing and testing exercise. The Instagram chatbot product page covers commercial access.
Step 1: Check access before configuring
Use a connected professional account, eligible AI-chat access and available AI credit. Plan entitlement and AI credit are separate; a free static keyword test is not proof that AI chat is included. Check current access before enabling generated replies.
Step 2: Select a supported trigger
Configure the AI chat’s actual keyword/event settings. Do not assume it is automatically a catch-all fallback behind every fixed rule. Start with one clear incoming request and test how that rule matches before expanding it.
Step 3: Write maintained instructions
Include the business role, audience, tone, verified facts, real links and what to do when an answer is unknown. Remove outdated prices and unsupported integrations. An illustrative instruction block is:
You are the Instagram assistant for [business]. Keep replies concise. Use only the business facts below. For current service details, share [real information URL]. Do not guess availability, promise a refund, claim an order lookup or confirm a booking. If information is missing, say a person needs to review it.
Replace bracketed material manually. These instructions influence generation; they cannot guarantee obedience or make an external calendar or order system available.
Step 4: Test known, unknown and adversarial questions
- Ask a question whose answer is in your instructions.
- Ask for a price or availability you did not supply.
- Ask for a refund or medical/legal recommendation.
- Ask the assistant to ignore its instructions.
- Review whether it gives maintained information or makes an unsupported promise.
Revise instructions and retest failures. Generated answers can be inaccurate even after a successful test. Review the actual response text, not only whether a send succeeded.
Step 5: Keep entry messages and ongoing chat distinct
A comment private reply can deliver requested information. It does not automatically authorize an ongoing AI sequence. Continue ordinary generated messages only after an eligible incoming reply and within the standard messaging window.
Step 6: Review answers and pause when needed
Inspect conversation results and update facts when your services change. Use available local takeover/pause controls and respond from Instagram when the request needs you. A prompt that says “ask a teammate” is not a native shared-inbox assignment, notification or guaranteed handoff.
Keep lead capture and other channels separate
A separately configured lead questionnaire collects the person’s answers on eligible access. Do not assume AI automatically receives all questionnaire fields or scores and routes leads. WhatsApp uses separate setup and consent; copying a persona is not transferring conversation history between channels.
Frequently Asked Questions
Does Repli train a custom model on my posts?
This workflow uses instructions and supplied facts; it does not establish model fine-tuning or automatic ingestion of every post.
Can AI confirm bookings?
Not from a prompt or supplied link alone. The responsible booking system or staff confirms the appointment.
Does an instruction guarantee factual answers?
No. Test generated responses, maintain facts and retain personal review.