How to set up an AI assistant for customer messages without writing code
You can set up an AI assistant for customer messages without code by briefing it the way you would brief a new hire: who you are, how you speak, where the facts are and what to do at key moments. In AdaptiveReply that is a set of forms and plain sentences, and nothing reaches customers until you publish.
By the AdaptiveReply teamPublished Updated 8 min read
Key takeaways
- Setup is a briefing, not a build: forms, your own sources and guidelines written in plain language.
- Behaviour is sentences you choose: rules in the form “When this happens, do that”.
- Nothing reaches customers until you publish: edits stay a draft, you try them in the test chat, and version history lets you go back.
What does “no code” mean here?
No code means you configure the assistant by describing things in words and choosing options. There are no scripts to write, no decision trees to draw and no prompts to engineer.
Most owners know the scene. The shop is closed and the phone is still lighting up: how much is it, do you deliver to my city, are you open on Sunday. The same ten questions, every night, answered by hand. Automating them sounds like a software project with a developer and a budget, so nothing changes and the evenings stay the same.
Think of a new colleague’s first day instead. You tell them what the business sells, how to address customers, where the price list is and when to call you. An assistant needs the same four things, written down once.
The five setup steps
Setting up an assistant in AdaptiveReply takes five steps, and each one is a screen with a plain name.
- Describe your business. When you sign up you answer a few questions: what you sell, when you are open, how you handle payment, delivery and returns. AdaptiveReply builds a first assistant from your answers.
- Set the Personality. Give the assistant a name, choose a tone (friendly, neutral or formal), a reply length and a reply language, and add guidelines in your own words about what it must always or never do. See keeping replies in your brand voice.
- Add Sources and Business info. Tick the catalogue and the knowledge sources the assistant may look things up in, and fill in short facts such as shipping, returns and payment. Products can come from your store; see connecting a store catalogue.
- Write Rules. Decide what the assistant does at key moments: when someone wants to buy, asks for a person or is unhappy.
- Try it, then publish. Talk to the assistant in the Test chat, choose the Channels it answers on, and press Publish.
Rules written as sentences
A rule is a sentence with two halves: When something happens, Do something. You pick both halves from a list, so a rule reads the way you would say it to a colleague.
| When… | Do… |
|---|---|
| The customer wants to buy | Ask for name, phone and address, then hand the conversation to an agent |
| The customer asks for a person | Hand the conversation to an agent without asking further questions |
| The assistant cannot answer | Hand the conversation to an agent |
| The customer is unhappy | Hand the conversation to an agent |
| The customer asks about a topic you name | Give a fixed reply, word for word as you wrote it |
| The message comes from an ad | Add a tag to the customer |
You do not have to start from an empty page. The Rules screen offers ready-made rules such as Asks for a person, Does not know the answer, Outside working hours, Wants to buy and Unhappy customer. Pick one, adjust the wording, done. Other conditions include the channel, the customer’s first message and a customer who has not replied for a while.
Draft and live: change things without customers noticing
Every assistant has two versions: the draft you edit and the live version customers talk to. Your edits change only the draft.
- Save draft keeps your changes so you can try them in the Test chat. Customers do not see them yet.
- Publish makes the saved draft the version your customers talk to.
- The Answer customers switch decides whether the assistant answers at all, and it takes effect the moment you flip it.
The status next to the assistant’s name always says where you are: Draft, Live, or Live with unpublished changes. Version history keeps a copy of every save; when you restore an older version, your current draft is kept as a new version, so you can always go back.
What will the assistant not do?
Some rules are fixed for every business and cannot be switched off, because an assistant that improvises is worse than no assistant.
- It does not invent facts. Product, price, availability, delivery and opening-hours information must come from your catalogue, your sources and your business info. If a box is empty, it says it does not have that information.
- It does not claim an action that did not happen, such as a booking that was not made.
- It does not recommend competitors.
The product also has limits that are better known in advance:
- The assistant does not write AI replies to public comments. Comments get fixed texts by keyword.
- It does not create an order in your store. It records the request for your team to confirm.
- It reads text and photos, but not the contents of videos, voice notes or files.
- Availability means “available” or “not on sale”; it does not quote stock counts.
When none of your sources covers a question, the assistant does not guess. It follows your rule, which usually means handing the conversation to your team as a case; see how handoff works. Like any AI model it can still misread a question, which is the reason to test before you publish.
Start with ten real questions
The quickest route to a useful assistant is ten questions your customers really asked last week.
- Copy ten real questions from your messages, exactly as customers wrote them, typos included.
- Ask each one in the Test chat. Nothing there reaches real customers.
- Fix what you see: add a missing fact to Sources or Business info, add a guideline when the tone is off, add a rule when a person should step in.
- Repeat until all ten are right, then press Publish.
- After a week, read the cases that were opened and add the answers that were missing.
AdaptiveReply is free to start, without a credit card. Create your account and brief your first assistant, or see how it works step by step.
Frequently asked questions
Do I need a developer to set up an AI assistant?
No. In AdaptiveReply the setup is a series of forms and guidelines written in plain language: Personality, Sources, Business info, Rules and Channels. Connecting Instagram, Facebook and WhatsApp uses Meta’s own login screens. A developer is only useful if you want the assistant to query your own systems through an API connection.
Where does the assistant get its answers?
From your catalogue, your business information and the knowledge sources you add, such as FAQ and policy texts. AdaptiveReply instructs the assistant to take product, price, delivery and opening-hours facts only from those. When none of them covers a question, it does not guess; it follows your rule, which usually means a handoff to your team.
Can I change the assistant without affecting customers?
Yes. Every edit goes into a draft. Customers keep talking to the live version until you press Publish, and you can try the draft in the Test chat first. Version history keeps a copy of each save, so an earlier version can be restored if a change turns out worse than expected.
What happens when the assistant cannot answer?
It follows the rule you set for that moment. With the ready-made rule “Does not know the answer”, AdaptiveReply hands the conversation to your team as a case with a summary, and the assistant stops replying in that conversation while the case is open. Outside business hours it tells the customer that a person will follow up.
This guide is part of a series. Start with the overview: One inbox for Instagram, WhatsApp, Facebook, email and web chat: why scattered channels lose customers