Max Igoshev.
Email

Chat support manager and site moderator: what an AI agent takes

A typical chat support and moderation job post: replies to customers in the chat and messengers, taking in tickets, checking comments and reviews on the site. The agent takes the routine questions and the first pass against the rules, and the operator keeps the difficult tickets and disputed decisions.

How the support agent works: channels and a knowledge base feed the agent, the agent works with the CRM or order sheet, the ticket log and the moderation queue, hands the whole conversation to an operator or moderator, and a weekly report closes the loop
Channels
Website chat, Telegram, WhatsApp
Tools
CRM, ticket log, moderation queue
Goes to a person
Refunds, complaints, bans
Rollout
Drafts for approval first

A portfolio demo. The scenario is built from typical job posts for a chat support manager and site moderator, not from work with a specific company. The dialogue below is written as a sample.

What the job post usually asks for

  • answer questions about orders, delivery, payment and returns in the website chat, Telegram and WhatsApp;
  • take in tickets and pass them to the right team;
  • check users’ comments, reviews and listings against the rules;
  • keep a ticket log;
  • collect common questions for the knowledge base.

What the agent takes and what stays with a person

The agent takes

  • answers to routine questions from the knowledge base, at night and at weekends too
  • order status from the CRM or a sheet, by order number
  • a ticket in the CRM with the topic, the customer's details and a short summary
  • a first pass on reviews and comments against the rules: publish, hold for review, pass to a moderator
  • a list of questions the base had no answer for

Stays with a person

  • refunds, compensation and exceptions to the rules
  • complaints and annoyed customers
  • the final call on banning a user or removing disputed content
  • changes to the rules and the knowledge base
  • replies that need the company's position

How it works

One agent and one knowledge base cover the website chat, Telegram and WhatsApp. The agent answers from the knowledge base: delivery, payment, returns, common questions. Given an order number it takes the status from the CRM or a sheet, and it opens every ticket with a topic and a short summary.

New reviews and comments on the site also come to the agent. It checks them against the site rules and sorts them into three groups: publish, hold for review, pass to a moderator. It only holds what the rules name outright, such as links to other stores or insults, and anything disputed goes to a person.

A refund, a complaint and anything missing from the base go to an operator with the whole conversation and a summary: who is writing, the order number, what the agent has already said and what needs deciding. The full map is the picture at the top of the page.

I build chat agents on Watbot, the platform my partner and I have been building since 2018. The knowledge base of such an agent can also be opened to operators in ChatGPT or Claude, as in the plugin case, so people and the agent answer from the same material.

Sample script

Sample script, not a log of a real conversation

Customer: Where is my order? Number 48213, it was supposed to come yesterday.

Agent: Checked: order 48213 was handed to the courier the day before yesterday, and the courier shows delivery today by 9 pm. I'll send the tracking link in the next message.

Customer: I don't need it any more, I want my money back.

Agent: I understand. You can refuse the order on delivery, and the money comes back under the store's rules. If you want to start a return now, I'll pass your request to an operator with this conversation, and they will reply in this chat.

Customer: Please do.

Agent: Done. The operator sees the order number and the whole conversation, so you won't need to repeat anything.

New review on the site: "Ordered twice, everything arrived, but check out this store instead" with a link.

Agent, in the moderation log: Held for review under the rule "links to other stores", passed to a moderator.

The operator got a ticket "return, order 48213" with a summary and the conversation. The moderator got the review tagged with the rule, and the final call is theirs.

What I need from you

  • the job post or the support and moderation task list;
  • answers to common questions, the delivery, payment and return terms;
  • the site rules for reviews and comments;
  • access to the CRM or order sheet and to the moderation queue;
  • who takes the tickets the agent hands over, and in which hours.

How we roll it out

  1. I build the knowledge base and the agent’s instructions from your materials and past conversations and connect the channels, the CRM and the moderation queue.
  2. The agent works in draft mode: an operator sees every customer reply and every review decision first, and approves or edits it. Every edit goes into the knowledge base.
  3. Routine replies and decisions the operator approved without edits, the agent sends without approval. Refunds, complaints and disputed moderation still go through a person.
  4. A weekly report arrives: ticket topics, what went to an operator and why, questions missing from the base and moderation decisions.

Knowledge base template

A file with the sections the agent reads before it starts: delivery, payment and returns, common questions, moderation rules, reply templates, what to hand to an operator. Filled in, it becomes a ready brief for an agent on your support team.

Download: template for chat support and moderation, a Markdown file that opens in any text editor.

Send me your job post