AI sales agent: how AI answers customers in social media DMs

An AI sales agent is the most common way to use AI for selling on social media. It answers customers in direct messages and comments, quotes prices and terms from your business knowledge base, asks what the person needs and leads them to a booking, a request or a payment, handing complex conversations to your team. It pays off when 3 conditions are met: the knowledge base has current prices, you decide in advance when a person steps in, and the chat follows each platform’s rules.

What is an AI sales agent and how is it different from a chatbot?

In short: an AI sales agent is a program built on a language model that holds a conversation with a customer in its own words, based on your knowledge base. A regular chatbot follows a pre-built script of buttons and keywords, so a question outside the script gets silence or an off-topic answer.

A language model is a neural network that reads text and writes a reply. ChatGPT, for example, runs on one. In sales, the model works within 2 limits: it answers only from the knowledge your business gave it, and it steers the conversation toward the goal the owner set.

Which sales tasks can you give to AI?

In short: hand AI the repetitive part of messaging: the first reply at any hour, questions about price, stock and delivery, finding out what the customer needs, sending the right materials and offering the next step. Decisions the business is accountable for stay with people.

• First reply. A customer writes in the evening or on a weekend and gets an answer right away, while they are still choosing.
• Price, stock, timing and delivery. The AI takes these from the knowledge base, so the numbers in the chat match the numbers in your posts.
• Finding out what the customer needs. A few questions about size, budget, city or a convenient time. The team member who joins later does not have to ask again.
• Materials for each stage of the purchase. A catalog, examples of work, an answer to a common doubt.
• Next step. Book a meeting, take a request or send a payment link.
• Replies to comments. A short answer under the post and an invitation to continue in direct messages.
• Messages to past customers. A reminder about a repeat order in a messaging app, within the platform’s rules.

For each of these tasks, the answer is already in what your business knows or in rules the owner set in advance.

What should AI not do?

In short: an AI sales agent must not promise results, invent prices or terms, pretend to be human or take on decisions the business is accountable for: contracts, discounts beyond the rules, refunds, handling complaints. It hands those conversations to your team.

• Promises of income, medical results or other outcomes. Social media rules prohibit such promises, and they mislead the customer.
• Prices and terms that are not in the knowledge base. If there is no answer, the AI says it will check and passes the question to a person.
• Pretending to be human. The customer should know they are talking to AI. In the European Union the law requires it.
• Contracts and special terms. Discounts beyond the rules, refunds, complaints and disputes are handled by your team.

Where does an AI sales agent get prices and terms?

In short: from your business knowledge base: your products, prices, payment and delivery terms, common customer questions and answers to their doubts, all in one place. The AI sales agent answers only from it, so the more complete it is, the more questions the agent can handle on its own.

What to put into the knowledge base before launch:

• Products and prices, including bundles and subscriptions.
• Payment, delivery and return terms.
• Opening hours, addresses and ways to book.
• Common customer questions and their answers.
• Customer doubts and an honest answer to each.
• Handoff rules: in which cases and to whom.

When a price changes, update it in the knowledge base the same day. It works best when your posts and the agent’s replies take numbers from one knowledge base: then the video and the chat show the same price. Once a week, review the questions the AI could not answer: each answer you add to the knowledge base covers similar questions that come later.

In Agentix, the AI sales agent and your posts draw on one knowledge base, and the chat picks up at the sales funnel stage the person came from.

Will AI replace your sales rep?

In short: on the front line, partly. AI answers routine questions around the clock and moves simple purchases to the next step. Complex deals, contracts, special terms and disputes stay with a person, and by the time a conversation reaches your rep, the customer’s need is already clear.

How much work to give AI depends on the mode. There are usually 3 modes:

Switch modes based on facts: count how many drafts your team sends without edits.

Which platform rules do you need to follow?

In short: each platform sets its own limits for automated replies. On Instagram and Facebook you have 24 hours to reply to a customer message, and you can answer a comment with 1 private message. WhatsApp is closed to general-purpose AI assistants, and a Telegram business bot can reply in chats that had an incoming message in the last 24 hours.

Automated replies on Instagram need a path to a person. Meta requires that it be possible to pass the conversation to a team member (by a keyword, by a button or when the bot does not understand the question) and that a person then continue the chat in the shared inbox.

The WhatsApp ban covers services that offer AI itself as the product. Meta told TechCrunch that the change does not affect businesses using AI to serve their customers. The terms themselves say a business may use an AI provider as its solution provider.

Article 50 of the EU AI Act has applied since August 2, 2026. If a person is chatting with an AI system, they must be informed, unless it is obvious from the circumstances. A simple approach works in any country: in its first message, the AI sales agent says it is an AI assistant and offers to bring in a team member.

How do you launch an AI sales agent step by step?

In short: start with the goal of the conversation and the knowledge base, connect only the channels where customers already message you, set the handoff rules and run in drafts mode for the first days. Turn on automated replies once your team stops editing the drafts.

1. Choose the goal of the conversation: a booking, a request, a payment by link or a product recommendation.
2. Build the knowledge base: products and prices, terms, common questions and answers to doubts.
3. Using each platform’s official sign-in, connect the channels where customers already message you, without sharing your password with the service.
4. Decide when a person steps in: contracts, discounts, complaints, refunds, a question with no answer in the knowledge base, a direct request from the customer.
5. Write the first message, in which the AI says it is an AI assistant and offers to bring in a team member.
6. Run in drafts mode for the first days: the AI writes, a team member checks and sends.
7. Once a week, review the questions the AI could not answer and add the answers to the knowledge base.
8. Move routine questions to automated replies once drafts go out without edits.

How can you tell whether the AI is selling?

In short: the main metric is the conversion rate, the share of conversations that reach the goal: a booking, a request or a payment. Alongside it, track the time to first reply, the share of conversations handed to a person and the list of unanswered questions. Together they show where AI helps and where it loses customers.

The last metric connects messaging with posts. When you see which video led to a payment, you can build the next content plan around what sells. How to compare social networks by inquiries is covered in Cross-posting to social media, and in Agentix, analytics shows a summary by post.

Common mistakes

In short: an AI sales agent most often fails because of an empty or outdated knowledge base, missing handoff rules or messages sent outside a platform’s reply window.

• Launching without a knowledge base. The AI answers in general terms or makes things up.
• An old price in the knowledge base. The video shows one price, the chat another.
• No way to reach a person. A customer with a complex question leaves, and on Instagram it also breaks Meta’s requirements.
• Messaging days later. The reply window has closed, and the message will not go through.
• A series of messages to someone who commented. Instagram allows only 1 private message.
• AI pretending to be human. The customer loses trust, and in the European Union it breaks the law.
• Judging by the number of chats. Lots of messaging without bookings or payments means the AI is not leading conversations to the goal.

How it works in Agentix

In short: the Agentix AI sales agent answers from the same knowledge base your posts are written from, and it knows which post and which funnel stage the person came from. It works in 3 modes and hands complex conversations to your team.

• Channels: Telegram, WhatsApp, Instagram, Facebook and TikTok; on Instagram, Facebook and TikTok, comments too.
• Modes: replies on its own; replies on its own and hands difficult questions to a person; only drafts replies, which go to the customer after your approval. You can take over any conversation at any time.
• The AI sales agent does not promise income or medical results, does not pretend to be human and does not sign contracts.
• In development: booking and a payment link right in the chat, materials for each funnel stage and handing the conversation to a CRM (Bitrix24, amoCRM) at the point set in the funnel.

More on the AI sales agent page.

Sources

1. Meta for Developers. Messenger Platform Policy Overview: the 24-hour reply window and the Human Agent tag; the rules also apply to Instagram.
2. Meta for Developers. Instagram Messaging: Private Replies.
3. Meta for Developers. Instagram Messaging: Human Agent Escalation.
4. Meta. Meta Terms for WhatsApp Business Platform, section 4.7.
5. TechCrunch. WhatsApp changes its terms to bar general-purpose chatbots from its platform, 18 October 2025.
6. Telegram. Bot Features: Business Bots.
7. Telegram. Telegram Business, 31 March 2024.
8. TikTok for Business. About API for Business.
9. EUR-Lex. Regulation (EU) 2024/1689 on artificial intelligence, Article 50.

Answers to customers from your knowledge base, a timely handoff of complex conversations to a person and a link between every chat and a post: this is how the Agentix AI sales agent works.