Yes, for the routine part. An AI receptionist can answer Instagram and Facebook messages about opening hours, services, how to book and what to bring, and it can sort the rest. A person must handle advice, complaints, fees that vary, anything sensitive and any message where the sender sounds upset or uncertain.
For a practice, the more useful question is which messages you trust the bot to answer without anyone seeing them first. For most practices the safe list is short, and the rest of the inbox needs a drafted reply that a human approves or a quick handover. Set that line deliberately before you switch anything on, because social inboxes are where a poorly judged automated answer travels fastest.
Which messages an AI can answer, and which it should not
Sort the last 50 messages your practice received on Instagram and Facebook into the groups below. You will probably find that most are in the first two.
| Type of message | Example | AI role |
|---|---|---|
| Fixed facts | "What time do you open?" "Do you have parking?" | Answer automatically from a written list |
| Process questions | "How do I book a first meeting?" "What should I bring?" | Answer automatically, with a booking link |
| Price or scope | "How much do you charge for a tax return?" | Answer only if the price is fixed and written down; otherwise route to a person |
| Advice | "Should I move my pension?" "Can I dismiss someone for this?" | Never answer; tell them how to book a conversation |
| Complaints and distress | "You gave me the wrong date." "I am really worried." | Handover to a person at once |
| Spam, scams and sales pitches | "We can get you 10,000 followers" | Sort to a folder for a person to clear |
| Press, partners, jobseekers | "I would like to interview your partner" | Route to the named contact |
Anything that involves a client's own details deserves caution. A message such as "Is my return finished?" cannot be answered by a bot that does not know the client, and a bot that tries to answer is guessing. Treat any message that names a client's situation as one for a person.
What the platforms require of automated replies
Platform rules are the part to check first, because they decide what you can automate and for how long. The following points come from Meta's Messenger Platform policy overview, which covers Messenger and Instagram messaging for businesses; check the current text before you rely on any of it, since terms are revised.
- Reply window. Businesses have a standard messaging window to respond, which for most user actions is 24 hours. Promotional content is allowed inside it. After it closes, what you can send is restricted.
- Human agent route. A "Human Agent" tag allows a person to reply manually within a separate seven-day period. The policy describes it as a route for manual replies, so do not send bot messages through it.
- Disclosure. Where the law requires it, an automated chat must tell people they are interacting with an automated service, at the start of a thread, after a long gap or when a chat moves from a person to automation. Meta also recommends disclosure even when not required.
- Responsiveness. The policy sets response-time expectations for fully automated bots on Messenger, including answering any input from a user within 30 seconds. Check whether and how that applies to your set-up.
- Some features differ by channel. Several Messenger features, including some message tags, one-time notifications and sponsored messages, are not available through Instagram messaging.
Practical consequence: tell people at the first reply that they are talking to an automated assistant and how to reach a person. Do not use an automated tool to send promotional messages outside the permitted window, and do not use the Human Agent tag for automated content. If your tool is a Meta-provided one, read its own terms. Meta Business Agent, for example, is billed per token on the WhatsApp Business Platform route (Meta's developer pricing page puts a typical message at around 4 to 5 cents), and that page does not cover Instagram or Messenger, so check the current pricing for your channel before you plan a budget.
For what gets accounts restricted in practice, the guide to Instagram DM automation safety goes through the behaviours that trigger limits, such as bulk sending and misuse of keywords.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowThree ways to set it up, from lightest to heaviest
Saved replies and away messages (no AI)
Start here. Meta's business inbox tools offer saved replies and simple automated responses such as an away message, though the exact options differ between the Instagram app and Meta Business Suite, so check what your account shows. Write six or seven short answers to your most frequent questions, with the booking link and the opening hours, and use them by hand. This costs nothing and tells you which questions repeat. Many practices stop here and find it enough.
An assistant that drafts, and a person who sends
This is the middle path and the one I would recommend for most professional practices. A person opens the inbox, copies a message into an AI assistant that has your FAQ and tone in its instructions, reads the draft, edits it and sends it. It saves typing, keeps wording consistent and keeps a human responsible for every word. Put the instructions in a shared project in a business-plan assistant, and keep client names out of what you paste.
A connected bot that answers for you
A connected tool replies automatically to some messages, using your FAQ and rules, and hands over the rest. Options include inbox tools that connect to Meta's messaging service and Meta's own business agent. Compare them on handover, logging and price using the Instagram DM automation tools comparison and the ManyChat, Meta agent and booking-app comparison. Prices and availability change often, and Meta's own agent may not be offered on every channel or in every market, so check what is available to your account before you plan around it.
The FAQ file that makes answers safe
A bot is only as safe as the document it draws from. Write a single page, in plain words, with each question and a short answer you would be comfortable seeing on a screenshot. Use the format below, filled in here for an illustrative three-partner accountancy practice.
| Question | Approved answer | If asked more |
|---|---|---|
| Opening hours | Monday to Friday, 9am to 5pm. Closed on public holidays. | Offer a callback request |
| How to book a first meeting | Use the booking page at [link]. The first call is 20 minutes and free. | Hand over if they cannot use the page |
| What to bring | A list of income sources, last year's return if you have it, and your questions. | Do not list tax rules; hand over |
| Price of a personal tax return | "From $[amount], depending on how many income sources you have. We confirm in writing before we start." | Do not quote a price for anything else |
| Do you take new clients? | Yes for personal returns; for businesses please book a call. | Hand over for anything unusual |
Then add a "never answer" list, such as anything starting "should I", "is it legal", "can I claim", "how much will I get back" or "is it safe to". On those, the only reply is: "That depends on your situation, so we would want to talk it through. You can book a call here: [link]." Put a short note in the instructions that if the FAQ does not contain the answer, the assistant must say so and offer a handover, never improvise.
A prompt that bakes this in:
You are the first-line messaging assistant for a small accountancy practice. Answer only from the FAQ below. If the answer is not in the FAQ, say "I do not have that to hand, so I will ask one of the team to reply" and add [HANDOVER]. Never give tax, legal or financial advice. Never state a price except the ones in the FAQ. If the person sounds upset, mentions a mistake, a deadline or a complaint, reply with one calm sentence, then add [HANDOVER] [URGENT]. In your first message in any thread, say you are an automated assistant for the practice and that a person can take over at any time. Keep replies under 60 words, in plain language.
FAQ:
[paste]A message-by-message example
These examples are illustrative, and the practice is a three-partner accountancy firm. A sample thread might look like this.
Message 1: "Hi, do you do tax returns for freelancers?" The assistant replies: "Hello, I am the practice's automated assistant, and a person can take over any time. Yes, we prepare personal returns, including for freelancers. A first call takes 20 minutes and is free: [link]." That is a fixed-fact answer and it is correct.
Message 2, same thread: "How much would I owe this year?" The assistant must not answer. A safe reply: "That depends on your income and costs, so one of the team would want to look at it with you. The booking link is above, or I can ask someone to message you. [HANDOVER]". If it had guessed, it would have made a numerical statement about a person's tax with no information.
Message 3, on another day: "I filed with you last year and the refund was wrong. I am furious." Handover, flagged urgent, with an alert to a named partner within the working day. A reply that is too polished makes a complaint worse; the bot should say one sentence of acknowledgement and no more. A person writes the real reply, ideally by a call.
The prompt's flags (HANDOVER, URGENT) become the working part of the system. Your tool must turn them into a visible notification, such as an email or a task to a named person, and not simply log them. A flag nobody sees is worse than no bot, because the sender was told a person would reply.
Comments on public posts
Comments are a different job. Most are friendly ("great tip") and need nothing or a quick thank-you. Some are questions that people with the same worry will read, and some are complaints. Use AI to read the day's comments and list: which need a reply, which are questions, which are complaints, and which look like spam. Then draft a short reply for the questions and have a person post them. Never let a tool reply automatically to a complaint in public. A public, scripted reply under a post can do more damage than an unanswered comment.
A before-and-after helps. Before: an automated reply to "Your advice cost me money" says "Thank you for your feedback! We love to hear from our followers!" After: a person replies, "I am sorry to hear that. Please message us your contact details so that a partner can look into it today." The second is slower and it is the only one that helps.
Profession-specific cautions
- Financial advisers. A message that could be read as a recommendation is advice. Nothing automated should name a product, a return or a "good time to invest". Keep records of messages, as your regulator may require.
- HR consultancies. Employers write in about disputes and dismissals; those messages are sensitive and often time-critical. Automate only the booking path and fixed facts.
- Accountancy and bookkeeping. Fixed deadlines and fees can be answered; anything about a particular person's affairs cannot.
- Any practice. Never ask for sensitive information in a DM, such as account numbers, identity documents or health details. The bot's instructions should say: "If someone starts to share sensitive details, tell them not to and offer a secure route."
The same principle runs through phone-based receptionist tools, where handover and logging decide whether the service is safe; what an AI receptionist is and how it handles bookings covers the basics, and the realtor call-routing example shows a professional setting.
A seven-day test before you go live
Run the set-up in shadow mode for a week. The assistant drafts replies to real messages, but a person sends them. Keep a simple table: message type, draft quality (good, edited, wrong), time saved, and whether it handed over correctly. At the end, count. If fewer than about nine in ten drafts for fixed-fact and process questions needed no edit, tighten the FAQ. If any draft for an advice or complaint message tried to answer, fix the instructions and repeat the test before it goes anywhere near automation.
Then set a time budget. For a practice that gets 40 messages a week, an illustrative split might be 25 fixed-fact or booking questions, 8 that need a person, 4 spam and 3 complaints or sensitive messages. The tool handles the first group and sorts the third; a person spends perhaps 50 minutes a week on the 11 messages that need them plus a quick spam sweep, instead of about two hours on all 40. Reassess monthly: look at every message the bot answered on its own in a sample of 20, read them as a stranger would, and delete any answer you would not stand behind.
A worked example: an HR consultancy's Friday inbox
An illustration with invented numbers. A four-person HR consultancy gets 60 social messages in a typical week: 34 on Instagram and 26 on Facebook. After sorting a fortnight of messages, the owner finds 28 are fixed-fact or booking questions, 9 ask about fees, 11 are spam or promotions, 8 describe a real workplace problem and 4 are from people looking for work.
The rules she writes: the assistant answers the 28 from the FAQ file; for the 9 fee questions it quotes only the published price for a fixed-price contract review and routes the rest; it files the 11 spam messages in a folder for Friday; it replies to the 8 workplace problems with "That sounds like something to talk through properly. Here is how to book a call", flagged for a person; and it points jobseekers to the careers page. The people-heavy part drops from about 60 messages to 17 that need a person, of which 8 are the ones that matter most.
The sum on time: if each message takes about three minutes to read and answer by hand, 60 messages take 3 hours a week. With triage, the person spends about 17 messages at four minutes each (about 68 minutes) plus 15 minutes on the Friday sweep. That is a saving of roughly 1.5 hours a week, and more importantly the eight real enquiries are never buried between spam. The tool might cost anywhere from nothing, for saved replies, to a monthly fee; the saving is worth comparing against that fee only after the shadow week confirms the 28 answers are right.
After hours and weekends
An inbox bot earns its keep out of office hours. A message arriving at 9pm on a Saturday should get an immediate, honest reply: "Thank you for your message. I am the practice's automated assistant. The office is open from Monday at 9am, and a person will reply by Tuesday at noon. If you would like to book a call, here is the link." That sets an expectation you can meet. Then make sure you meet it. Add a Monday morning routine: open the handover list, sort by urgent first, and reply to each message. The stated deadline is a promise.
Be careful with the messaging window described above. If a person writes on Saturday and nobody replies until Tuesday, the standard window may have closed; a person can use the human-agent route, which Meta's policy describes as covering a separate seven-day period for manual replies. Ask your tool's support how it handles this, since it is a common source of undelivered replies.
Records, privacy and retention
Direct messages become records. Decide before you start how long you keep them, who can read them and where they are exported. If a person tells you something sensitive in a DM, move the conversation to a secure channel and delete or restrict the message according to your own rules. Add one line to your privacy notice saying that messages may be handled by an automated assistant and read by staff. Do not copy message text into a general-purpose AI tool on a free plan; use a business-plan assistant or the tool's own setting, and strip names.
If you operate under professional rules, for example those that govern advisers or health practitioners, check whether message content counts as a client record, and whether your regulator expects you to keep it. This is the point to ask your professional body, not an AI assistant.
Inbox-bot errors practices hit in the first month
- The confident wrong price. A bot answered a fee question from an old FAQ line, after prices rose. Fix: put a review date on the FAQ page and a rule that any change to prices updates the file the same day.
- The silent handover. A bot said "a person will reply soon" and nobody was notified. Fix: test the notification yourself by sending a message that should hand over, and see whether the right person receives it within five minutes.
- Looping. A user who asks the same question twice gets the same answer twice and gives up. Fix: after two repeated messages, hand over.
- Over-friendly tone. An exclamation-heavy tone sits badly with a serious practice. Fix: add "no exclamation marks, no emojis" to the instructions and read ten replies aloud.
- Answering for someone else. A message was from a client's relative asking about the client's file. Fix: the assistant never discusses an identified person's matter, whoever asks.
What stays with a person, always
- Any advice, opinion or recommendation.
- Complaints, errors, deadlines and anything with distress in it.
- Fees that depend on circumstances.
- Messages that mention a client's identity or case.
- Any request for sensitive information.
- Press, regulators and legal letters.
- The weekly review of what the bot did.
If you want an inbox that joins WhatsApp, Instagram and Facebook into one place with a clear handover, one AI inbox for WhatsApp, Instagram and Facebook enquiries goes through the set-up. Platform rules and prices move, so re-check Meta's current messaging policy and your tool's pricing page every quarter.
AI social media receptionist questions
Do I have to tell people they are talking to an AI?
Meta's messaging policy says automated chats must disclose where the law requires it, and recommends disclosure anyway. Tell people at the start, for example in the first reply, and say how to reach a person. It builds trust and avoids awkward moments later.
Can AI reply to public comments as well as DMs?
It can draft them, but public replies are visible to everyone and are harder to retract. Use AI to suggest wording, and have a person post, especially for anything mentioning a service, a price, a complaint or a named client. Hide or delete nothing automatically.
What happens when someone messages after the reply window has closed?
Meta's messaging rules give businesses a window to reply, 24 hours for most actions, and limit what can be sent after it. A person can use the human-agent route for a longer, separate period. Check your tool's current rules, because they change.
Is a chatbot cheaper than a part-time receptionist?
Sometimes, but the comparison is misleading. A triage and FAQ tool carries a monthly fee (check the vendor's pricing page for your message volume), while a person still handles the calls, advice and complaints it cannot. Count the person's time saved on repetitive questions, not the whole role.
Further reads
- How to Turn Instagram Comments Into Leads With Keyword Auto-DMs — Keyword auto-DMs from comments, and their limits.
- How to Automate Instagram DMs Without Sounding Like a Bot — Automating Instagram DMs without sounding like a bot.
- AI Email Triage for Shared Inboxes: Sales, Support, and Invoices — Triage rules for any shared inbox, not just social.
- How to Measure an AI Receptionist's Return in the First 90 Days — How to measure whether a receptionist tool paid off.
- What to Do When an AI Receptionist Gets a Booking Wrong — What to do when an automated booking goes wrong.
- AI Receptionist or Answering Service for a Booking-Based Business? — Compare AI with a human answering service.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Meta's Messenger Platform policy overview (messaging windows, automated-chat disclosure, Human Agent tag, 30-second bot response, Instagram feature limits); Meta's WhatsApp Business Platform pricing page for Meta Business Agent messages; ChatGPT and Claude pricing pages.