AI replies: answer DMs from your FAQs
Updated
Keywords are great for “comment LINK, get the link”. But fans also send things you never planned for: “do you ship to Dubai?”, “is this vegan?”, “what time do you open?”. AI replies answer those for you, in the same thread, using only the FAQs you wrote. If the answer isn’t in your FAQs, it doesn’t guess — it sends your fallback message instead.
It is deliberately narrow. It is not a chatbot that chats about anything, and it never invents a price, a policy or a date. Everything it says traces back to something you typed.
What it does — and what it never does
| It does | It never does |
|---|---|
| Answers a DM, a story reply or a comment that no keyword matched, from your FAQs and facts. | Answer from outside your FAQs. No open-ended chat, no made-up answers. |
| Replies in the fan’s own language and script (Hinglish included) when Answer in their language is on. | Replace a keyword rule. A keyword that matches always wins; AI only steps in when nothing matched. |
| Sends your fallback message when the FAQs can’t answer, so the person is never ignored. | Answer someone who asks for a human. “Talk to a person” goes straight to your fallback, no AI call. |
| Shows the fan a typing indicator while it works on a DM. | Answer a photo, voice note or sticker. Only messages with words are answered. |
| Records every attempt — answered, fell back, out of credits, failed — on the Usage tab. | Fake a reply. No setup, no credits or no answer means the fallback goes out, and the reason is logged. |
Set it up
- Open the AI Agent pageClick AI Agent in the sidebar. The page has two tabs, Build and Usage. Everything below happens on Build.
- Add your FAQsStep 1, Add your FAQs: click Add a FAQ, write the question a fan would ask and the answer you’d give, then save. Use Fact instead of FAQ for something the agent may state with no particular question — your opening hours, say. You only need one phrasing per question; it recognises other wordings. Writing FAQs it can use has examples.
- Define the voiceStep 2, Define the voice: who the agent is, how long replies run, whether it uses emojis. Optional, but it’s what makes replies sound like you. See AI voice and tone.
- Check the Advanced stepStep 3, Advanced: the language, the When the FAQs can’t answer message, the Automated reply label and the greeting reply. Details below.
- Turn on AI replies and saveFlip the AI replies switch in the status bar at the top, then press Save changes if it’s lit. The Test your agent chat on the right answers from what is *saved*, so save first.
- Switch on AI reply in an automationThe account-level switch isn’t enough on its own. Open the automation that should hand unmatched messages to the agent, go to its Advanced settings and turn on AI reply (FAQ). See the next section.
Choosing which automations use it
The agent speaks inside an automation, because every reply needs a home in your queue and your per-person limits. In the automation’s Advanced settings you’ll find AI reply (FAQ) with the note “Answers when no keyword matches”. It only shows in Advanced mode (the Simple/Advanced switch is described in Build your first automation).
Once it’s on, a second control, When AI answers, has two options: On no match (AI steps in only when no keyword matched) and Every message (AI is consulted on every message, but keywords still win first). Either way, a keyword that matches is handled by your normal automation, not by AI.
- Unmatched DMs: switch it on in a Someone DMs a keyword automation. See DM keyword triggers.
- Unmatched story replies: switch it on in a Reply to my story automation. See Story replies and mentions.
- Comments on a post: if the automation on that post has AI reply on, a comment that matches none of its keywords can be answered too. That answer goes out as the one private reply Instagram allows per comment, so use this deliberately.
The status bar: what “Live” really means
The coloured badge at the top of the AI Agent page always names the first thing that’s stopping the agent, so you never have to guess why it’s quiet. Reading top to bottom:
| Badge | What it means | What to do |
|---|---|---|
| Setup pending | AI isn’t available on our side right now. Nothing is faked; you can keep building and it answers once it is. | Keep building. If it lasts, contact us. See no_key. |
| Not on this plan | Your plan has no AI credits to spend. | Compare plans in Plans and features. |
| Off | The AI replies switch is off. | Turn it on. |
| Out of AI credits | This month’s credits are used up. Greetings still go out; everything else gets your fallback message until the credits reset. | See AI credits. |
| Needs FAQs | No FAQ is switched on. | Add at least one. |
| Needs an automation | No active automation has AI reply on. | Turn on AI reply (FAQ) in one. |
| Live | Answering unmatched messages from your FAQs. | Nothing. Test it. |
If you use your own AI key, the “Out of AI credits” gate doesn’t apply, because the replies aren’t spending Dartzo credits. See Use your own AI key.
The Advanced step, control by control
- Answer in their language — on by default. Replies in the language *and script* the fan wrote in (romanised Hindi comes back romanised). Turn it off to pick one language from a list: English, Hindi, Hinglish, Tamil, Telugu, Marathi, Bengali, Gujarati or Kannada.
- When the FAQs can’t answer — the message sent instead of a guess. Leave it empty and Dartzo sends: “Thanks for reaching out! I’ll have someone follow up with you shortly.” Write your own so it sounds like you and promises only what you’ll really do.
- Flag me — on by default. The conversation is marked so you can pick it up, and the fan still gets the message above. It shows as Handed to you in the Usage tab.
- Mark replies as automated — adds a blank line and “Automated reply” under every AI-written message. Off by default. Turn it on if you want fans to know a machine wrote it. See AI voice and tone.
- Reply to greetings — a plain “hi”, “hello!!”, “hey there” or “good morning” gets a fixed reply straight away, with no AI call and no credit spent. Leave the greeting empty for “Hey! Thanks for reaching out. What can I help you with?” or write your own.
The typing indicator
About a second after a fan’s DM lands, the agent shows them Instagram’s “…” typing indicator, then sends the reply. It isn’t a message, so it costs nothing. It isn’t shown for comment replies, because a comment’s private reply has no typing state. If Instagram refuses the indicator, the reply goes out anyway.
Test before you trust it
Test your agent is a live chat on the AI Agent page. Ask the questions fans really ask and look at what comes back. Each test answer is a real call and it does use credits, so keep tests short. If it answers wrongly, the fix is nearly always a clearer FAQ, not a different setting. If the FAQs and facts together get too long, a “Prompt size” meter in Step 1 shows how full they are (the limit is 6,000 tokens; a token is a small chunk of a word). Past it, the oldest entries stop reaching the model.