
Yes, AI can answer most Instagram DMs a support team gets: sizing, stock, shipping and return policy, and "where is my order" once it can look the order up. The rule that shapes everything is Meta's: an automated reply has to go out within 24 hours of the customer's last message, and only a person can reply after that, for up to 7 days.
So set it up in three lanes. AI answers the policy and product questions on its own, looks up orders and returns before replying, and hands refunds, complaints, creator DMs and anything past the 24-hour mark to a person. Ignore story reactions and GIFs; most tools can't read them anyway.
For tools, Gorgias, Zendesk and Freshdesk Omni put the AI inside your helpdesk next to email, Manychat is built for comment-to-DM marketing flows, and Meta now has its own free-to-start Business Agent. eesel works as an AI teammate on the DMs in your helpdesk. Question Beauty runs it on Gorgias, where short social messages like "price?" get a direct answer and a product link, and 280 tickets a month close without a person.
What do Instagram DMs to a support team actually look like?
I've spent a lot of time on the plumbing behind helpdesk integrations at eesel, and the first thing I'd tell anyone moving Instagram into their support queue is that an Instagram inbox doesn't look like an email inbox. The messages are short, they arrive at all hours, and a good share of them aren't support at all.
Question Beauty, a direct-to-consumer cosmetics brand on Gorgias, describes its social and chat volume as "short messages from social media and chat: how much is this, do you ship to my country, where is my order, is this in stock." A customer typing "price?" under a product video "wants a link, not a form to fill in."
Rúngne, a Norwegian climbing apparel brand, sees the other half of the picture. Its Gorgias queue gets "a lot of Gorgias tickets that are not support at all: ad comments, story mentions, out-of-office replies, SEO pitches." Out of 2,100 tickets a month across email, chat and social, the useful work started with sorting the noise out.
The other problem is ownership. Before Instagram is connected to a shared inbox, DMs usually live on whoever runs the brand account. One Zendesk customer put it plainly in a G2 review:
"A lot of our customers reach out through Instagram DMs first, and before this whoever managed our social account was fielding those separately from actual support tickets, so if that person was out, messages just sat unanswered."
That is the order I'd fix things in: get DMs into a shared queue, sort the noise, then let AI take the repeat questions. If you skip straight to the AI, it ends up answering emoji reactions and creator pitches.
Can AI legally answer Instagram DMs? The 24-hour rule
Yes, through Meta's official messaging API, which is what every helpdesk and tool in this post uses. The constraint is time. Per Meta's Instagram messaging docs, a business can only message someone who messaged first, and the standard window "lasts 24 hours" for most actions. When the chat starts from a Click-to-Direct ad, the window "may last up to 7 days." Meta added that ad rule on September 28, 2026.
After 24 hours, there is exactly one way to keep talking. Meta's Page messages reference lists five message tags and then says "Only HUMAN_AGENT is available for Instagram Messaging." That tag lets a person reply up to 7 days after the customer's last message. It is for a human typing, not for your AI. Manychat's help center says the same thing in plain words: "Automated messages are not allowed during the 7-day window" (Manychat).

This changes how you plan coverage. On email, a ticket that sits over a weekend is slow. On Instagram, a DM that sits for 25 hours can no longer get an automated answer at all, and after 7 days nobody on your team can reply until the customer writes again. The AI's real job on Instagram is to make sure nothing crosses the 24-hour line unanswered, and anything it can't answer needs to reach a person well before day 7.

The same Meta policy page has two more rules worth knowing before you switch anything on. Automated chats "must disclose that a person is interacting with an automated service" when the law requires it, and Meta names California and German users specifically. And an automated experience needs an escalation path to a human. Gorgias builds the first one in as an optional AI disclosure footer on the first message (Gorgias). I'd turn it on everywhere, not just where it is required.
What does Instagram actually send to your AI?
Less than you'd think, and this is the part most setup guides skip. Your AI only sees what Meta's webhooks deliver, and the Instagram messaging webhooks page lists the gaps:
- "If a person sends a message with a gif or sticker a webhook will not be triggered." Story replies don't support GIFs or stickers either.
- Disappearing photos ("view once, allow replay") arrive as
ephemeralwith no URL. - A shared post arrives as a link, not as the post.

Story mentions have their own rules. Per Meta's story mention docs, a mention only comes through if the account is public, or private and following you. The media link stops working when the story expires after 24 hours, and you're told not to store the media. Comments are separate again: Meta's private replies feature lets you send "one message" to a commenter, "within 7 days of the comment," and follow-ups only if they answer.
Two smaller side effects catch teams out. Replying from a third-party app moves the conversation to the General folder in the Instagram app, whatever your settings are, and a message only shows as Read after a reply is sent (Meta). Message requests that sit for 30 days drop out of the API entirely. If your social manager still lives in the Instagram app, tell them before go-live, or they'll think DMs are disappearing.
What this means for the AI: a story reply is often just an emoji or a sticker the AI never sees, so don't build a flow that depends on reacting to it. Gorgias made the same call. Story reactions, replies and mentions create a ticket, but its AI Agent won't reply because "they aren't real questions" (Gorgias). Rúngne does the same on eesel: "Social comments, story mentions, out-of-office replies and spam are tagged and closed."
Which Instagram DMs should AI answer on its own?
The ones where the answer already exists somewhere the AI can read: your help center, your product catalog or your order system. I'd split every Instagram DM into three lanes.

AI answers. Sizing, fit, materials, shipping countries, return policy, "is this in stock?" and "how much is this?" These are the bulk of Instagram DMs for most brands, and they're also the ones with a clear right answer. Question Beauty trains its agent on "about 270 product pages from Shopify, 230 pages from questionbeauty.com and 120 past tickets and macros from Gorgias," and short messages like "pp", "price" or "link" "get a direct answer and a product link, without a clarifying question." It replies in the language the customer wrote in, which matters on Instagram, where one post can pull DMs from a dozen countries. See my multilingual support guide and pre-sales questions guide for those two angles.
AI looks it up. "Where is my order", return status and address changes before shipping. The AI needs your order system to answer these honestly, and it needs to confirm who it's talking to, because an Instagram handle isn't an email address. Gorgias handles this by asking for the order email or phone and sending a 6-digit code that expires in 10 minutes; a wrong or expired code hands the chat to a person (Gorgias). My guides on order tracking support, WISMO automation and returns management go deeper on each.
Person only. Refunds and exceptions, angry customers, anything that could turn into a public post, creator and press DMs, and anything that has crossed the 24-hour line. Question Beauty sends "refunds, returns, complaints, chargebacks, wholesale and creator or PR requests" to a person, with a short reply and an internal note. Rúngne adds one rule I'd copy word for word: its agent "never states a restock date, a return fee figure or an unconfirmed delivery date." Also keep account changes away from the AI. Anyone can type a name and an order number into a DM, which is why account recovery and fraud tickets stay with people.
One more thing about tone. In eesel's own trials with ecommerce teams, support agents rewrote 8 to 15 sentence AI drafts into 1 to 3 sentence replies. On Instagram that gap is even wider. Poseidon Animal Health, an equine supplement brand on Gorgias, sets this as a rule: "Social media replies are kept short and suited to a comment thread. Email replies can be fuller" (Poseidon Animal Health). Put a length limit in your DM instructions on day one.
Which tools can answer Instagram DMs with AI?
Five are worth comparing for a support team, and they fall into three groups: helpdesks with an AI agent (Gorgias, Zendesk, Freshdesk Omni, Help Scout), a marketing automation tool (Manychat), and Meta itself. Here is what each one does on Instagram specifically, from each vendor's own docs and pricing page as of November 2026.
| Tool | Instagram DMs | Story replies and mentions | Comments | AI auto-replies to DMs | Order lookup in the DM | AI billing unit | Starting price |
|---|---|---|---|---|---|---|---|
| Gorgias | Yes | Ticket created, AI doesn't reply | Yes (people), AI doesn't reply | Yes, AI Agent, within 24 hours | Yes, after a 6-digit code | $1.50 per automated interaction | $10/mo Starter (50 tickets) |
| Zendesk | Yes | Story replies become tickets | No tickets for comments | Yes, AI agents on social messaging | Not documented for Instagram | $1.50 committed / $2.00 pay-as-you-go per automated resolution | $55/agent/mo Suite Team, annual |
| Freshdesk Omni | Yes | Yes, each toggleable | Yes, AI doesn't reply | Yes, Freddy AI Agent | Shopify actions exist, not documented for Instagram | $49 per 100 sessions | $29/agent/mo Growth, annual |
| Help Scout | Yes | Not documented | Not documented | No, AI covers chat widget and email | No | $0.75 per resolution (not on Instagram) | $25/user/mo Standard |
| Manychat | Yes | Yes, as triggers | Yes, AI on posts and reels | Yes, Manychat AI (Instagram only) | No built-in lookup documented | Plan by active contacts | $29/mo Pro, annual (2,500 contacts) |
| Meta Business Agent | Yes | Not documented | Not documented | Yes | Agent platform connects to Shopify | Meta One plans | Free to start |
Gorgias
Gorgias made its AI Agent generally available on Instagram DMs, Messenger and WhatsApp in its AI Agent on Socials update. It's the most Instagram-specific of the helpdesk options: order support inside the thread after the code check, add-to-cart from the DM, and image reading, including image-only messages. It can't read videos, voice notes or documents, can't send attachments, and Instagram and Messenger switch on together because they share one integration.

One detail in its FAQ is easy to miss: the AI stops when a teammate replies from Gorgias, but "replying from the Instagram app does not stop AI Agent." If your social manager answers from their phone, the AI keeps talking. On cost, each AI-resolved DM carries the ticket fee plus $1.50, which a G2 reviewer flagged for busy seasons: "During peak seasons like the holidays when support volume spikes, the costs can increase significantly in a short period" (Mitchell O., G2). An Instagram post that goes viral is exactly that kind of spike. Full numbers are in my Gorgias AI Agent pricing breakdown and the best AI for Gorgias roundup.
Zendesk
Zendesk's Instagram Direct channel turns DMs and story replies into tickets, but "Tickets are not created for comments or replies on posts, reels." Setup starts in the Instagram mobile app, and replies must go out from Zendesk. Its AI agents can be added to "WhatsApp or your other social messaging channels" if you're on Zendesk messaging. I couldn't find an Instagram-specific order lookup in Zendesk's docs. More in my Zendesk AI agents guide, Zendesk pricing and best AI for Zendesk.
Freshdesk Omni
Freshdesk Omni has the widest Instagram coverage of the helpdesks: DMs, story replies and reactions, story mentions and comments, up to 50 accounts (Freshdesk). Freddy AI Agent runs on messages, "not Instagram Comments." The billing unit is the thing to understand: a chat session is every interaction within 24 hours, charged whether or not the AI resolves it. My Freddy session guide and Freshdesk pricing post explain the maths.
Help Scout
Help Scout connects Instagram on every paid plan, and on its free plan up to 100 contacts a month. Its AI agent, Scout, is "designed to resolve customer conversations across Beacon and email" (Help Scout), so Instagram DMs are answered by people. It's a fair pick for a team that wants Instagram in a calm shared inbox and is happy to type replies. See Help Scout pricing and the best AI for Help Scout.
Manychat
Manychat is the tool ChatGPT reaches for first when you ask about Instagram automation, and for good reason: comment-to-DM, keyword triggers, story reply and mention triggers, and Manychat AI, which answers DMs from a knowledge library and is "currently available only for the Instagram channel" (Manychat).

It's a marketing tool first, though. There's no documented order lookup, the shared inbox starts at 2 seats on Pro, and billing counts every person you talk to. One G2 reviewer's warning is worth reading before you sign up:
"I assumed an "active contact" meant someone I reached through an automation I built. In practice, a contact counts as "active" any time a message is sent or received through the platform."
Plenty of brands run Manychat for campaigns and a helpdesk for support, side by side. My Manychat review, Manychat pricing and Manychat alternatives cover it in full.
Meta Business Agent
Meta's own option is new. On June 3, 2026, Meta said it was "expanding Business Agent to Instagram," that "getting started is free," and that paid capacity would come by subscription. That arrived in September as Meta One business plans at $14.99, $49.99, $149 and $499 a month. It hands off to Zendesk, Gorgias and others with a reason attached (purchase_intent, payment_received, human_requested or other).
The catch is coexistence. Meta's conversation routing docs say that when Meta's AI owns a conversation, your app should "not respond to standby messages," and the setting that lets your helpdesk take over "is off by default." Turning it on also pauses your Business Suite away messages and instant replies. Pick one AI per Instagram account; two bots on one thread is how customers get two different answers.
How do you set up AI for Instagram DMs?
About a week for most teams, in this order. The first three steps are plumbing; the last three decide whether customers like the result.
- Check the account. You need an Instagram professional account (business or creator), with Allow access to messages turned on under Settings, Messages and story replies, Message controls, Connected tools (Meta). Most helpdesks also want it linked to a Facebook Page in the same business portfolio. Gorgias says Instagram "can only be added to Gorgias through a Facebook Page," and history isn't imported, so you start from today.
- Connect Instagram to the helpdesk your team already works in. One queue for email, chat and DMs means the order history sits next to the DM and anyone can pick it up. That was the whole point of the G2 review above.
- Connect your store. Shopify (or your order system) is what turns "where is my order" from a canned answer into a real one. Decide how the AI confirms who it's talking to before it shares order details.
- Write the lane rules. Which topics the AI answers, which it looks up, which go to a person, and a hard rule that nothing waits past about 20 hours without either an answer or a person on it. Add the "never promise" list: delivery dates, restock dates, refund amounts.
- Decide what to do with noise. Story reactions, emoji-only replies, collab pitches, spam. Tag and close them, or route them to whoever runs social, so they don't eat your agents' time.
- Test on your own past DMs before going live. Not on 10 questions you made up. A CX lead at a supplements brand doing about 7,000 tickets a month told eesel what every team is really afraid of:
"I need an AI who is only handling the tickets that it's confident to handle and all the other ones, leave them alone."
That's the right bar. At eesel, I run every rollout as a simulation against historical tickets first, so you can see which topics the AI gets right before a single customer sees a reply. My simulation testing guide explains how many tickets per topic you need for the result to mean anything.
If you're nervous, start draft-only. Poseidon Animal Health runs eesel that way on Gorgias: 1,000 draft replies a month, and "a person reads and sends each one." Moving a topic from "draft" to "send" later is a settings change, not a new project. For the wider rollout plan, see my ecommerce support automation guide and handoff best practices.
What does AI for Instagram DMs cost per month?
Each tool bills a different unit, so the only fair way to compare is one example. Pick how many Instagram DM conversations the AI works on each month. I've assumed the AI resolves 60% of them without a person, which matters for the tools that bill per resolution. Helpdesk seat fees, plan allowances and the one-time free credits are left out, so this is the AI line only.
| Tool | AI cost | How it's worked out |
|---|---|---|
| Gorgias | $270 | 180 resolved x $1.50, plus ticket fees |
| Zendesk | $270 to $360 | 180 resolved x $1.50 committed or $2.00 pay-as-you-go |
| Freshdesk Omni | $147 | 3 blocks of 100 sessions x $49, resolved or not |
| eesel | $299 | 500-credit plan, 1 credit per conversation |
| Manychat | $29 | Pro, annual, up to 2,500 contacts (no order lookup) |
| Tool | AI cost | How it's worked out |
|---|---|---|
| Gorgias | $900 | 600 resolved x $1.50, plus ticket fees |
| Zendesk | $900 to $1,200 | 600 resolved x $1.50 committed or $2.00 pay-as-you-go |
| Freshdesk Omni | $490 | 10 blocks of 100 sessions x $49, resolved or not |
| eesel | $499 | 1,000-credit plan, 1 credit per conversation |
| Manychat | $29 | Pro, annual, up to 2,500 contacts (no order lookup) |
| Tool | AI cost | How it's worked out |
|---|---|---|
| Gorgias | $2,700 | 1,800 resolved x $1.50, plus ticket fees |
| Zendesk | $2,700 to $3,600 | 1,800 resolved x $1.50 committed or $2.00 pay-as-you-go |
| Freshdesk Omni | $1,470 | 30 blocks of 100 sessions x $49, resolved or not |
| eesel | $1,149 | 3,000-credit plan, 1 credit per conversation |
| Manychat | $69 | Business, annual, up to 7,500 contacts (no order lookup) |
Two things stand out. Manychat is by far the cheapest, because it's doing a different job: answering from a knowledge library without your order system or a ticket history. And the per-resolution tools get expensive fastest as volume grows, which is the worry in the G2 review above. Prices come from Gorgias, Zendesk, Freshdesk Omni, Manychat and eesel as of November 2026. If DM volume swings with launches and drops, my guide to seasonal support spikes is worth a read.
If you'd rather nobody on your team touches the inbox at all, eesel's Managed Support covers "Instagram and Facebook DMs" along with email, SMS, WhatsApp, reviews and phone for $5k a month flat, inside your own tools. That's the comparison to make against outsourcing ecommerce support, not against a helpdesk add-on.
Try eesel for Instagram DMs
If your Instagram DMs already land in Gorgias or Zendesk, eesel joins that queue like a new teammate who already knows your help center and your Shopify store. It answers sizing, stock and policy questions, looks up orders and tracking before replying, keeps replies short, and hands refunds, complaints and creator DMs to your team with a note. Question Beauty runs exactly this on Gorgias: 960 replies sent to customers and 290 Shopify orders looked up in a month.
You can run a simulation on your own past DMs before anything goes live, and every action can be set to run on its own, wait for approval, or stay off. Plans start at $299 a month for 500 credits, one credit per conversation however many messages it takes. Try eesel free with 100 credits, or read how Rúngne split its social noise from real support first.