Instagram: A New Way to Sell When Your Shop Lives in the Feed
Your profile already looks like a storefront, but sales get stuck in DMs. Learn how an AI agent turns posts into a catalog and answers customers for you — without a website or development.
An Instagram store has a peculiar disconnect. The storefront is already built: hundreds of photos, descriptions, reviews, live content. Yet, sales still bottleneck in direct messages, where one person manually answers "do you have size 38?" forty times a week.
Anthropic recently analyzed what makes commercial AI agents truly effective in production: The anatomy of effective commerce agents. One of the conclusions explains why previous DM bots failed to take off, while current agents are changing the landscape.
Why scripted bots didn't solve the problem#
An old bot operates on a tree structure: button, branch, another button. A live customer doesn't fit into that tree. They ask about size, then delivery, then return to size and say "and show me in beige, and how much will it be with delivery to Krakow".
The article formulates it this way: a purchase conversation is a single, tightly connected session with multiple intents. Therefore, an effective architecture is one agent that maintains the entire context, not a set of bots passing the baton. This is precisely what distinguishes current solutions from scripted ones: the answer is built from your catalog and your conditions, not chosen from pre-written phrases.
What happens to your posts#
The agent reads the feed just like a human does: it looks at photos, analyzes captions, listens to Reels. This generates product cards: name, price, sizes, composition, link to the original post.
Fifty posts are processed in three to six minutes. You can then retrieve more of your archive using the “Continue older” button, without fear of duplicates: posts are identified by their ID, and old cards are updated.
Reels are a separate story. For deep video analysis, Gemini listens: it transcribes speech and reads on-screen text. For accounts where half the content is video, the difference between quick and deep analysis is twofold, because people verbally share details they never write in captions: what it's made from, how long it takes to dry, who it's not suitable for.
Link in bio instead of a website#
Next, the most interesting part for those without a website, or who don't plan to have one. The agent lives at a short link you place in your bio. The customer clicks and lands not on a "leave a request" form, but in a chat where they can ask questions, choose items, add them to a cart, and pay by card, Apple Pay, or Google Pay.
The customer can add this same page to their phone's home screen, creating an icon with your logo. After this, you can send them push notifications about new collections. Your own app, without developing an app.
The link is also convenient to share in DMs: in the messenger, it expands into a card with a cover image, store name, and description.
Where the return is most noticeable#
Evenings and weekends. Half of messages arrive when no one is available to respond. This is revenue you already have, but no one to serve.
"Price in DM". A classic of Instagram commerce that kills conversion. The agent states the price immediately and guides the customer to an order while they are engaged. If the price isn't in the posts, it will honestly say "I'll clarify" and take contact information — a lead instead of a lost conversation.
Services and bookings. Masters, studios, consultations. Before purchasing, people ask five to ten questions: how long it lasts, what to bring, if there are openings on Saturday. Every unanswered question is a booking lost to whoever responded faster.
Customers in different languages. A store in Warsaw receives messages in Polish, Ukrainian, English. The agent responds in the customer's language, even if the catalog is described in just one.
How difficult is it to implement#
This is perhaps the main change compared to previous generation bots. You don't need to configure scenarios — the agent learns from your link.
- You provide a link to your profile.
- The agent reads posts and videos and builds the catalog itself.
- You add conditions: delivery, payment, returns, bookings.
- You place the link in your bio.
This usually takes fifteen minutes. After that, the agent pulls in new posts as they appear, and you monitor conversations and leads in your dashboard.
There's no neural network training "on your data" in the frightening sense: the model doesn't retrain; your catalog and conditions simply become the basis for its responses. What isn't written in the conditions, the agent won't invent — it will say it needs to clarify.
How to get started#
Check out the first launch — it has seven steps with pictures. If you want to understand in advance how the agent analyzes Instagram and what to do if the catalog turns out imperfect, refer to the article on sources.
And a small practical tip: before connecting, write the price as a number in your three most recent posts. The catalog will be built noticeably better.