OpenAI Put a Business-Apps Shelf Inside ChatGPT. Here Is How to Use It.

ChatGPT's Small Business Collection: what launched on September 9

On September 9, 2026, OpenAI released the Small Business Collection inside ChatGPT: a curated set of sixteen ready-to-integrate business apps. HoneyBook, the client relationship platform for independent businesses, announced that day that it was one of the tools selected for the collection, and that its members can connect their accounts and ask ChatGPT questions about their own clients, projects, schedules, contracts and invoices, then have it prepare the work that follows. Dropbox announced its own integration a day earlier, on September 8.

The collection covers the tools a small business already runs on, according to a September roundup from ai.advalorem.io: client management through HoneyBook, design through Canva and Figma, payments through Stripe and PayPal, payroll and HR through Gusto, team messaging through Slack, file storage through Dropbox, project tracking through ClickUp, websites through Wix, accounting through Intuit QuickBooks, CRM through HubSpot, ecommerce through Shopify, e-signatures through Docusign, business banking through Mercury, and sales intelligence through ZoomInfo.

The practical shift is simple. Until now, chatting with an AI meant explaining your business from scratch every time, or pasting in the data it needed. With your real accounts connected, the assistant starts from your own records. You ask what is happening. It answers from your systems, not from its imagination.

Why a ChatGPT connector shelf matters more than another app

Small businesses do not need another app. They need the apps they already pay for to talk to each other without a new dashboard, a new login habit, or a new monthly training session for staff.

That is what makes this launch different from the usual AI news. The collection is a connector layer, not a product launch. The assistant you already use reads the systems of record you already run. There is no migration, no data import, no week-long setup. You link an account once and start asking questions.

Think of it the way you would think of a good counter assistant. A new app is a new employee who needs training, supervision, and a chair. A connector is your existing assistant suddenly able to read the ledger, the order book, and the stock register without being handed each one. The business does not change. The distance between your question and the answer shrinks to one sentence.

This is also why the honest-advice version of this news is worth saying out loud: if you already use Shopify, QuickBooks, Slack, or Dropbox, this launch is immediately useful to you. If your business runs entirely on registers and paper, it is not your launch yet, and no honest write-up should pretend otherwise. Connect what you actually use. Leave the rest on the shelf.

How the connect-once, ask-anything loop works in ChatGPT

The workflow has four steps, and the partner announcements describe the same pattern in every case.

First, find the app inside ChatGPT's plugin directory, under the Small Business Collection. Second, link your real account once, the way you would sign in to the app itself. Third, ask in plain language. HoneyBook's announcement lists the kinds of questions its members ask: which invoices are overdue, who is on this week's schedule, where a specific client stands, or which new leads should be answered first. Broader questions work too: which months typically bring the strongest leads, what similar past projects required, or how recent bookings compare by value.

Fourth, and this is the step that matters most, the assistant prepares rather than sends. HoneyBook says everything it creates arrives as a draft for the member to review and approve, and that nothing is sent, changed or issued without that approval. Dropbox's partner announcement puts it the same way: ChatGPT prepares the draft or the request, and you approve what goes out. It can draft replies to enquiries in your style, create proposals, invoices or contracts from your saved templates, add contacts, build lead forms, or prepare payment requests. You always hit send.

That draft-and-approve design is the whole safety model. A chatbot that can read your business records and send emails on its own would be a liability. A chatbot that reads your records and lines up drafts for your approval is a junior assistant with a built-in review step. The technology's capability and the approval gate arrive together, and the partner announcements describe both as part of the product.

The practical use case: one Monday-morning ChatGPT briefing for your small business

Here is a deep dive on the single routine that pays for the whole setup: a Monday-morning briefing in one chat window.

Take a home-goods retailer in Bengaluru selling through her own Shopify store and taking payments that land in Stripe, with accounts in QuickBooks and the three-person team coordinating in Slack. Right now her Monday starts with four open tabs and a notebook. With the collection, it starts with one chat.

Step one: connect your core apps

This is a one-time setup, done on a Friday afternoon: link Shopify, Stripe, QuickBooks, and Slack inside ChatGPT. Four sign-ins, five minutes, and the collection remembers the connections.

Step two: run the briefing, every Monday

She opens one chat and pastes in a saved set of five questions:

1. Which invoices are still unpaid from last week, and how overdue is each?
2. What sold over the weekend, and which products are running low?
3. What is my cash position after this week's expected payouts?
4. Which tasks are still open in our team channel, and who owns each?
5. What needs my personal decision before noon today?

The answers come from her own data, read across the four connected apps.

Step three: turn answers into drafts

The assistant drafts a reminder message for each overdue invoice in her own tone, ready to review; a restock note listing exactly the low-stock products; and a short task list she can post to the team channel herself. Drafts, not sends.

Step four: approve and send

The approval gate stays with the owner: read, fix anything the AI misread, send. Nothing about this removes her judgement. It removes the thirty minutes of tab-switching before her judgement is needed.

The habit that makes it work is dull and deliberate. Same questions, same day, every week. The first Monday feels slow because you are checking the answers carefully. By the fourth, you know which answers to trust at a glance and which to verify. That is the same rollout discipline behind the workflow automation solutions we build for clients: one routine, connected to real systems, with a human approval step that never gets skipped.

What ChatGPT's app shelf still cannot do

A launch announcement is a promise of what the product does, not a map of its edges. The edges matter for a business owner, so here they are.

It cannot send anything without you. The draft-and-approve workflow is the product, not a temporary setting. If you are worried about an AI emailing your customers at 3 a.m., the design already answers that worry.

It can only answer from data it can reach. The briefing is only as honest as your records. If invoices live in a notebook the app cannot see, the answer will be confident, tidy, and wrong. The first honest test of the collection is whether your business records live inside the apps you connect.

Not every app on the shelf will apply to you. The collection is international in flavour: Gusto handles payroll and HR in the US sense, and Mercury is a US business-banking product. If you use those tools, good. If you do not, connect the ones you do. The value is in the connection, not the count of connections.

Verify numbers before you act on them. An assistant reading your QuickBooks figures can summarise them beautifully and still round a figure or mix up a date range. Check anything that moves money before it moves. The launch announcements from HoneyBook and Dropbox describe a careful product: ChatGPT prepares the draft or the request, and you approve what goes out. The safe habit is the same one the announcements model — nothing reaches a customer or a money record until you have reviewed it.

Getting started with ChatGPT's business apps this week

Do not connect all sixteen apps. Connect one, ask one real question, and see what happens. The whole pilot fits in an afternoon.

1. Map: write down the three tools your business actually lives in. For most small retailers, service firms, and workshops, the list is some mix of a payments tool, an accounting tool, and a messaging or task tool.
2. Connect: link one of those inside ChatGPT's Small Business Collection. One account, one sign-in, done.
3. Prompt: ask one question you currently answer by opening three tabs. "Which invoices from last month are still unpaid?" is a good first question.
4. Measure: did the answer save you thirty minutes, or did you spend thirty minutes fixing it? Keep the connection only if the answer is the former. Expand to a second app only when the first one earns it.

The pattern underneath this launch will not stay inside ChatGPT. Amazon's September announcements showed AI assistants for sellers, PayU put an agent team inside its payment flow for small merchants, and Google folded more AI into the tools businesses already run. The direction is consistent: the AI comes to the systems you already use, and it stops asking you to type things in twice. A connector shelf inside the chat app you already open is the most literal version of that trend so far. Businesses that learn the connect-once habit now will spend less time feeding software and more time running the shop.