OpenAI's June 2026 Workspace Agents update added native connectors that let ChatGPT read from and act inside Slack, Salesforce and Notion. For an Australian small business already paying for those tools, the appeal is obvious. Ask one assistant a question and have it pull the answer from your CRM, your team chat and your documents at once. We build on Claude at Automata AI, but our clients run mixed stacks, so this is a practical walkthrough of how to wire ChatGPT into all three, and the access questions you should settle before you do.
Settle the access question first
A connector is a standing grant of access, not a one-off lookup. Once you link Salesforce, ChatGPT can read whatever the connected account can read, and the same holds for Slack and Notion. That is fine when the scope is tight and a real problem when it is not. Under the Privacy Act 1988 and the Australian Privacy Principles, your business stays accountable for personal information even after it passes to a third-party tool, so a connection you set up in five minutes is a decision you may have to defend later.
Before you click connect, settle four things:
Use a dedicated service account, not a director's personal login, so access can be revoked without locking anyone out of their own work.
Scope each connector to read-only first, then add write access only for the one or two workflows that genuinely need it.
Check where the data is processed and stored. ChatGPT processes prompts on OpenAI infrastructure offshore, which matters for client records covered by the Privacy Act.
Record who approved the connection and what it can reach, because that note is exactly what an auditor or a security-conscious client will ask to see.
Connecting ChatGPT to Slack
Slack is the most straightforward of the three. In the ChatGPT admin console, open Settings, then Connectors, choose Slack and select Connect. You are redirected to Slack's authorisation screen, where you sign in as the service account and approve the workspace. Slack asks which channels the connector may see, so grant the smallest set that covers the work, usually a handful of project or support channels rather than the entire workspace. Once approved, ChatGPT can summarise long threads, find decisions buried in old messages and draft replies. It cannot post on your behalf unless you separately switch on write access, which most teams should leave off until they trust the output.
Connecting ChatGPT to Salesforce
Salesforce is where most of the value sits for a sales-led business, and where the access question bites hardest, because a CRM holds the personal information of every customer and prospect you have. From Connectors, choose Salesforce and Connect, then authenticate through Salesforce's login flow. A Salesforce administrator usually has to approve the connected app on the org side before the link completes. Set the integration user's profile so it sees only the objects ChatGPT needs, for example Accounts, Contacts and Opportunities, and nothing under Setup. Test it with a read-only question such as this quarter's open pipeline before you trust it with anything that writes back to records.
Connecting ChatGPT to Notion
Notion holds your internal knowledge, so the goal here is retrieval: ask a question, get an answer drawn from your own documents. From Connectors, choose Notion and Connect, sign in, then pick the specific pages or teamspaces to share rather than the whole workspace. Notion's sharing is granular, so you can share the staff handbook and process docs while keeping HR, finance and board spaces out of scope. After connecting, confirm the assistant only surfaces what you intended by asking it to list the pages it can see.
Check each connection before you rely on it
Connecting a tool is not the same as trusting it. Run the same short test after each connector goes live:
Ask a read-only question and confirm the answer matches what the source system actually holds.
Ask something it should not be able to answer, like a salary or a private channel, and confirm it returns nothing.
Open the access log in Slack, Salesforce or Notion and confirm the connector shows up as its own actor, not as a staff member.
Revoke the connection and reconnect once, so you know the off switch works before the workflow becomes part of someone's day.
What this costs an Australian small business
The connectors are included with ChatGPT's business plans rather than billed as a separate line. The real cost is the per-seat licensing across your team and the time to set it up properly. A ten-person business on a ChatGPT business plan runs to roughly $4,800 a year in AUD on subscriptions alone, and a careful rollout with scoped service accounts and a tested access model is usually a one to two day piece of work, on the order of $3,500 to $7,000 if you bring in outside help. Against that, an automation that saves each of ten staff thirty minutes a day is worth well over $45,000 a year in recovered time, so a connector pays for itself quickly when the workflow behind it is real and not just a novelty.
A Claude-first note on the same problem
The connector pattern is not unique to ChatGPT. Claude uses an open standard called the Model Context Protocol for the same job, which means a connector built once can work across different tools instead of being tied to a single vendor. For clients who handle regulated or sensitive data, we tend to build on Claude because the governance story is cleaner: tighter control over what a connector can reach, and clearer commitments on how your data is handled. Either way, the discipline in this guide carries over. Scoped service accounts, read-only by default and a tested off switch matter whichever assistant you put in front of your team.
Where to start
Pick the single connector that removes the most manual lookups, usually Salesforce for a sales-led team or Notion for a knowledge-heavy one, and get that working safely before you add the next. If you would like a hand mapping which workflows are worth connecting and which should stay manual, book a brainstorm and we will work through it with you.
