Getting started with AI — Layer 4: connecting AI to your own documents and tools

This is the fourth in the series. In Layer 1 we had a conversation. In Layer 2 we got good at feeding it context. In Layer 3 we handed over the keyboard and let it do small tasks in a browser. At the end of each one I promised the same thing: connecting AI to your own documents and tools. This is that post.

Go back to Layer 2 for a moment. Its big lesson was that the AI only knows what you tell it, so the quality of the answer depends on the context you paste in. That's true, and it's also tiring. Every conversation starts from a blank page, and you end up re-explaining your prices, your terms and your customers every time.

The one idea in this layer:

The AI can now read your world for itself: your files, your email, your calendar, your price list. And most small businesses are already paying for a version that does this, without having switched it on.

That second half is the reason for this post. When I sit down with a small business, the most common thing I find isn't a missing tool. It's a tool they're already paying for, half set up, that nobody has pointed at their own documents.

What "connected" actually means

There are two quite different ways of giving AI your documents, and the difference matters.

  • Uploading. You drag a file into a chat. The AI reads that copy, for that conversation. It doesn't see the next version, and it doesn't see the file next to it. Useful, and it's where Layer 2 left us.
  • Connecting. You let the AI look inside the place your files already live (your SharePoint or OneDrive, your Google Drive, your mailbox) with your permission. It searches when it needs to, finds the current version, and tells you which document the answer came from.

Connecting is the step change. It's the difference between briefing a temp every morning and working with a colleague who has access to the shared drive.

Where you probably already have it

As I write, in early November 2026, here's what the two big office suites include. Plans and names change often, so treat this as a map, not a contract.

If you're on Microsoft 365. Eligible Microsoft 365 business plans now include Copilot Chat at no extra cost, and you can give it files to work with. The step up is a paid Microsoft 365 Copilot licence, sold per person on top of your existing plan. That's the version grounded in your own work: your email, files, meetings and chats. It also sits inside Word, Excel, PowerPoint and Teams, and comes with Microsoft's ready-made agents for research and analysis.

If you're on Google Workspace. Gemini is now built into the business plans rather than sold as an extra: the Gemini app, Gemini in Gmail and Docs, and Gemini Notebook, which is the closest thing to a ready-made "office manual you can ask questions of". The part that matters for this layer, asking Gemini questions across the files in your Drive, needs Business Standard or above.

So if you're on Business Standard or already have Copilot licences, you may own most of this layer. It just hasn't been set up.

And if you prefer Claude or ChatGPT

You don't have to use the AI that came with your office suite. Both of the big independent assistants can now reach into the same files:

  • Claude has a Microsoft 365 connector that searches SharePoint, OneDrive, Outlook and Teams. Anthropic lists it on every Claude plan, but it needs a Microsoft business account; a personal Outlook or Hotmail address won't connect. On a team plan an administrator switches it on once, then each person connects their own account.
  • ChatGPT's business plan has what OpenAI calls company knowledge. It answers from connected sources such as SharePoint, Google Drive and Slack, respects each person's existing permissions, and cites the documents it used.

Which one is right depends on how you work, not on which is "best". If your team lives in Outlook and Teams all day, Copilot inside those apps is hard to beat for convenience. If one or two people do most of the thinking and writing, a Claude or ChatGPT seat connected to the shared drive can be the better buy. I'm not tied to any of them, and quite often the honest answer is "the one you already have".

Beyond Microsoft and Google: the rest of your tools

Your documents aren't only in an office suite. The price list might be in Notion, the customer history in HubSpot, the invoices in Xero or QuickBooks, the files in Dropbox or Box. This is where Claude and ChatGPT pull ahead of the suite's own AI: both now have a directory of ready-made connections to the other tools a small business runs on.

  • Claude's connector directory includes Notion, Slack, HubSpot, QuickBooks, Box, Asana, monday.com, Shopify and Canva, alongside Google Drive and Gmail. Xero has its own Claude connector too, which Xero describes as view-only: Claude can answer questions about your figures but can't change anything in your books.
  • ChatGPT calls these apps now (they used to be "connectors"). On the business plan that includes Dropbox, Box, Notion, Slack, Outlook and Intercom, and the list keeps growing.

Many of these are built by the software company itself rather than by the AI vendor, and most use an open standard called MCP, so one connection pattern works across tools. That's also why the lists change month to month: check the directory for the tools you actually use, rather than trusting any list in a blog post, including this one.

Two practical notes. First, the same "reading before writing" advice applies: some of these apps can now create and edit as well as read, and your administrator can usually choose which actions are allowed. Second, if a tool you rely on isn't listed, that isn't the end of it. Bridging services such as Zapier can link it up, and for something central to your business a small custom connection is usually a modest piece of work rather than a big project.

What to set up first

Don't start by connecting everything. Start with one shared space loaded with the documents people ask about most. Depending on the tool, that's a Copilot Notebook, a Gemini Notebook, or a Claude or ChatGPT project. For most small businesses, the first one looks like this:

  • your current price list and what's included at each price;
  • your standard terms, or the paragraph you always paste into quotes;
  • your FAQs: the ten questions customers ask every week;
  • two or three good past proposals or quotes;
  • a short "how we do things here" note, even if you have to write it first.

Then try it on real questions:

  • "A customer wants X by Friday. What do we normally charge, and what's in our terms about rush jobs?"
  • "Draft a quote for this enquiry in the style of our last three."
  • "What did we agree with this supplier last spring?" (with email connected)

You'll know it's working when the answers start from your world, not a blank chat, and when it can tell you which document each answer came from. You'll know it isn't when it confidently answers from the open internet instead. That usually means the document isn't where the tool is looking, which is a set-up problem, not an AI problem.

The one catch: it sees what you can see

Here's the part almost nobody mentions until it bites. These tools respect your existing permissions. That's the right design, and it's also the catch.

If the old "HR" folder was shared with everyone in the company five years ago, then everyone's AI can now find it, summarise it and quote it back, instantly. Nothing was hacked. The permission was always there; the AI just makes it effortless to use. Before you switch on organisation-wide search, spend an hour on the sharing settings of anything sensitive: pay, HR, client contracts, anything personal.

The rest of the sensible-caution list:

  • Check the citations. A connected answer should point you to its source. If it can't, treat it as a guess.
  • Business accounts, not personal ones. Read your plan's data terms before you connect anything, in particular whether your documents can be used to train the vendor's models. Business plans give your administrator proper controls; on personal accounts the settings vary, so keep company documents off them unless you've checked.
  • Keep it to reading at first. Some connections can do more than read: draft email, edit documents. Get used to the reading before you allow the writing.
  • Personal data is still personal data. Connecting a mailbox full of customer details doesn't change your obligations under UK GDPR. It just makes them easier to forget.

That's Layer 4

One idea again: your AI can now read your world for itself. The assistant from Layer 1, which learned to take context in Layer 2 and got hands in Layer 3, can now start from your own files, email and calendar. And the office software you already pay for probably includes most of it.

The hard part isn't buying anything. It's the fiddly hour of switching the right features on, deciding what goes in the first shared space, and fixing the sharing settings before you start, which is exactly the hour most busy owners never find.

I'm Simon. I run Simon Studios, an AI-first development studio in Teddington. I help small businesses and individuals get real work out of the AI and software they already pay for, and build something custom only where there's a genuine gap. No jargon, no hype. If this post describes tools you're paying for but not really using, a Setup Day is a day of exactly this, done properly around how your business works. Or book a free intro call at cal.com/simonstudios/free-intro-call and I'll tell you honestly what would change.