What Are AI Connectors? Letting Your Chatbot Read Your Drive and Email

"Can you pull up last month's meeting notes from Drive and draft me a report?"

The answer is yes, once you connect it. A connector links your AI chatbot to services like Google Drive, Gmail, Calendar, or Notion. Once it's set up, the AI searches for the file itself and reads it, no copying and pasting required. ChatGPT, Claude, and Gemini all have this feature, but each one calls it something different. If you came from the previous track already comfortable working with AI inside a browser tab, this is where you go a level deeper, outside the browser. If you just want to know what a connector is, this lesson alone covers it. This reflects how things stood in August 2026. Menu names and locations for connections change often, so if you're reading this well after that, check each company's current documentation too.

Connector

A pathway that lets an AI chatbot reach into your apps and services to pull data, and sometimes take action on your behalf. It's less like handing over a copy of your keys and more like issuing a badge. Anything you can't see, the AI can't see either, and you can revoke the badge at any time.

How Is This Different from Copy-Paste?

Up to this point in the track, there's been exactly one way to hand AI your material: open a file, copy the text, and paste it into the chat window. That approach has three problems. You have to carry every piece of information over by hand. Long documents have to be cut into chunks. And the AI can only ever see the fragment you decided to give it.

Connect a service, and the whole thing flips. Say "find last month's meeting notes and summarize them," and the AI searches your Drive, finds the document, reads it, and answers. Claude even cites which document it used, with a link back to the source. Instead of you carrying material to the AI, the AI goes and gets it.

Same Feature, Three Different Names (as of August 2026)

The three companies built the same feature and gave it three different names, which makes it confusing to even search for. Here's the breakdown.

Claude's own documentation states plainly that Google Workspace connectors are open to every user. Gemini carries the same restriction you saw in the memory lesson: personal Google accounts only, no work or school accounts. ChatGPT's supported plans and regions vary app by app, so you'll need to check each one individually in the directory.

Read-Only, or Can It Write Too?

How far a connector reaches depends on the tool. Claude's Google Workspace connectors draw a clear line.

• Gmail: Search, read, and draft emails, plus send, reply to, and forward them. By default, though, Claude asks for your approval before it sends anything.
• Calendar: Read events, plus create, edit, and delete them.
• Drive: Search and read documents, and save files Claude creates back to your Drive. Sharing, moving, or trashing a file goes through an approval step first.

There's one principle worth pulling out of this list: write actions default to requiring your approval. Claude states that asking for explicit permission before acting on your behalf is the default across the board. ChatGPT instead lets you pick from four permission levels: always ask, ask on any change, ask only for important actions, or never ask. The default is "read automatically, ask before anything important." Sending an email, deleting a file, or making a purchase all count as important.

I learned this principle the hard way, from the opposite direction. I'd connected Drive, asked Claude to create a document and drop it in a specific folder, and it just didn't happen. Claude could only reference my Drive, not write to it. I'd assumed that once something was connected, it could do anything.

Being connected and being able to do anything are two different things. Each connector opens up a different range of actions, and even within that range, writing requires approval. What was and wasn't allowed had been spelled out on the screen I saw when I connected it. I just hadn't read it.

Read the Permission Screen Before You Connect

Hit connect, and you'll land on a login screen followed by a permission consent screen. Most people just click "Allow" here, but that screen is really a contract. On the Google side, permissions break down into three tiers.

1. Basic profile: name, email, photo. That's it.
2. View access: read and copy data like contacts or photos.
3. Manage access: edit, upload, create, and delete data.

Two things to check on that consent screen: what tier of access it's asking for, and whether that tier actually matches what you're trying to do. If a tool that only needs to read your documents is asking for manage access, that's a signal worth pausing on.

It helps to know how to disconnect, too, before you ever need to. You can do it inside each service (Claude: Customize > Connectors; ChatGPT: Settings > Apps), or handle it all from your Google account at once. Go to myaccount.google.com/connections and you'll see every service currently accessing your Google account, with the option to remove access any time.

What Should You Connect, and What Shouldn't You?

Being able to connect something and having a good reason to are two different questions. Here's how to draw the line.

Fine to connect: material in your personal account that's no problem for AI to read. Your own blog drafts, a personal calendar, a folder of public materials. This extends the same principle you set up in the memory lesson.

Think twice: company material. On a work account, an admin usually controls connector policy already (both Claude and ChatGPT gate organization plans behind admin approval). Moving confidential company files into a personal account just to connect them is close to a security incident. The same goes for any folder holding customer personal data, contracts, or financial records.

One thing worth remembering: data pulled through a connector passes through that service's own servers. Claude's documentation says the transfer is encrypted, but also notes that third-party services process data on their own infrastructure. On ChatGPT's personal plans, if "Improve the model for everyone" is turned on, information pulled from a connected app can be used for training. If you're dealing with anything sensitive, check that setting first.

Behind This Connection Sits MCP

Scroll through a connector list long enough and you'll run into the term MCP. It shows up in Claude's custom connector menu and in ChatGPT's instructions for building custom apps. For now, one line is enough: MCP is the shared standard that makes these connections work the same way everywhere, and it's the door that lets you connect services that aren't in any directory. We'll get into how it actually works in the MCP lesson. The next lesson has you connecting something in Claude yourself.