Trigger: they're copy-pasting things into the chat that already exist somewhere on their computer or in their accounts. Or they're describing something to the AI that could just be shown.
Establish context
Ask: when you need the AI to look at something — an email, a document, a spreadsheet — how do you get it in? If the answer is copy-paste, or "I just describe it," that's the signal.
What changes
Most AI tools can connect directly to where information already lives. Gmail, Google Drive, Outlook, files on the computer, uploaded documents. The AI reads the source rather than relying on what the person remembers to paste in.
This is the step from chat toy to working tool, and it's the one most people never take — because it feels like a big technical thing when it's usually one click.
What to do right now:
- Check what's already connected. Ask the AI: can you see my emails? Can you read files on my computer? What are you connected to? The answer shows the current state.
- Connect the main source. Whatever they spend the most time copying from — usually email or a document store — connect that first. Most providers have this as a settings toggle or a plugin.
- Try one real task with the connection. Not a test. A real thing they were going to do anyway. "Look at the last five emails from [this person] and tell me what's outstanding." The difference between describing the situation and showing the situation is the difference between a vague answer and a useful one.
Why this matters more than it looks: the quality of what AI gives back is determined by the quality of what goes in. A person describing their situation from memory is giving a lossy, biased, incomplete version. Their actual emails, documents and data are the uncompressed version. The same model, the same question, dramatically different output — just because the input was real instead of remembered.
What to watch for:
- Privacy. Once connected, the AI can see what's there. If there are things in the inbox or drive that shouldn't be seen by the provider, that's worth knowing before connecting. Different providers have different data policies. Worth asking the AI itself: what happens to the data you see through this connection?
- Not everything connects. Some tools, some file types, some providers don't have connectors yet. That's when files get uploaded manually or content gets pasted — but know that it's a fallback, not the normal way.
The principle: don't describe what you can show. Don't remember what you can read. The AI is better with evidence than with testimony, same as any other intelligence.