Guide
How to stop pasting context into ChatGPT
A lot of every ChatGPT prompt is background: who the client is, what was said on the call, what you tried already. Here's how to give ChatGPT that once, and keep it current without retyping it.
Why you keep pasting
ChatGPT is only as useful as what it knows about your situation. When the answer comes back generic, the usual reason is missing context, not a weak model. So you paste in more: the email thread, the meeting notes, the doc. Each paste takes a minute, and you lose a little each time because you summarize.
There are four layers of context you can give ChatGPT. The first three are built in. The fourth is the one most people are missing.
1. Custom instructions: who you are
Custom instructions apply to every chat. Use them for things that are always true: your role, your company, the tone you want, the formats you like. Keep them short and concrete. “I run marketing at a 20-person B2B SaaS company; answer in plain English, no bullet-point summaries” does more than a page of adjectives.
2. Projects: one body of work
A Project groups chats with shared instructions and files. Make one per client, product or big piece of work. Put in:
- Instructions that say what the project is, who it's for, and what good looks like.
- A short summary document that you keep current.
- The few files you keep pasting from.
The weak spot is “keep current”. A Project with last month's notes gives last month's answers.
3. Memory: what comes up in chats
ChatGPT's memory picks up facts from your conversations and can refer back to earlier chats. It's convenient, and you can see and delete what it keeps in settings. It only knows what you've told ChatGPT, though. The client call, the Slack thread and the doc you edited all happened somewhere else.
4. Context from outside ChatGPT
Most of your work happens outside ChatGPT, which is why the first three layers leave a gap. To close it you need something that records your work as it happens and turns it into documents ChatGPT can read.
That's what Cachejay does on a Mac. It reads the text in the apps you use, transcribes your calls on the Mac, and writes it all up as Markdown in ~/Cachejay, with a page per project and a note for each call. You can add those files to a ChatGPT Project if you like.
You can also skip the files: add Cachejay to ChatGPT as a custom connector (https://cachejay.com/mcp), and ChatGPT reads your days, calls and projects itself, on the web and on your phone. How to connect it.
A routine that works
- Once: write custom instructions, and create a Project for each client or product.
- Weekly: refresh each Project's files with the latest project page and weekly summary from your Library.
- Daily: start chats inside the right Project and ask directly: “draft the update for Acme from this week.”
Once ChatGPT has your context, it can also act on it. With its connectors for tools like Jira, HubSpot or Notion, you can ask it to post the update rather than just write it. See keeping your tools current with AI.