Interface and chats: how not to get in your own way
AI from scratch: your digital assistant · Lesson 3 / 20
Interface and chats
ChatGPT, Claude and Gemini are laid out almost identically: input box at the bottom, message feed in the middle, your chat list to the side. Let's cover what actually affects the quality of your work, not just where the buttons are.
A chat is the unit of work
One chat = one context. Inside a chat the model remembers everything you've discussed and builds on it. That's a huge advantage: you don't re-explain the situation each time — "make it shorter" or "now the same for the second client" is enough.
There's a flip side. If one chat covers your holiday, then your CV, then a borscht recipe, the model drags all of it along. Answers get blurry, and fragments of old topics surface.
The new-chat rule
New, unrelated task — new chat. It's the simplest habit there is, it noticeably improves answers, and it costs nothing.
Worth learning on day one
- Start a new chat — one per separate task.
- Refine in the conversation — write "shorter", "add an example" as the next message instead of starting over.
- Copy the answer — use the copy button rather than dragging your mouse.
- Ask for alternatives — "give me three more versions", or hit regenerate.
- Rename important chats — so you can find your work a week later.
Example: how a junk chat ruins an answer
Open a new chat and the identical request produces a proper business letter. It's not the model — it's the accumulated context.
Cheat sheet
- One chat = one context.
- New topic — new chat.
- Refine in-conversation instead of restarting.
- Standing, named chats for recurring work.