How to Start a New AI Chat Without Re-Explaining Everything
Every new chat starts with the same five minutes. The project, the client, the constraints, what you already tried, why the obvious approach will not work. You have typed it so many times you could recite it, and you will type it again tomorrow.
The fix is not a better assistant. It is ten lines you write once and paste at the start of anything new. Here is the template, and the part most people get wrong about keeping it current.
Why the assistant cannot do this for you
It has the history, so it feels like it should be able to. Two reasons it cannot.
What it keeps about you is a summary rather than a record, assembled in the background and only partly visible. It knows you work on a thing. It does not know that on Thursday you dropped the second option and why. The mechanics of that are in what AI actually remembers.
More importantly, deciding what is still true is a judgement, not a retrieval. Your thread contains the decision and the later reversal, the assumption and the correction, in the order they happened. Nothing in there marks which one survived. You are the only one who knows, so the two minutes have to come from you.
The ten lines
Copy this. Fill it in once, then keep it current.
Objective. What this work is for, in one sentence.
Current status. Where it actually stands today. A version, a stage, a number.
Decisions made. The ones that hold, each with a short reason.
Rejected approaches. What you tried or considered and turned down, and why. Most valuable line in the list.
Constraints. Budget, deadline, platform, the client’s one rule, anything non negotiable.
Open questions. What is genuinely undecided, so it does not get answered by assumption.
Next action. The specific thing you are doing in this session.
Sources. Where the detail lives: the file, the doc, the thread.
Evidence. What you have checked yourself, as opposed to what an assistant told you.
Last updated. The date, and who touched it if more than one person does.
Ten lines, one screen. If yours runs longer, it has started absorbing history, and the detail belongs behind the sources line instead.
How to actually keep it up to date
Write it at the end, not the beginning. The two minutes after you stop working are worth ten the next morning, because you still know what happened. Beginning of session updates turn into archaeology.
Paste it as the first message of a new chat. Not as an attachment. A short block of text at the top of the conversation is the most reliable way to guarantee the assistant actually has it, for the reasons that make attachments unreliable.
Update three lines, not ten. Most sessions change the status, one decision and the next action. Rewriting the whole thing every time is what makes people quit.
Delete rather than append. When a decision is reversed, replace it and move the old one into your notes. A state document that only grows becomes a history document, and then nobody reads it.
If a thread has already grown unmanageably long before you started doing this, the recovery routine is slightly different, and a thread that has grown too long covers it.
Two ways people get this wrong
They write background instead of state. Three paragraphs on the history of the project, nothing on where it stands. Background is what you would tell someone at a party. State is what you would tell a colleague taking over tomorrow.
They keep it inside the assistant. A state document that lives in one app’s project folder is invisible the moment you open a different assistant, which is precisely when you needed it most. Keep it somewhere neutral: a text file, a note, a document, anywhere you can copy from in five seconds regardless of which tool you are in. The manual routine for carrying it between assistants is in moving context by hand.
When a file stops being enough
For one person on one project, a text file is the whole answer and you can stop reading here.
Three situations break it.
Several projects at once. Ten files in a folder, and you are now maintaining a filing system rather than a note.
Several people. The moment someone else needs to read or update it, a file on your machine is a bottleneck, and a shared document means chasing people to keep it current.
State that needs its sources attached. When half the entries need a link back to the conversation or document they came from, the note stops being enough on its own. You end up wanting the state and the material it came from in the same place, which is a different kind of tool. Which form solves which problem covers what those look like.
None of that changes the habit, only where it lives. The habit is the part that works immediately: two minutes at the end of a session, ten lines, pasted into whatever you open next.