TL;DR:
An assistant can only consider a limited amount of text at once. That limit is its context.
In a very long conversation, early details can fall out of view or get blurred.
Summarise what matters, restate key facts, and start a fresh conversation when a thread gets long.
You have spent an hour with an assistant on a school project. Early on you told it the project was about water use in your town, and that the report is for ninth grade. Now, forty messages later, it suggests a statistic about a different country and writes in a tone that suits a college paper. Nothing broke. The assistant just no longer has your early instructions clearly in view.
This is a context problem, and understanding it makes long conversations much easier to manage.
What context means
Context is everything the assistant can look at while it writes its next reply: your messages, its own earlier replies, any documents or screen content you gave it, and any instructions it was set up with. You can think of it as a desk. Whatever is on the desk can be used. Whatever is not, cannot.
The desk has a limited size, usually called the context window. Text is measured in small pieces called tokens, which are roughly parts of words. Different systems have different sizes of desk. When a conversation grows past the size, something has to go.
Walk-through: a long conversation about one project
Start: Message 1: "I am writing a ninth grade report on water use in my town. Keep the language simple." This is the most important message in the whole chat.
Middle: Over many messages you explore sources, outlines and drafts. Each reply adds more text to the desk.
Drift: Late in the conversation, the early message may be far back, or squeezed out, or just given less weight than all the text since. The assistant's replies slowly stop fitting your first instructions.
Fix 1: restate: You write: "Reminder: ninth grade report, water use in my town, simple language." The key facts are back at the front.
Fix 2: summarise and restart: You ask: "Summarise what we decided so far in ten lines." You open a fresh conversation, paste the summary, and continue from there.
The fresh start has a cost, because you lose the detail of the old chat, but a clean desk with the right ten lines on it often works better than a cluttered one with everything.
A useful way to picture the drift is a whiteboard that keeps filling up. Eventually, someone has to erase the oldest notes to write new ones, and the person erasing does not always know which notes you cared about most. You are the only one who knows which lines on the board must stay, so you are the best person to write them again.
Signs that context is the problem
It forgets something you stated clearly earlier.
It contradicts a decision you both made.
Its style drifts from what you asked for.
It repeats suggestions you already rejected.
It starts making up details to fill gaps.
Habits that help
Put the important facts in one place at the start of a message or in a short summary, not scattered across a long chat.
Restate constraints when you change direction: audience, length, tone.
Summarise at milestones. After each stage, ask for a short list of decisions.
Start fresh for a new topic. Unrelated threads add clutter.
Paste in what it needs. If a document matters, share it again instead of assuming it remembers.
Tip: A good test: if a new teammate had only your last message, could they do the job? If not, add what is missing to the message.
Context is not the same as memory
The desk empties when a conversation ends, unless a tool is built to save things. Some assistants have a separate memory feature that keeps selected facts between conversations. That is a design decision, with privacy trade-offs, and it is a different thing from the context window. See what an assistant should remember about you.
Pasting versus pointing
When a document matters, give it to the assistant in the current conversation, whether by pasting the relevant part or by showing it on screen, instead of saying "remember the file from earlier". Only the portion you share is on the desk. Sharing a short relevant extract usually works better than sharing a whole long file, because less irrelevant text competes for attention.
What this does not mean
A bigger desk does not guarantee better answers. Even when everything fits, an assistant can pay less attention to material buried in the middle of a very long text. Shorter, well-organised context beats a huge pile.
Privacy and control
Anything you put in a conversation is part of its context, including what is visible on your screen when you ask. DuckyHelper looks at your screen only while the pill is open and for that answer, which keeps what it sees limited to the moment you chose.
Try it yourself
In your next long chat, write a three-line "project card" with the goal, the audience and the rules. Paste it at the top whenever the conversation shifts. See if the replies stay on track longer.
Frequently asked questions
Can I just keep chatting forever in one conversation?
You can, but quality can drop as it grows. Summarise and restart when you notice drift.
Does the assistant forget on purpose?
No. It has a limited view, and older text can fall outside it or be weighed less.
Is a longer context always better?
Not always. Clear, relevant context beats a large pile of text.
Get started
