TL;DR:
An assistant answers a follow-up by reading the earlier turns of the conversation again, together with your new question.
Short words like "it", "that" and "why" work because the earlier turns tell the assistant what they refer to.
This memory is limited to one conversation. Long chats can lose early details, so restate anything important.
You ask a spreadsheet helper why a cell shows an error, and it explains. Then you type two words: "and why?" A person would know you mean the part of the explanation you did not understand. A computer program has no such instinct, so it is fair to wonder how that two-word question gets a useful answer at all.
The short answer
An assistant does not hold a thought between your messages the way you do. Each time you send something, it is given the conversation so far, your earlier questions and its own earlier answers, followed by your new message. It reads all of that as one piece of text and writes the next reply.
That bundle of earlier turns is the conversation context. When your new message says "and why?", the words that tell the assistant what you mean are sitting right above it in the context. Nothing mysterious is remembered. The earlier turns are simply read again.
What happens step by step
You ask the first question and the assistant answers it. Both are added to the conversation.
You send a short follow-up. It is added to the end of the same conversation.
The assistant reads the whole thing again, from the first message down to your latest one.
It works out what the short words point to. This is called clarifying references: "it" becomes the cell, "that" becomes the last step it described, "why" becomes a request for a reason behind its last claim.
It writes the reply and that reply is added too, ready for the next follow-up.
Because the previous answer is part of the input, you can also steer it with the smallest corrections: "shorter", "in simpler words", "the second one". Each of those only makes sense next to what came before.
Walk-through: asking "and why?" after an answer
Imagine a spreadsheet where cell D2 holds the formula =B2/C2 and shows #DIV/0!. You ask, "What does this error mean?"
The assistant answers that the formula divides by zero, which a spreadsheet cannot do, so it shows an error instead of a number. That is turn one.
Now you type, "and why?" The assistant has to decide which claim you want justified. Reading the context, the most recent claim is "the formula divides by zero", so it explains the cause: C2 is empty, and a spreadsheet treats an empty cell as 0 in a division. Dividing any number by 0 has no answer, because no number multiplied by 0 gives back a non-zero result.
You then say, "so how do I stop it showing that?" The word "it" means the error in D2, and the assistant can suggest filling in C2, or wrapping the formula so it shows a blank when C2 is empty. Three turns, and you never repeated the cell name or the formula.
Check the logic yourself: if B2 = 12 and C2 = 4, the result is 12 / 4 = 3, and multiplying back gives 3 × 4 = 12. With C2 = 0 there is no number that multiplied by 0 gives 12, which is why the spreadsheet refuses.
What it can't do
Remember across separate conversations by default. A new conversation starts with an empty context unless the product deliberately stores notes about you. Do not assume it knows what you asked yesterday.
Hold unlimited history. There is a limit to how much text it can read at once. In a very long chat, early details can be dropped or get less attention.
Read your mind when the reference is ambiguous. If your last message could mean two things, a good assistant asks which one. A careless one picks and may be wrong.
Notice that you changed topics. If you say "and why?" after a break, it will still tie the words to the last answer, even if you meant something older.
When a conversation goes badly, the most reliable fix is to restate the key facts in one message: what you are working on, what you have tried, and what you want now. That gives the assistant a clean copy of what matters instead of relying on what is buried far up the chat.
Privacy and control
Everything in the conversation is input to the assistant each turn, so anything you paste or show early on keeps being part of what it reads. If you shared something private for one question, start a fresh conversation for unrelated topics. Treat the conversation like a document you keep adding to: it contains all of it.
Check where the product keeps conversations and whether you can delete them. Those settings differ from one product to the next, so look at the product's own privacy page rather than guessing.
DuckyHelper is one assistant you can talk to out loud, and a follow-up spoken a moment later works the same way: it is understood next to the answer you just heard. It only listens and looks while the pill is open, and for that answer.
Try it yourself
Ask any assistant a question you only half understand. Then follow up with the shortest possible questions: "why?", "what does that mean?", "give me an example". Notice that each one works because of the answer above it. Next, switch to a brand new conversation and type only "and why?" It has nothing to refer to, and you will see the difference context makes.
Frequently asked questions
Why does the assistant sometimes forget something I said earlier?
Only so much text can be read at once, and details far back in a long conversation can be lost or given less weight. Restate the important fact in a new message.
Should I repeat my whole question in a follow-up?
Not usually. Short follow-ups are fine while the topic is the same. If the answer drifts, add the missing detail in one clear sentence.
Does a new conversation know what I asked in an old one?
Not unless the product stores notes about you on purpose. By default, each new conversation starts empty.
Get started
