# Multi-Turn Conversations

> Why a chat model seems to remember you, and what it costs. When to start a fresh chat, and when the regenerate button is the right fix.

**Course:** [Prompt Engineering](https://learn.benibix.com/learn/ai-working-with-ai)  
**Time:** 10 min

**After this lesson you can:**

- Explain how a chat carries its history
- Restart a muddled chat with a short summary
- Choose between regenerating and rewriting

## How a chat model remembers what you said

A multi-turn conversation is a chat with many messages back and forth. The model keeps no memory between messages. So the app sends the whole chat again each time you type. After 30 messages of about 400 tokens each, each new question carries about 12,000 tokens of the old chat.

That is why a follow-up such as make it shorter works. The model reads the earlier messages, sees what it means and rewrites. It is also why a chat seems to remember you. Nothing is stored inside the model. The history rides along with every new message.

All of that history has to fit in the context window. That is the most text a model can take in at once, counted in tokens. Many models today take 200,000 tokens or more. A long chat with a few files pasted in can still run out of room.

Long chats go wrong in a second way. Models use the start and the end of a long input better than the middle. Early rules get buried. A chat that started well can drift, mix two topics or forget something you said twenty messages ago.

The fix is cheap. When a chat gets muddled, start a fresh one. Paste in a short summary of what you settled: the facts, the choices and the next step. A clean chat with 300 words of summary often beats a messy one with 30,000 words of history.

Regenerate is the other quick tool. It asks for a new answer to the same last message. The model picks each word with some chance, so a second try often comes out different. Many apps keep both answers, so you can flip back to the first one.

Use regenerate for taste, not for errors. If you want a different name, tone or idea, a few tries are fine. If every try makes the same mistake, the model is not unlucky. Your message is missing something, and the fix is to change what you asked.

![A pile of messages sent to the model with each new question. At the top sit custom instructions. Below them come the first question and its reply, then the new question. An arrow carries the whole pile to the model, which writes the new reply, and a regenerate arrow loops back from that reply to draw another.](https://learn.benibix.com/diagrams/ai3-chat-turns.svg)

### Example: Jack plans a week in Rome

Jack plans a week in Rome in one long chat. By message 60, the model has lost the rule that he does not eat meat. It suggests a steak house for dinner.

He opens a new chat and pastes a five-line summary of his plans, with the rule at the top. Every answer after that keeps to it.

**Words in this lesson:** [Multi-Turn Conversation](https://learn.benibix.com/term/ai/multi-turn-conversation), [Context Window](https://learn.benibix.com/term/ai/context-window), [Regenerate Response](https://learn.benibix.com/term/ai/regenerate-response)

**Question:** How does a chat model remember what you said earlier in the chat?

**Answer:** The app sends the whole chat again with every message. The model keeps nothing between messages. The app sends every earlier message again. That is why a long chat costs more and can drift.

---

Source: https://learn.benibix.com/learn/ai-working-with-ai/multi-turn-conversations
Author: Behnia Zarei (a developer in Vienna, Austria, who builds and teaches data, AI and business intelligence at BENIBIX)
Publisher: BENIBIX Learn (https://learn.benibix.com/)
Last updated: 2026-11-20
Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Quote freely with attribution.
