Why long Claude chats get worse and when to start fresh

If you use Claude for work, you have probably noticed something odd. The conversation starts well. Claude understands what you need. But after a while, it forgets things or makes mistakes you already fixed.

This is not random. It happens because of how Claude's memory works.

Why this matters

A long conversation can cost you time. You repeat yourself. You fix the same error twice. The answers get worse instead of better.

Knowing when to start fresh saves that time. It also gets you better results.

How it works

Anthropic's guide explains that Claude has a context window. That is its working memory for the conversation.

The context window holds every message you send. It holds every file Claude reads. It holds every result Claude writes.

It fills up fast.

When the context window fills, performance gets worse. Claude may start forgetting earlier instructions. It may make more mistakes.

Think of it like a desk covered in papers. At first you can find what you need. After a while, important notes get buried under new ones.

How to do it

Here is when to start a new conversation.

1. Between unrelated tasks

If you finish one piece of work and start something different, open a new chat. Anthropic recommends clearing or starting a new session between unrelated tasks.

2. After correcting the same mistake twice

Anthropic says that if you have corrected Claude more than twice on the same issue, the conversation is cluttered with failed attempts. A fresh session with a better first prompt almost always does better.

When you start the new chat, write a clearer first message. Say what went wrong last time and what you need instead.

3. Before a big piece of work

If you are about to ask Claude to write a long document or work through a complex problem, start fresh. Give it the full context in one clear opening message.

4. Save decisions before you switch

At the end of a long piece of work, ask Claude to write a short summary. Then paste that into a new conversation.

Try this prompt:

> Please write a short summary of the decisions we made and the next steps, so I can continue in a new conversation.

Copy the answer. Open a new chat. Paste it in with your next question.

Common mistakes

Pushing through when things go wrong

If Claude keeps misunderstanding you, stop. Starting fresh almost always works better than trying to correct the same problem again.

Losing track of what you decided

If you close a long conversation without saving the key points, you lose that work. Ask for a summary first.

Starting fresh too often

You do not need a new chat after every reply. If the conversation is working and the task is related, keep going.

One thing to try this week

Next time you notice Claude repeating a mistake, stop. Ask it to summarise what you have done so far. Then open a new conversation. Paste the summary in and write a clearer version of what you need.

I find this saves more time than it takes. The second attempt usually lands.

You can read more in Anthropic's Claude Code best practices guide.

Claude and other AI tools can save your business time. What could your business automate? is a free check that shows what else might work for you.