Search

FOUNDATIONS

Context: The ongoing conversation

When you ask a new question in an existing chat, the AI does not only have your latest message available. The previous conversation also feeds into the next answer. This context is what makes a conversation feel coherent. You can write: “And what do you think about point 3?” – and the system knows what you are referring to.

Important: Context is not the same as Memory. Memory can make information available across multiple conversations. Context is the temporary working space within this one conversation.

The limited desk

This working space is not a static store. It is more like a desk with limited room to keep everything in view:

  • Attention gets diluted: As a conversation becomes very long, older details can slip out of focus or carry less weight than what was said just moments ago. An instruction from the beginning of the chat can gradually lose influence.
  • Old topics can interfere: If you first talk about holiday destinations and then switch to tax questions, the beach holiday may still be sitting on the desk. That can affect which connections the model makes.
  • Errors can carry forward: If a wrong number or assumption remains in the chat, the AI may continue to use it as part of the conversation’s working basis in later answers. Even a later correction does not simply erase the original statement from the conversation – both versions can remain part of the context.

THE RUNNING EXAMPLE

Mara plans a trip

Mara tells the AI that she wants to travel with her sister for five days and spend no more than €1,200 per person. The AI suggests three cities. A few messages later, Mara simply asks:

“Which of those would you recommend?”

The AI can understand which cities “those” refers to and can still take the budget into account. Both pieces of information are already part of the current conversation. As long as they remain available to the AI, its answer can build on them.

Here, the previous conversation provides the information. The budget comes from this chat – not from an earlier one.

Your leverage in everyday use

  • New topic, new chat: Start a new chat for each independent task. That reduces the chance that old baggage will blur the next answer.
  • Reset when things go wrong: If the AI gets stuck, repeats the same mistakes or ignores instructions, do not argue with it indefinitely. Ask it to produce a clean handover for a new chat:

“Summarise our current goal, the established results and all remaining constraints concisely so I can continue in a new chat. Leave out approaches we have already discarded.”

The key point: Context is the desk of your conversation. What is sitting on it can influence the next answer – for better or worse.