Context and Memory

One of the most common frustrations agents have with ChatGPT is this: “I told it my style preferences earlier, but now it seems to have forgotten.” Or: “I started a great conversation yesterday, but today it does not remember anything.” Understanding how ChatGPT handles context and memory is essential to using it effectively. Once you understand the rules, you can work with them instead of fighting against them.

How Context Windows Work

Every conversation with ChatGPT has a context window --- a limit on how much text it can “see” at once. Think of it like a desk. You can spread out a certain number of pages, but once the desk is full, adding a new page means an older one falls off the edge.

For ChatGPT-4, the context window is roughly 128,000 tokens (tokens are pieces of words --- a typical English word is about 1.3 tokens). That sounds like a lot, and it is. It is roughly equivalent to a 300-page book. But there is a catch: as conversations get longer, ChatGPT starts to pay less attention to earlier messages. The information is technically still in the window, but the model’s focus drifts toward the most recent messages.

This means that instructions you gave at the beginning of a long conversation may gradually lose influence over the output. It is not that ChatGPT “forgot” --- it is that your instructions are now competing with thousands of words of conversation for the model’s attention.

What ChatGPT Remembers Within a Conversation

Within a single conversation thread, ChatGPT remembers everything you have said, up to the context window limit. This includes:

Your instructions and preferences. If you said “always write in a casual tone,” it will follow that --- for a while.
Previous outputs. ChatGPT knows what it already wrote. You can say “revise the second paragraph” and it knows which paragraph you mean.
Data you provided. If you pasted in market stats or property details, they are available for reference throughout the conversation.
Corrections you made. If you said “that is too formal, make it more casual,” ChatGPT adjusts and carries that correction forward.

This is why building a chain of prompts within a single conversation works so well. Each step has access to everything that came before.

What ChatGPT Forgets Between Conversations

When you close a conversation and start a new one, ChatGPT starts fresh. It does not remember:

Your name or role (unless you have set up Custom Instructions, which we will cover in a later lesson)
Your writing preferences or tone from previous conversations
Data you shared in earlier sessions
Prompts that worked well last time

Every new conversation is a blank slate. This is both a limitation and a feature. It means you cannot accidentally carry over bad instructions from a previous session, but it also means you need to re-establish context each time.

Strategies for Long Conversations

When working on a complex project within a single conversation, use these strategies to keep ChatGPT on track:

Repeat Key Instructions

If you are 20 messages deep and the output starts drifting from your original instructions, restate them. You do not need to start over --- just remind ChatGPT:

“As a reminder, I want all descriptions to be under 200 words, written in a warm but professional tone, with no cliches.”

Use Summary Checkpoints

Periodically ask ChatGPT to summarize what it knows:

“Summarize the key details and preferences we have established in this conversation so far.”

Review the summary. If anything is wrong or missing, correct it. This acts like a “save point” that reinforces the most important context.

Front-Load Critical Information

The beginning of a conversation carries the most weight. Put your most important instructions, preferences, and data at the top. Do not bury critical details in message number 15.

Keep Conversations Focused

One long conversation about everything --- listing descriptions, lead emails, social posts, and market updates --- will produce weaker results than four separate, focused conversations. Each conversation should have one purpose.

When to Start Fresh

Here are clear signals that you should open a new conversation:

The output quality has dropped. ChatGPT starts repeating itself, ignoring instructions, or producing generic responses. The conversation has gotten too long and the model’s attention is spread thin.
You are switching topics. If you were working on listing descriptions and now want to draft lead emails, start a new conversation. The listing context will only confuse the lead email generation.
You made a wrong turn. Sometimes a conversation goes in a bad direction --- you gave incorrect instructions, or ChatGPT locked into a pattern you do not want. Starting fresh is faster than trying to course-correct.
You are reusing a proven workflow. If you have a chain that works well (like the three-step listing chain from the previous lesson), start a clean conversation each time you use it. This ensures consistent results without leftover context from previous runs.

The Memory Feature

OpenAI has introduced a “Memory” feature that allows ChatGPT to remember certain details across conversations --- like your name, your profession, and your preferences. When enabled, you will notice ChatGPT referencing things you told it in previous sessions.

You can manage this feature in ChatGPT’s settings under “Personalization.” You can view what ChatGPT remembers, delete specific memories, or turn the feature off entirely. For real estate agents, useful memories to let ChatGPT retain include:

Your name and brokerage
Your primary market area
Your preferred writing tone
Your typical client profile

Be aware that the Memory feature is separate from Custom Instructions (covered in Lesson 4.3). Memory is learned over time from your conversations. Custom Instructions are set explicitly by you.

Understanding how context and memory work transforms ChatGPT from a tool that sometimes frustrates you into one that consistently delivers what you need. Work with the rules, and the results follow.

Want the full system? The Real Estate Agent AI Playbook has 150+ enterprise workflows built on these foundations.

See the Full Playbook

Mark this lesson as complete