The essentials in 30 seconds
ChatGPT has of memory systemsthe saved memories (explicit data you ask to remember, limit of ~1,200 to 1,750 words (according to the plan)) and the Reference to chat history (lessons from past conversations). Furthermore, you can use Personalised instructions y Custom GPTs as complementary ways to provide context. The current context window in GPT-5.4 is 272K tokens (expandable to 1M in Codex). All this is managed from Settings > Personalisation.
ChatGPT no tiene memoria en el mismo sentido que los seres humanos. No tiene recuerdos personales ni experiencias. En cambio, su "memoria" se refiere a su capacidad para retener el contexto dentro de una conversación.Esto se logra de varias maneras:* **Memoria contextual a corto plazo:** ChatGPT puede recordar y hacer referencia a lo que se ha dicho anteriormente en la conversación actual. Cuanto más reciente sea la información, mejor podrá recordarla y utilizarla. Sin embargo, esta memoria tiene un límite. Una vez que la conversación se alarga demasiado, la información anterior puede empezar a "olvidarse" o a tener menos impacto en sus respuestas. * **Sin memoria a largo plazo entre conversaciones:** Cada conversación con ChatGPT es, en gran medida, independiente. No recuerda conversaciones pasadas que hayas tenido con él en el pasado. Cada vez que inicias una nueva conversación, es una "pizarra en blanco" en términos de historial de interacciones previas. * **Entrenamiento en datos masivos:** La "memoria" más profunda de ChatGPT proviene de los datos masivos con los que fue entrenado. Estos datos le han enseñado patrones, información, hechos y estilos de lenguaje. Es esta vasta base de conocimientos, y no una memoria personal, lo que le permite responder a una amplia gama de preguntas y generar texto coherente.En resumen, ChatGPT no tiene "tipos de memoria" como la memoria a corto plazo, a largo plazo o episódica, tal como la entendemos en los humanos. Su capacidad se basa en mantener el contexto de la conversación actual y en el conocimiento derivado de su extenso entrenamiento.
When we talk about “memory” in ChatGPT, we're not referring to a single mechanism. In reality, there are four distinct ways How AI retains and uses information about you.
Conversation context
Everything you write within an active chat remains in the context window. GPT-5.4 handles up to 272K tokens per session, enough for extensive documents.
Stored memories (Memory)
Data that ChatGPT explicitly saves between conversations: your name, preferences, projects. Limit of ~1,200 to 1,750 words (depending on the plan) in total.
Personalised instructions
Fixed directives that you configure manually to define tone, format, or rules. They apply to all conversations without taking up memory space.
Custom GPTs
Custom versions of ChatGPT with their own instructions and reference files. Each GPT has its own context, independent of your main memory.
How does saved memory work?
From the official announcement from OpenAI, the Memory feature allows ChatGPT retain information between conversations. It works in two ways:
Explicit memories You tell it “remember I am vegetarian” and ChatGPT stores this as a memory entry. These memories are automatically injected into the context of every new conversation.
Reference to history: From April 2025, ChatGPT can also consult your previous conversations to find relevant context, even if you haven't explicitly saved anything. Plus and Pro users get a deeper, long-term understanding, while Free users receive a lighter version with short-term continuity.
Comparison: types of memory, functioning and limits
| Type of memory | How does it work | Limits |
|---|---|---|
| Conversation context | The entire active session remains in the model's context window. | 272K tokens (GPT-5.4); lost when closing chat |
| Stored memories ⭐ | ChatGPT stores data between sessions (name, preferences, projects) | 1,200 to 1,750 words (depending on the plan); 100–200 entries |
| Chat history | The AI consults past conversations to extract relevant context. | Plus/Pro: long-term understanding; Free: basic continuity |
| Personalised instructions | Fixed rules that you write in Settings > Customisation | Approximately 1,500 characters per section |
| Custom GPTs | Instructions and associated files for each GPT created | No shared memory with your main account; files limited by plan |
How to optimise ChatGPT's memory. 5 practical steps.
Memory is powerful, but it has limits. These steps help you make the most of it without wasting space.
- Be explicit about what you want me to remember — Don't expect ChatGPT to guess. Say clearly: “Remember I work in digital marketing and prefer direct answers.” The more specific you are, the better it works.
- Regularly audit your memories — Ask them “What do you remember about me?” to see the full list. Remove obsolete entries since Settings > Personalisation > Manage Memory.
- Separate memory and custom instructions — Use the Personalised instructions for fixed rules (tone, format, language). Use the Memory for dynamic data (your company's name, current projects).
- Use Projects for sustained work The ChatGPT Projects they allow uploading reference files (up to 20 files per project) and maintaining context for weeks without filling general memory.
- Use Temporal Chat for private conversations — Si no quieres que una conversación afecte tu memoria ni tu historial, activa el modo de incógnito. Temporary Chat. It does not read existing memories or create new ones.
Pro tip: If your memory is full and ChatGPT stops saving new things, don't delete everything. Review the entries and consolidate related items In a single, more concise instruction. This frees up space without losing important context.
Limitations and precautions you need to be aware of
- Memory is NOT an infinite database. With a capacity of around 1,400 words, it fills up quickly if you save a lot of details. Prioritise the essentials.
- Privacy and sensitive data. Everything ChatGPT remembers is stored on OpenAI's servers. Do not ask it to remember passwords, bank details, or sensitive medical information.
- GPTs DO NOT share memory with your account. Each custom GPT starts from scratch. It doesn't inherit your saved memories or your custom instructions.
- The reference to the history is not perfect. ChatGPT can recall previous chats, but it doesn't retain every detail. For critical data, keep an explicit record.
- You have full control. You can disable saved memories and history references at any time from Settings > Personalisation.
Frequently asked questions
There is no limit to the amount of information ChatGPT can recall, as it is a language model. However, the accuracy of the information it recalls may vary depending on the training data and the complexity of the query.
Saved memories are limited to approximately 1,200 to 1,400 words, which equates to between 100 and 200 individual entries. Furthermore, the context window per conversation in GPT-5.4 is 272K tokens. If you need it to retain more data long-term, use Personalised instructions as a supplement.
Does ChatGPT's memory work on the free version?
Yes, since June 2025, Free users have a Lightweight memory version which offers basic continuity between conversations. However, the full version with long-term user understanding is available only to Plus, Pro, and Enterprise.
How can I delete what ChatGPT remembers about me?
You have several options: you can tell them directly “forget I'm vegetarian”, go to Settings > Personalisation > Manage Memory to delete specific entries, or clear all memories at once. See the Official OpenAI FAQ for more details.
Do custom GPTs use my memory?
No. Custom GPTs do not access your saved memories, custom instructions, or conversation history. Each GPT has its own isolated context. If you'd like to know more, I can explain the differences in this guide to ChatGPT as a virtual assistant.
What is the difference between memory and context window?
The context window is the amount of text the model can process in a single conversation (272K tokens in GPT-5.4). It is emptied when a new chat is started. The Memory, However, it persists between conversations and is automatically injected into every new chat. They are complementary: the window handles the “now” and the memory handles the “always”. It learns to to empower both here.
Make ChatGPT work for you, not the other way around.
Configure your memory today and notice the difference from the very first conversation.
Updated: March 2026. Sources: OpenAI Blog, OpenAI Help Centre.
You may also be interested in
EN 
ES 
PT 
DE 
FR