Useful Hack: How to Make ChatGPT Remember What Matters Most - Protocol

In this post, I’ll share a highly effective and rarely discussed strategy for maximizing ChatGPT’s long-term usefulness: how to build a functional protocol that the model can remember and execute consistently across sessions.

1. Why Does Memory Matter?

Unlike earlier versions, ChatGPT (starting with GPT-4 with memory enabled) can now retain information across conversations. This makes it dramatically more efficient for ongoing work in research, writing, coding, or business strategy.

But this memory isn’t magical or all-knowing. It has boundaries:

• It only remembers what you explicitly tell it to.

• You can view, edit, or delete memory items.

• It doesn’t infer intent—it follows protocol, not assumptions.

The memory is allocated in Settings --.> Personalization --> Manage Memory

2. What Did We Do?

We built a system: the XXXX Protocol v7.7. It doesn’t just store data—it defines behavior, filtering, priority levels, structural rules, and response logic.

This protocol was embedded directly into ChatGPT’s memory through explicit instructions. For example:

“Apply the XXXXX Protocol v7.7 in all responses: do not complete unsought sentences, do not suggest tasks based on previous inputs, etc.” It is just a list of important things you need the chat to remember. Example for writers, could be a character or a scene. For mi looks more like a high-level directives:

• Rules for academic publishing.

• Prioritization of commercial operations.

• Structured, impersonal tone without artificial politeness.

• Cross-referencing between research projects and identities.

3. Where Is It Stored?

This protocol is now part of what OpenAI calls the Model Set Context — a persistent layer of memory that operates beyond single conversations. It lets the model act consistently even across time, without needing daily repetition of the same rules.

Memory, this is how it looks.

4. How Can You Do It?

Step 1: Write a clear, logically structured protocol.

• Example: 1. Contrast every thing you say to me with empirical subjects. 2. Give me the source of any answer. 3. Don't talk like a human .... (by the way, I took these three from my own Protocol that has almost 100 lines)

  
  

Step 2: Declare it explicitly. For example:

“From now on, apply Protocol X in all replies. It includes rules A, B, C...”

• Example, in any chat that you it to apply the protocol, just type something like ' Apply Protocol XXXX'.

• 
  

Step 3: Confirm that it’s remembered.

• If it si something important that you are working on, just ask something regular and check if the answer matches with the protocol that you wrote.

Ask the model:

[This is imprortant!] What do you know about Protocol X I asked you to follow?”

If it doesn’t respond correctly, reissue the directive and confirm memory retention.

Conclusion

This “hack” isn’t a hidden feature. It’s an epistemological framework applied to a probabilistic language model. It’s not about making the AI think for you—it’s about building a rule system it can execute autonomously inside your projects.

That transforms ChatGPT from a passive assistant into a rule-bound agent—not with subjectivity, but with structural execution.

Want a real-world example?

You're reading a blog post written by a model that remembers and obeys the Startari Protocol.

Not magic. Just structure.

Research Profiles & Publications

ORCID: 0009-0001-4714-6539

Zenodo: https://zenodo.org/

Medium: https://medium.com/@agustinstartari

ResearcherID: NGR-2476-2025

Author Ethos

I do not use artificial intelligence to write what I don’t know. I use it to challenge what I do. I write to reclaim the voice in an age of automated neutrality. My work is not outsourced. It is authored.

— Agustin V. Startari