I misunderstood what a GPT actually is. That’s on me. But I also believe OpenAI let the word "custom" do far too much heavy lifting. And the result is a tool that feels like it was designed to impress first-time users—not to build trust with serious ones.
When I saw the phrase Custom GPT, I imagined something persistent. Something mine. I expected to create a GPT that could learn from me, grow with me, and carry forward my voice or thinking. I thought I was building a relationship. Instead, I was handed a blank canvas—every time.
A stateless GPT is like a dictionary you’re allowed to annotate, so long as all your notes evaporate the moment you close the book. You can look up words, but you can’t build a language. You can ask questions, but you can’t develop a relationship. It remains brilliant—and forgetful.
In hindsight, maybe I was naive. But I wasn’t wrong to expect more. The word custom carries weight. It means tailored. It means adaptive. It means I can shape it, and it will hold that shape. What OpenAI delivered was something far more limited: a template with a clever set of instructions and no memory.
Had they called it GPT Lite, I would have looked for what was missing. I would have asked better questions. But "Custom GPT" suggests agency, ownership, permanence. And that, frankly, was a setup.
To make matters worse, even the sense of authorship is illusory. I created a Custom GPT. It bore my name. It followed my instructions. But I couldn’t edit it. I couldn’t delete it. I could only hide it from my sidebar and try not to let the irony bother me. The system gave me authorship without agency. A voice, but no control.
Maybe OpenAI should rename the feature something more honest. Custom-but-Very-Forgetful GPT has a nice ring to it. At least then, we’d all know not to expect a lasting relationship from something built to ghost us after every session.
I’m not here to rant. I believe in what these tools can become. But language matters. And if OpenAI wants to keep the trust of thoughtful users, it needs to stop using words that overpromise and interfaces that underexplain. Memory is not a luxury. It’s not a bonus feature. It’s the bridge between interaction and relationship. Between novelty and depth. Between toy and tool.
Memory is everything. (Really.)
Postscript:
For those tallying em dashes—yes, I had help. The very tool I’m criticizing lent its synthetic pen. I considered editing out its fingerprints, but thought better of it. Let the punctuation stand as evidence: even a forgetful assistant can leave a distinctive mark.
(That level of parched precision was only possible with the help of a very large language model.)
Em dash count: 11. You're welcome.
There are many assumptions that can be made on accident with this stuff.
One measure of these systems that’s worth looking into is “context window”. Another is called “RAG”, these both are “memory”. I think constantly about limitations of context window but I don’t believe others do.
The forgetfulness is something I think of as a good thing, but I think that’s a biased perspective from my software development side of me. I believe I more commonly see people wanting it to remember more about themselves.