You’ve introduced yourself to ChatGPT or Claude. Three messages later, the AI calls you by the wrong name—or suddenly switches to “you” and “your” without ever using your name again. It’s jarring, and it raises a practical question: do AI models actually remember who you’re talking about, or are they just guessing?
The short answer: AI models don’t have a reliable sense of identity within a conversation. They’re stateless systems that reconstruct context from scratch with every response. When they get your name right, it’s because your name appears in the recent conversation history. When they get it wrong, it’s usually because of how that history was processed—or because the model is hedging its bets.
Why Models Forget Names Mid-Conversation
When you chat with an AI, each response is generated by reading the entire conversation up to that point. The model doesn’t “remember” you in any human sense—it re-reads the transcript every time. If your name appeared early in a long chat, and the conversation has moved on, that early mention might fall outside the model’s active attention window.
Large language models use attention mechanisms to weigh which parts of the input matter most for the current response. If you introduced yourself ten messages ago and haven’t been referred to by name since, the model may deprioritize that information. The result: it defaults to second-person pronouns (“you,” “your”) or avoids reference entirely.
This is especially common in longer conversations where the context window is filling up. Even models with 128,000-token windows don’t treat all tokens equally—recent messages get more weight, and older details fade into the background.
When AI Uses the Wrong Name Entirely
Sometimes the model doesn’t just avoid your name—it calls you by someone else’s. This usually happens in one of three scenarios:
- Multiple people in the conversation: If you’re asking the AI to draft an email to “Sarah” or analyze a document written by “Michael,” the model may confuse pronouns and mistakenly address you by one of those names.
- Ambiguous references: If you say “my colleague thinks…” and later “he believes…,” the model may lose track of who “he” refers to, especially if the conversation shifts topics.
- Memory features: Tools like ChatGPT’s Memory or Gemini’s persistent memory store facts across sessions. If the AI saved a detail incorrectly—say, it stored your name as “Alex” when you’re actually “Alexis”—it will confidently use the wrong name every time.
You can check ChatGPT’s memory by going to Settings > Personalization > Memory, and delete or correct any entries that are wrong. Claude and Gemini Advanced have similar features, though the interface differs.
Why Models Avoid Names Altogether
Here’s a pattern you’ve probably noticed: AI models often avoid using your name even after you’ve introduced yourself. Instead, they’ll say “you mentioned,” “your question,” or “let’s work through this together.” This isn’t politeness—it’s risk avoidance.
Models are trained to be cautious about identity. Using the wrong name feels like a factual error, and modern AI systems are tuned to minimize confidently wrong statements. If the model isn’t certain of your name—because it’s been a while since you mentioned it, or because the context is ambiguous—it will default to second-person address. It’s safer.
This is also why AI rarely asks clarifying questions about identity. If you say “I’m working on this with Jordan,” the model won’t ask “Is Jordan your coworker or your manager?” It will infer from context or hedge with neutral phrasing. That’s a deliberate design choice: avoid making assumptions about people.
What You Can Do About It
If you want the AI to use your name consistently, the simplest fix is to reintroduce yourself periodically—especially in long conversations. You can also use a system-level instruction if the tool supports it. In ChatGPT, you can add “My name is [Your Name]” to Custom Instructions (Settings > Personalization > Custom Instructions). Claude’s Projects feature lets you set per-project context, including your name and role.
For conversations involving multiple people, be explicit. Instead of “he thinks,” say “Jordan thinks.” Instead of “my manager wants,” say “my manager, Sam, wants.” The clearer your references, the less the model has to guess.
And if the AI gets your name wrong? Just correct it in your next message. The model will adjust immediately—it’s reading the whole conversation again anyway.
Want one useful AI insight like this in your inbox every day? Subscribe to the One Two Three AI newsletter and get smarter about how AI actually works—no hype, just practical takeaways you can use.