ChatGPT can now reference your past chats across conversations

Starting a fresh chat with an AI usually means dealing with total amnesia. You have to explain your job, your tech stack, and your writing tone all over again.

OpenAI is finally fixing that friction. The company announced on Thursday that it has started rolling out a memory upgrade that lets ChatGPT pull context across your past conversations to inform new answers.

No more repeating yourself

Found in your settings under reference saved memories, the feature is designed to make ChatGPT feel much more personalized. Crucially, the added context works across text, voice mode, and image generation.

OpenAI rolled out a basic memory tool last year, but that version mostly relied on explicit user prompts to log specific facts, like telling the bot what programming language you prefer. This new update operates more seamlessly in the background, automatically referencing previous discussions.

If you already had ChatGPT’s memory toggled on, this new cross-chat feature will be enabled by default.

Paid users first, Europe waits

The feature is currently exclusive to paying ChatGPT Plus and Pro subscribers. An OpenAI spokesperson told TechCrunch that the team is focused solely on paid tiers for now, with no timeline shared for free users.

Geography also matters here. The rollout completely excludes the U.K., EU, Iceland, Liechtenstein, Norway, and Switzerland. OpenAI noted that these regions require additional external reviews to comply with local privacy regulations, though it plans to bring the feature there eventually.

For privacy-minded users who do not want their past conversations cataloged, OpenAI offers an opt-out in settings. You can review and delete specific stored memories, ask ChatGPT what it knows about you, or switch to Temporary Chat when you want an off-the-record session.

Why it matters

For everyday power users, this is a massive quality-of-life improvement. If you use ChatGPT to draft newsletters, write code, or brainstorm campaigns, you no longer have to paste three paragraphs of background context at the start of every single prompt. Being able to keep consistent context across voice and image generation makes the assistant feel like a continuous partner rather than a disposable scratchpad.

In our view, this also significantly raises the platform lock-in. Google introduced a similar memory upgrade to Gemini back in February, and OpenAI had to counter. Once an AI assistant accurately remembers your preferences, project histories, and quirks from hundreds of chats, the friction of switching to a rival model becomes immense.

OpenAI's rollout strategy looks very calculated. Restricting it to paid tiers gives free users a compelling reason to upgrade, while dodging the European regulatory buzzsaw prevents costly compliance headaches at launch. The tricky part will be managing retrieval mistakes. If ChatGPT pulls an outdated preference from three months ago and applies it to a fresh project, untangling that bad memory could become a real headache.