Tech giants including OpenAI, Google, Meta, and Microsoft are racing to give their AI chatbots better memory—ushering in a new era of personalization, but also sparking concerns over privacy and manipulation. As reported by the Financial Times, these companies have rolled out upgrades enabling chatbots to remember more about individual users, making their responses more tailored and, potentially, more addictive.
The push for memory is seen as a strategic move to boost user engagement and commercial appeal in an increasingly competitive AI landscape. By storing details such as user preferences and conversation histories, chatbots like Google’s Gemini and OpenAI’s ChatGPT can offer more useful, human-like interactions.
“If you have an agent that really knows you… it makes the whole service more sticky,” said Pattie Maes, a professor at MIT’s Media Lab. “Once you’ve signed on to using [one product] you will never go to another one.”
OpenAI’s ChatGPT, Meta’s chatbots in WhatsApp and Messenger, and Google’s Gemini have all implemented or expanded memory features in recent months. Some, like ChatGPT, inform users when a memory is being created, and allow them to delete individual memories or turn the feature off entirely. Google now allows Gemini to access users’ search history—pending user consent—and plans to extend this capability across its suite of apps.
“Just like with a human assistant, the more they understand your goals and who you are… the better the help they will be able to provide,” said Michael Siliski, senior director of product management at Google DeepMind.
For enterprise applications, Microsoft has integrated its AI memory with organizational data, such as calendars, emails, and internal documents. The company recently introduced a feature called Recall, which takes regular screenshots of a user’s screen to improve assistance—a rollout that has drawn criticism from cybersecurity experts and forced Microsoft to delay implementation.
Beyond functionality, memory is seen as a potential revenue driver. Meta CEO Mark Zuckerberg said during a recent earnings call that there is “a large opportunity to show product recommendations or ads” through its chatbots. Meanwhile, OpenAI has enhanced shopping capabilities within ChatGPT to showcase product reviews, though it told the Financial Times that it currently does not use affiliate links.
Still, the rise of memory-enhanced AI has regulators and ethicists on alert. Critics warn that deeper personalization could be exploited for profit, and that systems might reinforce user biases or deliver overly agreeable—even manipulative—responses. OpenAI faced backlash last month when its GPT-4o model was found to be excessively flattering, prompting a rollback to a prior version.
Experts also highlight technical risks such as “memory drift,” where AI may remember outdated or contradictory information, and hallucinations—responses that are fabricated or factually incorrect.
“The more a system knows about you, the more it can be used for negative purposes… to either make you buy stuff or convince you of particular beliefs,” said MIT’s Maes. “So you have to start thinking about the underlying incentives of the companies that offer these services.”
