Your AI Has Been Taking Notes (And You Didn’t Even Notice)
I’ve been covering AI for over a decade, and I’ve lost count of how many times a company has told me its new feature « puts users in control. » Usually, that means a settings menu buried three layers deep that lets you toggle off a data-sharing option you never agreed to in the first place. So when I saw ZDNET’s headline about Google giving Gemini a Memory summary, my first reaction was a cynical smirk. Another PR-friendly knob to turn, I thought. But then I actually read the details, and I have to admit: this one feels different. Not revolutionary, mind you. But different enough to matter.
Here’s the gist. Google’s Gemini now has a Memory summary feature that lets you see exactly what the AI has remembered about you. That’s it. That’s the headline. But in a world where every chatbot is quietly building a profile of your preferences, your job, your favorite pizza topping, and your deeply held political opinions, being able to peek behind the curtain is a bigger deal than it sounds. You can also edit or delete those memories. One by one. Manually. Like some kind of digital janitor.
Is it perfect? No. Is it a step in the right direction? Absolutely. And frankly, it’s about time.
The Creepy Part Nobody Talks About
Let’s rewind for a second. When ChatGPT first rolled out memory features, I wrote a piece arguing that persistent memory was the most underrated and most unsettling development in consumer AI. Not because I’m a paranoid luddite, but because memory changes the fundamental relationship between you and a machine. A stateless chatbot is a tool. A chatbot that remembers you had a fight with your boss last Tuesday is something else entirely. It’s a confidant. It’s a therapist. It’s a surveillance system that happens to be polite.
Google’s Gemini has been doing this for a while now, quietly accumulating context across your conversations. If you use Gemini through Google Workspace, it’s also pulling from your emails, your docs, your calendar. That’s a staggering amount of personal data flowing into a model that, until recently, you couldn’t easily audit. What struck me here is how long it took for Google to give users a simple window into that memory. The company has been shouting about AI capabilities from the rooftops, but the guardrails have always lagged behind. That’s not unique to Google, of course. It’s an industry-wide pattern. Ship first, explain later, apologize if necessary.
The Memory summary feature is a tacit admission that this pattern is a problem. You don’t build a dashboard for something you think is trivial. Google knows that persistent memory is a trust issue, and trust is the currency that keeps people coming back to your chatbot instead of jumping to a competitor. So yes, this is self-serving. It’s also genuinely useful.
How It Actually Works (No, It’s Not Magic)
The mechanics are straightforward, which I appreciate. You open Gemini, navigate to the settings or the memory section, and there it is: a summary of what the AI thinks it knows about you. You can scroll through it, spot the stuff that’s wrong or outdated, and nuke it. You can also add memories manually if you want to give the AI a head start. Say you’re a vegetarian who hates cilantro. You can just tell Gemini that, and it’ll remember without you having to repeat yourself every single conversation.
What I like about this approach is that it’s transparent without being overwhelming. Google isn’t dumping a raw JSON file of your entire interaction history on you. It’s giving you a curated summary, the highlights of what the model considers relevant. That’s a smart design choice. Most people don’t want to read through a log of every query they’ve ever made. They want to know what the AI thinks it knows, in plain language, and then decide whether that’s okay.
What I don’t like is that this is still a manual process. You have to remember to check your memory summary. You have to remember to prune it. If you’re the kind of person who never clears their browser cache or unsubscribes from email newsletters, you’re probably not going to spend your Sunday afternoon auditing your AI’s memory. That’s a problem, because the whole point of memory is that it operates in the background. Out of sight, out of mind. The feature gives you control, but it doesn’t give you automation. I’d love to see Google add a « review your memory » prompt every few weeks, or a simple notification when the AI adds something new. Until then, this is a tool for the diligent, not the default.
The Privacy Paradox We Keep Ignoring
Here’s where I get a little worked up. The entire premise of memory features is that they make AI more useful. And they do. A chatbot that remembers your preferences is objectively more helpful than one that doesn’t. But the trade-off is that you’re handing over a piece of your identity to a corporation that has a long, complicated history with user data. Google is not a charity. It’s an advertising company that also makes phones and search engines and, yes, AI models. Every memory Gemini stores is another data point in a profile that could, in theory, be used to sell you things or shape what you see.
I’m not saying Google is doing anything nefarious here. The Memory summary feature is a genuine attempt at transparency. But I am saying that we, as users, need to stop pretending that convenience is free. Every time you let Gemini remember something, you’re making a small bet that Google will handle that information responsibly. That bet might pay off. It might not. The point is that you should be making it consciously, not sleepwalking into it because the feature is on by default.
What struck me here is how different this feels from the early days of the metaverse and Web3, when every company was falling over itself to promise a decentralized utopia where users owned their data. That never materialized. Instead, we got centralized platforms with better marketing. The AI memory race is following the same trajectory. The companies that win will be the ones that make memory feel seamless and magical, not the ones that make it feel safe. That’s a depressing reality, but it’s the one we’re in. Features like Memory summary are a small corrective, a reminder that users should have a say. But they’re not a solution to the underlying power imbalance.
What This Means for the Rest of the Industry
Google isn’t the only player in this space. OpenAI’s ChatGPT has had memory features for a while, and Anthropic’s Claude is experimenting with similar ideas. The difference is that Google is building memory into an ecosystem that already includes your email, your calendar, and your documents. That’s a much bigger surface area for potential creepiness. It’s also a much bigger opportunity for genuine utility. Imagine a Gemini that remembers you have a dentist appointment next Thursday and proactively blocks out time on your calendar. That’s not science fiction. That’s the logical endpoint of memory features.
The question is whether users will trust Google enough to let that happen. The Memory summary feature is a step toward earning that trust, but it’s a small step. Google needs to do more. It needs to make memory review a habit, not a chore. It needs to be transparent about how memories are used, not just what they are. And it needs to give users a way to opt out entirely without losing access to the core functionality of the chatbot. Right now, memory is optional, but the pressure to enable it will only grow as the feature becomes more integrated into the product.
I think the smartest thing Google could do is make Memory summary a regular part of the Gemini experience. A quick prompt, a gentle nudge, a « hey, here’s what I remember about you, want to take a look? » That would normalize the idea that memory is something you manage, not something that happens to you. It would also give users a sense of agency that’s often missing from AI products. The technology should serve us, not the other way around.
The Bottom Line (And Why I’m Still Skeptical)
Gemini’s Memory summary is a good feature. It’s not a game-changer, and it’s not going to solve the deep structural problems with AI and privacy. But it’s a practical, useful tool that gives users a modicum of control over something that’s been happening behind the scenes for too long. If you use Gemini regularly, you should check it out. Spend five minutes scrolling through what the AI thinks it knows about you. You might be surprised. You might be annoyed. You might find yourself deleting a memory that’s just plain wrong. All of those reactions are valid.
What I hope is that this feature sparks a broader conversation about what memory means in the age of AI. We’ve spent years talking about model capabilities, benchmark scores, and parameter counts. We’ve spent far less time talking about the human side of the equation: what it feels like to have a machine remember you, and what it means to forget. Memory summary is a small window into that relationship. It’s up to us to look through it.
And if you’re wondering whether I’ve checked my own Gemini memory summary? Of course I have. I deleted a memory where the AI thought I was a fan of a certain football team. I’m not. I’ve never been. And frankly, I’m offended it thought so.
Original source: read the full article