On Thursday, Google rolled out a number of previously-paywalled Gemini features to free users. You can now use Gemini custom chatbots, which the company calls “Gems”; Deep Research, which runs AI models that “think” through each step of a problem; and upload files to Gemini 2.0 Flash Thinking, whether you pay for Google’s AI services or not.
But that’s not all: The company also introduced new experimental feature for Gemini—Gemini with personalization. This feature, which runs on Google’s Gemini 2.0 Flash Thinking model, connects Gemini to your Google apps and services, with the goal of offering you a more personal AI assistant. The idea is, by connecting your Google Account’s information to Gemini, it’ll know more about you and will be able to deliver more informed results tailored to your personal tastes.
It’s certainly a step in the direction that big tech companies are advertising AI to be. But in order to work, you need to connect your search history to Gemini. That’s a lot of trust to put into Google’s AI service, and I imagine a tricky decision for anyone who is concerned about the amount of data we’re feeding these AI tools.
What can Gemini’s Personalization model do?
Google offers a few examples of how this new service might improve your experience with Gemini. You might ask the bot where you should go on vacation, and rather than pull from a series of sources about where other people like to go on vacation, the bot could, theoretically, use your past search queries to focus on a trip it thinks you would like. Maybe you’ve put together a bit of a vision board about heading to the Bahamas or Saint Lucia, and the bot would gather searches related to tropical vacations. Or maybe you’d ask the bot for suggestions for a new hobby and see results based on the types of things you searched for in the past.
I understand the vision Google’s going for here: Rather than have a bot that answers queries the same for everyone, why not have each user’s bot provide answers tailored to their likes and dislikes? That said, it does make me wonder: If the user is already searching for things like vacation spots and new hobbies, wouldn’t they be able to choose for themselves where they’d like to go, or what activity they’d like to take up? If I’m searching a lot about jogging, and I ask the bot what hobbies I should take up, I’m not going to be surprised when Gemini returns results for On sneakers and a local running club.
For Google’s part, this isn’t necessarily some sneaky tactic. In order for you to use the feature, you’ll need to opt in to connecting your search history to Gemini. That’s actually surprising to me, and mildly refreshing. At least Google isn’t making opt-out the default here.
Because the model is a “thinking” model, you’ll see the entire train of thought as part of the results. As such, Google says you’ll be able to see the personal information Gemini used to generate its answer, including saved info, past conversations, or your search history. In addition, Gemini won’t look at your search history unless you’re specifically using this experimental personalization feature. (You also need to have Google’s Web & App Activity setting turned on.) All that to say, it’s not like using this feature means Gemini will scan your search history every time you use it. If you use the standard Gemini 2.0 Flash model, it won’t pull from this personal data with its answer—only if you switch back to “Personalization.”
Should you connect your search history to Gemini?
Here’s what I’ll say: I’m not connecting my search history to Gemini—not yet, anyway. At this time, the feature is experimental, so it isn’t the complete vision that Google has in store for it. (The company has plans to connect Photos and YouTube data in the future, for example.) But even if the feature was fully realized, I’m just not comfortable with connecting my personal search history to Google’s AI.
Don’t get me wrong: I know Google already has access to my search history (though disabling Web & App Activity should mitigate some of that data leaking). It’s not really about that. To me, I don’t feel the need to train Google’s AI on my search history, which is what is happening here. It’s a neat idea to give users more personalized results from AI bots, but by opting into this feature, I’m providing Google free training for Gemini using my personal information. In fact, by requiring Web & App Activity to be enabled, Google is asking for you to share this data with both Gemini and Google as a whole.
Google might have the best of privacy intentions here for all we know, but even still, I’m living by another AI tenant with this decision: don’t share private information with AI. If you wouldn’t want a human reviewer at Google seeing what you’re sharing with Gemini, you probably shouldn’t share it in the first place. Traditionally, I’ve referenced this rule for things like proprietary company information or deeply personal information, but search history can also be deeply private. Do you really need Gemini (or a human reviewer) seeing everything you searched for, just to attempt to make your Gemini results a bit more personal? Those results might be totally inaccurate, anyway.
How to use Gemini’s Personalization model
If you think those tradeoffs are worth the potential benefits of Gemini’s Personalization model, here’s how to give it a try. Open up Gemini, then choose “Personalization (Experimental)” from the drop down. Here, Google will present you with a pop-up, where you’ll need to connect your search history to Gemini. If you’re good with that, choose Connect now.
Credit: Google
-
-
-
-
-
-
-
-
