In January 2026, Google rolled out a feature called Personal Intelligence inside AI Mode. For the first time, Gemini can securely connect to your Gmail and your Google Photos library — referencing your travel bookings, personal emails, transaction histories, and photo memories to generate more “personalised” search answers. By March, it was free for every user in the United States.
Google calls it convenience. Critics call it something closer to the most significant expansion of personal-data access in the company’s recent history. Both descriptions are accurate, which is exactly what makes this worth understanding before you tap “connect.”
This piece walks through what Personal Intelligence actually does, why “opt-in” doesn’t fully resolve the structural concern underneath it, and what it means to keep your most personal archive somewhere that was never built to feed a search engine in the first place.
What Personal Intelligence Actually Does
Personal Intelligence is a setting inside Google’s AI Mode that links your Gmail and Google Photos to Gemini. Once connected, AI Mode can pull from both — surfacing a hotel confirmation buried in your inbox, recalling a restaurant from a photo you took eight months ago, or stitching together a trip itinerary from a mix of emails and images without you typing any of it out.
On paper, that’s a genuinely useful capability. Search has always been about retrieving what you already have, faster. Personal Intelligence extends that idea from the open web into the parts of your digital life that used to sit in separate, siloed apps.
The rollout matters because of scale. Gmail and Google Photos are two of the most widely used personal-data repositories on the planet. Connecting either one to a generative AI system — let alone both, simultaneously — is not a small product update. It’s an infrastructural shift in how much of your personal life sits inside a single reasoning loop.
”Opt-In” Is the Right Word — But Not the Whole Story
To Google’s credit, Personal Intelligence is not switched on by default. You have to go into Search personalisation settings, find Connected Content Apps, and explicitly choose to link Gmail, Photos, or both. You can connect one without the other. You can disconnect at any time. That’s a meaningfully better starting position than features that activate quietly in the background.
But “opt-in” answers the question of whether you’ve agreed — not the question of what you’ve agreed to over time. A few things are worth sitting with:
- The boundary of “connected” is not fixed. Today, Personal Intelligence reaches into Gmail and Photos. Nothing about the architecture suggests those are permanent limits — they’re simply where the rollout started. Once a feature like this proves popular, the natural next step is to expand its reach into Calendar, Drive, Maps timeline, and beyond.
- Your photos stop being just photos. The moment your library is connected, each image becomes a potential input to a reasoning system that also has access to your email, your purchase history, and your search behaviour. The value of any single photo to that system isn’t what it shows — it’s what it can be cross-referenced against.
- Google has been candid that the lines are blurry. Reporting on the rollout noted that Google’s policies don’t clearly establish whether content from connected services is fully excluded from broader AI training pipelines, or only the direct Photos and Gmail products. For most users, that ambiguity is the entire ballgame — because once content has been used to shape a model’s outputs, there is no meaningful way to “disconnect” it after the fact.
None of this means Personal Intelligence is a trap. It means the decision to connect your photo library to an AI reasoning system is bigger than the toggle makes it feel.
The Quiet-Expansion Pattern Isn’t New
If this rollout pattern feels familiar, it should. In 2021, Apple announced a system to scan photos on-device for known abusive material before they were uploaded to iCloud — a proposal with a defensible stated goal that nonetheless triggered intense backlash from cryptographers and civil liberties groups. The objection wasn’t about the goal. It was about the precedent: once a company builds infrastructure capable of inspecting your private content at scale, that infrastructure doesn’t disappear just because the original use case gets shelved.
Personal Intelligence follows a similar shape, just from a different angle. It starts narrow, opt-in, and clearly scoped. The capability it establishes — a generative AI system with standing access to your most personal archives — is the part that persists and tends to grow. That’s not a conspiracy theory. It’s simply how product surfaces evolve once users get comfortable with them.
The right question isn’t “is this safe today?” It’s “what does this look like in three years, once the convenience has become a habit and the scope has quietly widened?”
Your Photo Library Isn’t Like Your Inbox
Email is mostly transactional: receipts, confirmations, conversations with a clear beginning and end. Your photo library is something else entirely. It’s closer to an external memory of your actual life — your children’s faces over the years, the apartment you lived in before things changed, the trip you took with someone you’ve since lost touch with, the documents and screenshots you saved because they mattered in a moment you don’t want to forget.
That distinction matters because of how value compounds. A photo from five years ago doesn’t just show what happened — it can now be cross-referenced against everything an AI system has learned about you since. As inference techniques improve, older, more “harmless” images become more revealing, not less. The data you handed over casually in 2024 means something different to a more capable model in 2027.
This is also why “they already know everything about me” is the wrong way to think about it. Each new connection doesn’t just add a data point — it multiplies what can be inferred from every data point that came before it. Context compounds. It does not reset.
What a Privacy-Respecting Alternative Actually Looks Like
None of this is an argument that AI-assisted memory tools are bad. Search, transcription, and intelligent organisation are genuinely useful — daftei builds and ships these features too. The difference is who the AI works for, and what infrastructure it’s connected to.
A privacy-respecting approach to personal AI features looks like this:
- The AI serves you, not an advertising or search graph. There’s a meaningful difference between “search my files for the receipt from March” and “feed my files into a system that also profiles my email and browsing to build a more complete picture of me.”
- Your content does not train someone else’s model. If a provider uses your photos, voice notes, or documents to improve systems that other companies or other users benefit from, that’s a one-way transaction — and you’re the one paying for it with your privacy.
- Encryption is real, not a slogan. Data should be encrypted in transit and at rest using current standards — not vague “we take security seriously” language that can’t be verified.
- The business model doesn’t depend on knowing more about you. If a company’s revenue comes from your subscription rather than from what can be inferred about you, its incentives point in your direction by default.
- You can leave, cleanly, whenever you want. Exporting your archive and deleting your account should be straightforward — not a multi-hour ordeal designed to keep you locked in.
How daftei Approaches This Differently
daftei exists because the current model — where “free” storage means your content becomes raw material for advertising or model training — has a structural flaw. The product being offered and the product being sold are not the same thing, and most people never get to see the difference clearly.
Here’s what that looks like in practice:
Your data is encrypted, not just “protected.” Everything you upload to daftei is encrypted in transit using TLS 1.3 and at rest using AES-256 — the same standards applied to sensitive institutional data. This is server-side encryption, clearly stated as such: daftei can operate on your content to power features like search, but that operation happens within a system designed around your interests, not an advertiser’s.
Your content never trains someone else’s AI. daftei does not sell your data, does not run advertising, and does not use your photos, voice notes, or documents to train third-party AI systems. The AI features inside daftei — search, organisation, memory assistance — work on your behalf. They don’t double as a pipeline into a larger reasoning system tied to search and ads.
The business model is the subscription, full stop. daftei offers 5 GB free, with unlimited storage on Pro plans starting at $5.99/month (or ₹249/month in India), with annual and lifetime options. When you pay for storage, that’s the entire transaction. There’s no second, hidden exchange where your memories are the actual product.
Compliance is the floor, not the achievement. daftei is GDPR and CCPA compliant, available in English, Hindi, and Mandarin Chinese, and works across iOS, Android, and the web. You can export your full archive at any time. If you choose to delete your account, you get a 30-day grace window to change your mind — after which deletion is permanent and irreversible. No ambiguity, no “contact support to really delete this.”
The Question Worth Asking Before You Connect Anything
Personal Intelligence isn’t an isolated decision — it’s part of a much larger pattern in which AI systems are quietly being granted standing access to the most personal corners of our digital lives, one convenient feature at a time. Each individual connection feels small and reversible. The cumulative effect is neither.
Before connecting your photo library to any AI system — Google’s or anyone else’s — it’s worth asking a simpler question first: where do the memories that matter most actually belong? Not which service is most convenient today, but which one has no structural reason to look at your life more closely than you’ve asked it to.
That’s the test daftei is built to pass. Your memories stay yours — searchable, organised, accessible whenever you need them — without becoming an input to someone else’s bigger machine.