Do you use Android or iOS?
Does anyone you know have your phone number saved on their device?
Do you Google your questions? Use Chrome? Have you ever used Alexa or Siri? A loyalty card? A Facebook, Instagram, or TikTok account? Do you order from Amazon?
If you answered yes to even one of those: Google, Meta, and the rest already know considerably more about you than AI training ever would — even if you use it every single day.
This isn't a dismissal of AI privacy concerns. It's a reality check.
What Big Tech actually knows about you
Let's start with Google.
Google stores — unless you actively turn it off — your complete search history since the day you created your account. Location data, down to the minute, for years. Everything you've watched on YouTube. What you've written in Gmail. Every website you've visited in Chrome.
In 2026, Google agreed to a $135 million settlement over collecting Android cellular data for targeted advertising without proper consent. And the "Privacy Sandbox" project — Google's announced plan to phase out third-party cookies in Chrome — was quietly killed. Cookies live on.
Meta (Facebook, Instagram) knows your social graph, your political leanings, your purchasing habits — not just from your posts, but because Meta is connected to thousands of data brokers. Loyalty card providers sell aggregated purchase profiles to brand manufacturers. Amazon knows whether you hesitate before buying and how often you abandon your cart.
That's individual profiling. Going back years. Specifically about you.
What AI training actually is — and what it isn't
Here's the widespread misunderstanding.
AI training for large language models like ChatGPT or Claude is not a dossier on individuals. It's statistical pattern recognition.
Ten million users in Germany ask about the calories in bread rolls. Ten million users in the US ask about pancakes. The model learns: people in Germany eat bread rolls for breakfast, people in the US eat pancakes. No names. No individuals. Just weights in a neural network — a mathematical model — shifting by tiny fractions.
Your specific conversation is not stored or retrievable. There's no AI that knows you had three bread rolls at 7:30 a.m.
An honest nuance: texts that appear very frequently in training data can sometimes be reproduced verbatim by models. That's an active area of research. But it concerns patterns from massive datasets — not your individual conversation from yesterday.
For the record: since September 2025, Anthropic (Claude) uses conversations from free-tier accounts for training — opt-out available in settings. OpenAI has done this for longer. If you don't want that: turn it off. Or use a local model like Ollama — nothing leaves your device.
The speeding ticket test
Imagine your speeding fine is processed by your neighbor — the one who talks freely after a few drinks. Two days later, the whole neighborhood knows you got caught by the same speed camera three times last week.
Or: an algorithm matches the license plate to a vehicle. Fine issued automatically. Payment checked. No human looks at it. No gossip. No judgment.
Which would you prefer?
This question comes up more often than you'd think — credit decisions, medical records, job applications. The AI doesn't judge you morally. It has no bad days, no biases, no urge to gossip. Depending on context, that's actually an argument in its favor.
Advertising: The AI knows your pattern, not you
Personalized advertising via algorithms divides opinion. But looking at it objectively:
The algorithm wants to sell you something — not evaluate you as a person. It doesn't care about your political views or whether you ate too much yesterday. It knows your purchasing pattern, not your character.
That can be useful. If you're planning to buy a new laptop anyway, a targeted offer at the right moment might save you money instead of paying full price three months later. Or you end up buying three things you don't need because the algorithm knows you well. That happens too.
The point: your spending habits end up in an algorithm — not with someone who knows you and could judge you.
What still holds true
Legitimate AI privacy concerns exist — they just look different from the usual framing.
The better questions: Which provider am I using? EU-based providers fall under GDPR with stricter obligations. What am I typing in? Passwords, real names in sensitive contexts, and trade secrets don't belong in any cloud application — AI or otherwise. Do I need full control? Then a local model is the answer.
More on this: What happens to your AI inputs?
Fear is a poor advisor
AI privacy is a legitimate topic. But the outrage seems oddly one-sided when the same people who eye ChatGPT with suspicion have been typing every question into Google for years, sharing their lives on Instagram, and logging every purchase with a loyalty card.
That doesn't mean: nothing matters. It means: know your priorities. Then decide — with information, not fear of the wrong ghost.
Unsponsored opinion, personal perspective. As of May 2026.