lab-reports

No, Your Phone Isn't Listening. What It's Actually Doing Is Worse.

Think your phone is eavesdropping? It’s worse: precise data tracking. Learn what’s really happening and the 3 settings that actually help.

By Ben Kalkman · 13 min read · privacy ad targeting data tracking microphones smartphones surveillance capitalism online advertising settings and controls

You said it out loud, never typed it, and the ad showed up anyway. Here is what actually happened, and the three settings that do something about it.


I said something out loud in my kitchen that I have never once typed into a phone.

I won't tell you the exact product, because it's embarrassingly specific, but picture the kind of thing you'd only mention to a spouse standing at the counter. No search. No text. Just words in the air of my own house. And the next afternoon, there it was in my feed, the exact thing, staring back at me like it had been waiting.

You know this feeling. Everybody does now. You say "we should look into a hot tub" or "my knee's been bugging me," and within a day, the ads find you, and a little cold spot opens up in your stomach. It sits so far into the background of modern life that we've stopped even arguing about it. Of course, the phone is listening. We've all just decided that's the deal.

Here's what bothers me about that shrug: we're all so sure it's the microphone, and almost nobody can tell you how it works, why it only seems to happen sometimes, or what on earth to do about it. We've accepted a spooky story and skipped straight past the part where we understand it. So I did the thing I do now with anything that nags at me. I pointed AI at it and told it to go find out whether my phone is actually listening to me.

The first answer it gave me was the popular one. The popular one was wrong. And chasing down why it was wrong turned out to be the most useful thing I've learned about my phone in years.

The moment it felt real

Let me give you the exact thing that almost ended my investigation early, because for about ten minutes, it looked like the case was closed.

There's a real company, Cox Media Group, and in 2023, a pitch deck of theirs leaked. The product was called "Active Listening," and the deck said, in plain language, that they could capture "real-time intent data by listening to our conversations" through the microphones in your smart devices, then pair that with your behavior to sell you things. It even listed a price. And on one slide, it named its partners: Google, Amazon, and Facebook.

I'll be honest, when the AI surfaced that, I felt the hair go up on my arms. There it was. Proof. A company saying out loud that they do the exact creepy thing we all suspect, with the biggest names in tech riding shotgun.

This is the part where I had to slow the machine down. Because the confident answer and the correct answer, again, were not the same answer.

Blog image

Then it fell apart

The deck was a sales pitch. Not a confession. Not a product teardown. A slide show built to make advertisers open their wallets.

And the second you look at the "partners," the whole thing deflates. Amazon said it had "never worked with CMG on this program." Meta said it was only a general marketing partner and was "not in this program." Google quietly dropped Cox Media Group from its Partners Program once reporters started asking questions. Cox itself backpedaled, calling the materials outdated.

So the single most viral piece of "evidence" that your phone is listening turned out to be a marketing team bragging about something there's no proof they could actually do, name-dropping clients who said they weren't clients. The scary headline had sprinted about a mile ahead of the technology.

That's when I stopped looking for the confession and started looking for the evidence.

The biggest test anyone has run has found nothing

Here's the study that should be famous and isn't. A team of researchers at Northeastern University got suspicious of the same myth, so they tested 17,260 Android apps, watching exactly what each one did and what it sent off the phone. If apps were secretly switching on your mic to record you, this is where it would show up.

It didn't. Across all 17,260, they found no evidence of apps quietly turning on the microphone to spy on people for ads.

Blog image

But, and this is the part that made me put my ZOA down, they did find something. A number of apps were quietly grabbing screenshots and recordings of your screen and shipping them off to third parties, without asking. So the honest headline was never "nothing to worry about." It was: you were staring at the wrong sensor. Everyone's watching the microphone while something else walks out the door.

I want to be fair about the limits, because that's the whole point of doing this right. A study of 17,000 apps is enormous, but it isn't every app that has ever existed. It can't prove that the microphone thing never happens anywhere. What it can do, and did, is show that across the apps almost all of us actually use, the eavesdropping we're all certain about simply does not show up. When you go looking with real instruments, the ghost isn't there.

So if it isn't the mic, how did that ad find me?

The real machine has no wire to cut

Here's the reframe that changed how I see the phone in my pocket. The reason the microphone theory is so popular is that it's comforting. A microphone is one wire. You can imagine taping over it, unplugging it, catching it in the act. It makes the problem small enough to hold.

The real answer is that they don't need the microphone, because they already have almost everything else.

Start with the data brokers, the companies most people have never heard of, whose entire business is knowing you. Back in 2014, the Federal Trade Commission pulled the curtain back on them and found brokers holding billions of data points on nearly every American. One single broker was sitting on more than 1.4 billion consumer transactions and over 700 billion aggregated data elements. That was a decade ago. The industry has only gotten bigger since, so read those as a floor, not a ceiling.

Then there's your location and your device's advertising ID, a little identifier every phone carries. Last December, the FTC took action against a broker called Mobilewalla, alleging it had collected more than 500 million of those advertising IDs paired with precise location data. And here's the detail that genuinely rattled me: the FTC says Mobilewalla scooped up people's data from the behind-the-scenes ad auctions that decide which ad you see, even when it lost the auction. Sit with that. You can get profiled off an ad you were never even shown.

That advertising ID is the master key. It's what lets a dozen companies, each with a slice of you, quietly snap their files together into one picture. Privacy researchers at the EFF will tell you there's a whole industry built to tie that "anonymous" ID back to your actual name, email, and phone number. "Anonymous" is doing a lot of lying in that sentence.

And this isn't a fringe claim from privacy activists anymore. In September 2024 the FTC studied nine of the biggest platforms we all use, Amazon, Meta, YouTube, TikTok, Snap, and more, and described what they found as vast surveillance: companies collecting far more than we'd ever expect, buying still more from brokers, gathering data even on people who never used their services, keeping it more or less forever, and feeding all of it into algorithms and AI with little or no way for you to opt out. The EFF put it in the one line that ties this whole thing together: this isn't a bug or an accident. It's the business model working exactly as designed.

This is the part that matters now more than ever

Here's where it stops being a fun myth-busting exercise and starts being the thing I actually want you to walk away with.

All of that data, the location trails, the purchases, the app activity, the identifiers, is the fuel. Artificial intelligence is the engine that turns it into something that feels like mind-reading. A model doesn't need to hear you say "hot tub." It has watched where you drove, what you bought, who you're near, and what you paused on, and it can guess you're a hot-tub person before the thought fully forms in your own head. Audio would be a noisy, battery-draining, legally radioactive downgrade from what prediction already pulls off for free.

Blog image

And every one of those engines is being fed from the device in the palm of your hand. That's the shift worth sitting up for. We used to worry about what a company could record. The real question now is what a machine can predict, and we are handing it the training data every hour, mostly without noticing. Understanding that trade is not paranoia. In an AI world, it's basic literacy about the most personal tool you own.

Why it still feels like the phone is listening

So why does the "it heard me" feeling refuse to die, even after all this?

There's a name for it, the frequency illusion, sometimes called the Baader-Meinhof phenomenon. You notice a thing, and suddenly it's everywhere. The world didn't change. Your attention did. Buy a certain car, and you'll swear every third vehicle on the road is now that car.

Now stack that on top of ads that were already predictive. You mention the hot tub because something in your real life nudged you toward it, a friend, the season, a store you drove past with a location on. The ad system, running on those same signals, serves you the hot-tub ad. Your brain lines up "I said it" next to "then I saw it" and draws the only line that feels obvious. But you and the algorithm weren't connected by a microphone. You were both reacting to the same moment in your actual life. It just got there a half-step ahead of you.

What you can actually do about it tonight

This is the part I care about most, because knowing how it works is useless if you don't know which lever moves anything.

First, the bad news you need to hear: taping over your microphone does nothing. You'd be barricading the one door they mostly aren't using while the real ones stay wide open. If you take one thing from this whole piece, let it be that you stop defending the wrong door.

Here's what actually moves the needle.

Reset and rein in your advertising ID. This is the master key we talked about. On an iPhone, you can turn off app tracking and Apple's own ad targeting in your privacy settings. On Android, you can delete or reset your advertising ID outright. It's a couple of taps, and it kicks a leg out from under the whole profile-stitching operation. The EFF has a plain-language walkthrough worth keeping open while you do it.

Cut off location for apps that have no business having it. Your weather app needs a rough location. Your flashlight and your solitaire game do not. Go through your app permissions and set location to "never" or "while using" for anything that doesn't genuinely need it. Location is one of the richest signals in that whole tangle. Starve it.

Know that you can tell the brokers to let go. This is the one almost nobody uses. You can opt out of many data brokers directly, and tools like the Global Privacy Control let you signal "don't sell my data" automatically as you browse. If you're in a state with a data-deletion law, like California's, you have even more teeth. It's tedious, I won't pretend otherwise, but it's real, and it's yours to use.

Blog image

None of this makes you invisible. That's not the goal, and anyone promising it is selling something. The goal is to stop being the easiest profile in the room and to make the trade on purpose instead of having it made for you while you slept.

What I actually want you to walk away with

I have a house full of phones. Eleven of us under one roof means a small army of devices, open microphones in every room, screens in more hands than I can count. For a long time, the thought of all that made me vaguely uneasy in a way I couldn't name, so I did what most people do. Nothing.

Here's what changed for me, and it's oddly a relief. It was never magic. It's math. And math has settings.

The reason I wanted you to see behind the curtain isn't to scare you off your phone. I'm not throwing mine out, and neither are you. It's that a fear you understand becomes a choice you can make. The people who shrug and say "the phone's just listening, nothing I can do" have handed over the one thing they didn't have to: the decision. I'd rather my kids grow up knowing there's a curtain at all, and knowing their own hand is on it.

So go reset that advertising ID tonight. Turn off the location permissions that don't earn their keep. It'll take you five minutes, and you'll walk away having actually done something instead of just feeling watched. Somebody reading this needed to know those switches existed. If that's you, that's the whole reason I wrote it down.

One more thing, because it's the next question

The uncomfortable part isn't what your phone can hear. It's what the machine can already guess. We spent years worried about being recorded, and quietly walked into an era of being predicted, feeding the engines a little more of ourselves every hour without a second thought.

That's the thread I'm pulling next: not what AI hears, but what it can see coming. And it starts, for me, with a dumb little ad for a thing I only ever said out loud.


Come build alongside me. I document these rabbit holes as I go, what the AI got right, where it confidently steered me wrong, and what I'm still figuring out. If that's useful to you, join the lab. Someone out there needs the next one, too. And if you're a business owner wondering what all this prediction machinery means for how customers find you, that's exactly the kind of question I dig into over at Digital Ignitor and Rocket Media.


Where these numbers come from: Northeastern "Panoptispy" study of 17,260 apps, PoPETs 2018 · 404 Media on the Cox Media Group "Active Listening" deck · Gizmodo on the CMG deck and denials · FTC on the data-broker industry, 2014 · FTC action against Mobilewalla, 2024 · FTC "vast surveillance" report on nine platforms, 2024 · EFF on disabling ad-ID tracking · ProPublica on what data brokers know · Psychology Today on the frequency illusion. A note on honesty: several broker-scale figures date to 2014 and understate today's industry, and the Mobilewalla figures are FTC allegations. The point stands either way.