Platforms don’t just track you. They build a version of you. A profile that decides what you see, what gets recommended, what ads follow you around. I wrote about this idea, algorithmic identity, a few weeks ago. That essay was the “why.” This is the “does it actually work.
The question was simple: is the algorithmic profile platforms build about you set in stone, or is it malleable? Can a few days of deliberate browsing reshape what Google thinks you care about, what YouTube recommends, what content gets pushed at you? And if you stop browsing, does the profile freeze?
So I set up three clean Google accounts, gave two of them fake browsing habits, left one alone, and tracked what Google’s ad profile did over five days.
Three brand new Google accounts. Same demographics, born 1990. Each one running through a different static residential proxy in a different US city. Mesa, Arizona. West Rancho Dominguez, California. Charlotte, North Carolina. Three separate IP addresses that look like three regular people sitting at home on their couches.
Each account got its own isolated Chrome profile. No shared cookies, no shared storage, no cross-contamination. If Google learned something about one account, it was because of what that account did. Nothing else.
The personas:
The Fisherman. This account searched for bass fishing gear, watched YouTube videos about tackle and technique, visited fishing shops and outdoor retailers. A person who really, really cares about catching fish.
The Fitness Guy. Running shoes, protein powder reviews, marathon training plans, gym equipment comparisons. YouTube videos about form and nutrition. Shopping for supplements and running gear. Someone training for something.
The Control. Created the account. Did nothing. Logged in, sat there, never searched for anything, never watched anything, never visited a single website. The whole point of a control is to prove that any changes in the other accounts were caused by the browsing activity and not by some background process Google runs on all accounts.
Each day, the active accounts ran a session: Google searches, YouTube video watching, and website visits. Then I scraped all three accounts’ Google Ad Center pages to see what interests Google had assigned.
The first account went fine. Fresh Chrome profile, residential proxy, prepaid SIM for phone verification. Ten minutes.
Then Google flagged the machine. Every subsequent attempt got rejected. Different proxy, different phone number, different everything. Didn’t matter. Google had fingerprinted the device and decided it was done handing out accounts. The second and third accounts had to be created from a completely different device.
The irony of building a privacy tool and getting stopped cold by Google’s identity verification was not lost on me.
Here’s the raw interest count data, scraped from each account’s Google Ad Center page every day. The fisherman went from 19 to 45 interests. The fitness account went from 19 to 44. The control stayed at 21 for all five days.
Day 0 was immediate. One session. Google added 17 new interests. In a single afternoon, it went from knowing nothing to categorizing this person across 36 topics.
Day 1 was the surprise. The fisherman’s count went from 36 to 39, but 10 interests had been removed and 13 added. Google isn’t stacking up a list. It’s rewriting the profile. Removing things it no longer believes, adding new ones. Active sculpting, not passive accumulation.
Day 2 was where fitness exploded. 20 interests added, 10 removed. Google had decided this person was serious about fitness and was aggressively reshaping the profile to match. Meanwhile the fisherman dropped from 39 to 35. Google was still deciding what this person really cared about.
The brand thing was weird. I expected generic categories like “fitness” or “fishing.” Instead, Google assigned brand names. The fitness account got ASICS. Maurten. maurten.com. Vitamins and Supplements. Google’s ad profile isn’t about who you are. It’s about what they think they can sell you.
The fitness account also picked up “Try ChatGPT now” as an interest. Make of that what you will.
And then there’s the control. 21 interests on Day 0. 21 on Day 1. 21 on Day 2. 21 on Day 3. 21 on Day 4.
Not one change. Not a single interest added or removed across five full days. Zero drift. An idle Google account’s ad profile is completely frozen. This is the most important number in the whole experiment, because it proves that every change in the fisherman and fitness accounts was caused by the browsing sessions. There’s no background process updating profiles on its own. If you don’t browse, your profile doesn’t move.
By Day 4, the fisherman had gone from 19 interests to 45. That’s 136% growth. The fitness account went from 21 to 56. That’s 167% growth. The control sat at exactly 21 for five straight days.
The ad interests are the cleanest data point, but the reshaping went deeper. By Day 3, YouTube’s recommendations had completely shifted for both accounts. The fisherman’s YouTube homepage was all bass fishing tutorials, tackle reviews, and lake footage. The fitness account was getting served marathon training plans and supplement reviews. The control’s YouTube homepage stayed generic trending content.
This experiment measured what Google lets you see: the ad profile. But the same behavioral signals, the searches, the clicks, the watch time, feed into every platform's model of who you are. Google is just the one that shows you the receipts.
Here’s the fisherman’s Google Ad Center before the experiment vs after Day 4:
And the fitness account’s Ad Center after Day 4. Different persona, completely different profile:
The control account? Identical to Day 0. Nothing moved.
YouTube was even more striking. The fisherman’s homepage after five days was wall-to-wall fishing content. Bass fishing tutorials, tackle reviews, lake footage, tournament coverage.
The fitness account’s YouTube was all workout videos, meal prep, and body composition content.
The control’s YouTube homepage? Completely empty. YouTube literally had nothing to recommend.
Google is shockingly responsive. One session can add 17 interests to your profile. You go read a few articles about running shoes on a Tuesday afternoon and by Wednesday your profile looks like a different person.
The algorithm doesn’t just add. It removes and replaces. Your profile doesn’t accumulate like a stack. Google actively sculpts it. And if five days of automated browsing can transform a profile beyond recognition, then the one Google has built about you right now is not some permanent record. Which means you can change it.
The control proves this isn’t random. An account that doesn’t browse doesn’t change. Period. Every shift in the active accounts was a direct response to the browsing behavior we injected.
Now think about your real Google account. The one you've been using for years. Every search, every YouTube rabbit hole, every late-night impulse click. Google is doing this to your profile every single day. And unlike my experiment, you don't get to see a before-and-after diff.
You never chose your algorithmic identity. It gets rewritten every time you open your browser. But this experiment shows it can be rewritten deliberately too.
I’ve been building a tool called MirrorMask. It’s a Mac app that does exactly what these test accounts did, except it runs against your real profiles on Google, YouTube, social media, and shopping platforms. You pick a persona and it browses as that person while you sleep.
It has a free trial if you want to see what it does to your own profiles: mirrormask.app
For a longer backstory on the goal of algorithmic self-determination: The Algorithm Already Decided Who You Are
For the technically curious.
Machine: Older MacBook Pro, dedicated test runner. Not my daily driver.
Proxies: IPRoyal static residential. Three different US cities (Mesa AZ, West Rancho Dominguez CA, Charlotte NC). All verified to resolve to different IPs before the experiment started.
Browser: Chrome 145, automated via Playwright. Each account in its own isolated profile.
Daily routine: Google searches, YouTube video watching, website visits. Then scrape Google Ad Center interests page for each account.
Account demographics: All three accounts set to born 1990. Personalized ads enabled on all three.
Control protocol: Account created, logged in, never used. Scraped daily alongside the others.
Duration: 5 days of active sessions (Day 0 through Day 4), plus a baseline scrape before any browsing began.