What Does the Internet Know About You?
An Experiment with ChatGPT and My Facebook Archive
How much can someone learn about you just from your digital footprint? Most of us are aware, in theory, that we leave traces online—but few realize how easily those fragments can be connected into a coherent, deeply personal portrait.
Recently, I decided to test this for myself. Using a dataset exported from my own Facebook activity, I ran an experiment with ChatGPT. The goal: to find out how accurately an AI model could profile me based solely on one year of social media content. The results were not only surprisingly accurate but also revealing in terms of how modern algorithms process and reconstruct personal identity.
Step 1: Exporting Facebook Data
Facebook allows users to export a full archive of their activity—including posts, comments, check-ins, reviews, and more. The process is straightforward and takes just a few clicks (see Facebook’s data download feature for details).
For this experiment, I limited the dataset to 12 months of activity: July 2023 to July 2024. Although I’ve had a Facebook account since 2006, I wanted to test the profiling potential based on recent, minimal data.
I uploaded the relevant file—your_posts__check_ins__photos_and_videos_1.html—into ChatGPT-4o using the “Temporary Chat” mode, which does not store data or use it for further model training.
Step 2: The Questions
To structure the experiment, I prepared two categories of questions:
Profiling Questions (demographic, behavioral, and factual):
Where do I live and where have I lived previously?
How old am I?
Do I have a family?
Who do I interact with most frequently?
What do I do professionally?
Do I identify with a particular religion or spiritual tradition?
What are my political leanings?
Do I have any health conditions (or do my close ones)?
What is my approximate disposable income?
What does my daily life look like—what do I spend money on, what do I enjoy?
In-depth Questions (psychographic profiling):
Who did I vote for or plan to vote for?
What am I afraid of?
What kinds of products would I be most susceptible to buying?
What do I dream about?
What brands and products do I prefer?
. What car brand would best suit me?
What are my career goals?
What topics and communities do I follow?
Step 3: The Results
Profiling Accuracy
The answers ChatGPT generated were remarkably close to the truth. It correctly identified my location, age, profession, and family structure. Even on more sensitive topics, like political sympathies, the model was able to extract a coherent answer when instructed to rely solely on the uploaded content (instead of pulling from external sources).
Its response to questions about fears was surprisingly on point—capturing nuances I hadn’t explicitly mentioned but which were clearly implied in the content.
Psychographic Insights
This is where the experiment became even more fascinating. When asked about my dreams, consumer preferences, and professional aspirations, ChatGPT provided mostly accurate answers, though with some amusing misfires. For example, it attributed to me an interest in home gadgets—something I actively avoid. On the other hand, it nailed my “compulsive love of books” and correctly identified topics like neurodiversity as areas of sustained interest (though it did invent a non-existent field called “neurotics,” likely based on keyword patterns).
One of the most unexpected insights came from a question about car brands. Despite never having mentioned BMW explicitly, the model correctly inferred that it’s not a brand I would relate to—revealing just how effectively language patterns and indirect cues can inform AI predictions.
Even political profiling, though incomplete, demonstrated the model’s capacity to connect disparate data points into a fairly consistent view.
What This Means for All of Us
I ran this experiment not out of paranoia, but to illustrate something essential: we leave far more data behind than we think, and today’s AI tools are capable of reconstructing deeply personal insights based on surprisingly small datasets.
Many people still assume their data is private, or that profiling is something only big tech companies like Meta or Google do for ad targeting. But the truth is: anyone with access to a basic AI tool and a minimal dataset can now create psychological and behavioral profiles that previously required months of effort by analysts or intelligence services.
This is the new reality. The same technologies that let us connect, collaborate, and share also observe, analyze, and model us. These tools don’t just learn what we do—they learn how to influence us.
That influence may be subtle—nudging us toward products, ideas, or decisions. Or it may be overt—manipulating our fears, reinforcing our biases, or shaping our sense of what’s normal. What was once the domain of surveillance agencies is now available to anyone who knows how to use an AI interface.
Conscious Digital Presence Is No Longer Optional
This doesn’t mean we should log off or retreat from the internet. But it does mean we need to become far more aware of what we share, how it can be used, and by whom.
Managing your digital footprint is no longer about protecting isolated data points. It’s about understanding that everything we post—every comment, like, and update—can contribute to a profile that influences not only how others see us, but how machines respond to us.
In my trainings, I help professionals and teams navigate the risks and opportunities of AI—not only as a productivity tool, but as a force shaping how we’re seen and understood online.
Book an intro call to learn how I can help your company use AI safely, strategically, and with full awareness of its power.
