How Big Tech Uses AI to Fake Ad Performance

marketing

Let's dig into the truth behind marketing campaign performance and the illusion of data driven descision making in digital advertising.

Without realizing it, you’ve been feeding the most profitable scam in history, and no, it’s not that one weird trick your doctor doesn’t want you to know about. It’s the one weird trick Google, Meta, and TikTok don’t want you to understand. A system so effective it props up hundreds of billions in ad revenue every year.

How You're Part of the Scam

Ok, so when a business wants to advertise on Google, Meta, or TikTok, the first thing they do is slap a little piece of code — a pixel — onto their website. Think of it like a digital homing beacon that beams all your activity straight back to the ad platform.

Let’s make this real. Say you need new headphones. You visit some online store, scroll around, find a pair you like, toss them in your cart... but then you check your bank account and realize, oh right, payday isn’t until Monday. So you bounce.

The next day, your mom's coming over for dinner. You want to impress her with homemade lasagna, but for whatever reason, you forgot ChatGPT exists, so you Google a recipe. You end up on one of those janky recipe blogs plastered with ads. You’re scrolling, trying to get past all the origin stories of Nonna Maria’s tomato sauce just to find the damn ingredients. Ads keep popping up — banners, sidebars, some video you didn’t ask for — and somewhere in that ad soup is one for the headphones you were eyeing.

You don’t even notice it. You’re just there for the recipe.

Fast forward: you nail the lasagna, mom’s happy, and so are you. Payday hits, you head back to the headphone site, and you buy those headphones.

Meanwhile, the marketing team at the headphone company logs into their Google Ads dashboard. They see a conversion. Nice! Their ads must be crushing it. Promotions all around. Somebody tell the CEO.

But here’s the kicker: they’ve been scammed. Google didn’t convince you to buy those headphones. Google just made sure an ad for them showed up somewhere between the lasagna story and the comments about bechamel sauce, so they could claim the win.

That’s the game. The platforms aren’t persuading anyone. They’re predicting who’s likely to buy and making sure an ad gets in front of them just in time to take credit.mAnd the scheme works beautifully because what marketing manager doesn’t want to believe they’re a genius? That their campaign moved the needle. That they deserve that raise. So it keeps spinning. Money flows in. Ad platforms keep getting richer. Marketers keep patting themselves on the back.

But now that you know, let’s dig deeper. I’ll show you exactly how these models work and how platforms like Google, Meta, and TikTok run this play. Time to get technical.

How the Scam Actually Works

Once platforms like Google or Meta have your data, the real operation begins. They build machine learning models trained on millions of user sessions to guess exactly what you’re about to do next.

It sounds helpful at first: "We’ll show the right ad to the right person at the right time." But most of the time, that’s not what they’re doing. Here’s how it works:

Step 1: Clickstream Data

So, every action you take is logged as a sequence of events and willingly passed on from a business to ad networks like Meta. It might look like this:

user_pseudo_id       event_timestamp       event_name         page_location                       item_name      engagement_time_msec
839a1cfe29d3f8       2024-07-07 12:01:05   scroll             https://store.google.com/home       —              5000
839a1cfe29d3f8       2024-07-07 12:02:18   view_item          https://store.google.com/bidets     Classic 3000   11000
839a1cfe29d3f8       2024-07-07 12:03:12   add_to_cart        https://store.google.com/cart       Classic 3000   3000
839a1cfe29d3f8       2024-07-07 12:04:04   begin_checkout     https://store.google.com/checkout   Classic 3000   8000
839a1cfe29d3f8       2024-07-07 12:04:47   remove_from_cart   https://store.google.com/checkout   Classic 3000   2000

Once they've collected your data, the next step is to transform that chaos into structured signals the computers can learn from. This process is called feature engineering, and the goal is to summarize behavior to help the model spot patterns.

Here’s an example of how the above data might be transfomed for a machine:

user_pseudo_id       product_views   time_on_checkout_sec   cart_abandonment   returned_within_24h   category_engagement_bidets   label_purchase
839a1cfe29d3f8       3               22                     True               Yes                   High                         1

These form the model’s input, while the output is the label: in this case, label_purchase = 1, meaning the user converted. Everything else is just clues leading to that decision.

Step 2: Model Training

With your cleaned up data, they want to predict who will buy next. Every model training pipeline includes a few core steps:

  1. Train/Test Split: The data is split into training and test sets to simulate the real world. The training data teaches the model; the test data checks its predictive power.
  2. Hyperparameter Tuning: Settings like the number of trees or their depth are adjusted to improve performance.
  3. Performance Objectives: Models aren’t just optimized for accuracy but for business goals like ROAS, cost per acquisition, or conversion lift.
  4. Validation and Monitoring: Models are monitored in production and updated regularly because human behavior changes fast.

Step 4: Prediction + Ad Delivery

Once the model is ready, it gets deployed. As you visit a site, your behavior is fed back, turned into features, and passed through the model, producing a real-time probability you’ll convert:

prediction = model.predict_proba(current_features)[1]
# Output: 0.87

This is ad delivery powered by inference, not influence.

Why This Looks Like It Works

On paper, the platform looks amazing: it predicted a buyer, showed an ad, and you purchased. But that’s the illusion. The model predicted you, and the platform jumped in at the last second.

Algorithms perform great on small budgets, targeting users they’re very confident about. But when you scale, performance drops even though the algorithm keeps taking credit. Platforms cherry-pick attribution: tight targeting looks impressive, but scaling exposes its limits.

The Lie Built on Prediction

There’s an old quote in advertising: "Half the money I spend on advertising is wasted—the trouble is, I don’t know which half." Before machine learning, that was true. Now, platforms know exactly who’s likely to buy. They don’t convince anyone; they just predict outcomes, fire an ad, and claim the credit.

We still don’t know which half is wasted, but now it’s the half that might have actually needed convincing. The better these systems get at prediction, the harder it becomes to measure real influence. The algorithm doesn’t change minds; it finds those who already decided. Then, when they buy, it says: "That was me."

If You Actually Want to Know What Works...

Don’t just trust the dashboard. Test reality:

  • Hold back entire geographies and compare.
  • Run ghost ads to fake users and measure the difference.
  • Freeze campaigns and watch what decays.
  • Build uplift models, not just conversion models.

Without real-world tests, you’re not measuring impact. You’re measuring prediction. Every time the algorithm gets it right, you’re just more convinced it worked—even if it didn’t.