Written by Brook Schaaf

Why is it that walled gardens’ advertising revenue keeps going up when engagement keeps going down, at least for Facebook? On the most recent Retail Geek podcast, co-host Scot Wingo noted that “the artist formerly known as Facebook just absolutely crushed it at $58 billion” in fourth-quarter earnings, adding that “they’re getting much better at targeting with AI.”

Setting aside WhatsApp and Threads, Facebook and Instagram remain massive, with billions of users. But engagement has long been declining, with Mark Zuckerberg himself acknowledging that time spent has “gone down meaningfully.” I can observe as much in my personal feed.

For years now, most industry pundits have cited enhanced targeting as the explanation for higher ad yields, but I’ve always found this unsatisfying.

While the ads I see in walled gardens are usually well targeted to me, I rarely, if ever, click on them. Do they influence later purchase decisions? Yes. An Instagram ad did for my wife, who recently bought an overpriced book on mindfulness for our kids that she would almost certainly never have known existed were it not for the ad. So yes, targeted advertising works. I’m a believer. (I’d be in the wrong profession if I weren’t.)

But does incremental improvement in AI targeting plausibly explain tens of billions of dollars in additional annual revenue in an increasingly competitive environment? There are other explanations commonly offered:

  • The price per impression rises faster than engagement falls. Maybe, but what causes CPMs to rise in the first place?
  • Engagement is declining, but only among the least valuable users. Maybe, but this seems unlikely because the most valuable users tend to have the most competing demands on their time and attention.
  • Digital advertising continues to absorb offline spend. Clearly true, but only up to a point of diminishing returns. Wouldn’t we have reached that point years ago, especially with more competition from platforms like TikTok?

This brings us back to the standard explanation: performance advertising keeps performing better, so more dollars flow to it. Even if walled gardens over-attribute conversions (as they surely do), the associated sales and leads are real. So the targeting must genuinely be getting better… right? There seems to be no other explanation.

Except maybe there is.

What if AI isn’t primarily getting better at persuading buyers, but at predicting them?

You may remember the story from Charles Duhigg’s “How Companies Learn Your Secrets” about the predictive model Target built to identify customers likely to be pregnant. The signals included purchase patterns involving unscented lotions and soaps, along with certain vitamin combinations. When maternity coupons began arriving at a household, a father complained to a local store manager, who apologized. The punchline came later: the father called back to apologize after learning his daughter was, in fact, pregnant.

Now imagine those coupons were sent not by Target, but by a third party that received credit for any subsequent purchases. If that third party had access to Target’s data plus much more, something closer to a family’s entire digital life, wouldn’t it be exceptionally well positioned to selectively send coupons… or serve ad impressions?

In that case, the system isn’t creating buyers so much as identifying them early and inserting itself into the transaction path.

This dynamic is sometimes described as “predictive targeting capture,” “reverse causality advertising,” or a “tax on demand discovery.” Yet these ideas are rarely discussed explicitly in the hundreds of articles I read each week or the podcasts I listen to. They surfaced for me only after I queried ChatGPT directly about the concept.

So perhaps this is best understood as a conspiracy of silence. The platform has no incentive to raise the issue. Advertisers lack a true counterfactual, fear revenue loss, and see only platform-supplied reporting that largely assumes causality. Commenters and pundits in the space who are in the know likely benefit from the status quo in one way or another.

Moreover, this effect is far subtler and harder to demonstrate than traditional attribution problems. There’s no obvious fraud, no villain, and no dramatic reveal. The software is just doing its job.

This matters for affiliate marketing because, if platforms are increasingly monetizing foreknowledge rather than influencing prospective customers, then affiliates are competing for dollars not just with better ads, but with systems that already know who will buy. That sets a very different bar in a world supposedly governed by performance.

The post A Dry-Eyed Conspiracy appeared first on FMTC.