The Collapse of “Intent” as a Signal

Why intent was never that stable to begin with.

Sarah wasn’t pretending to want hiking boots.

For about forty-five minutes on a Wednesday night, she genuinely did. She compared waterproofing materials, flipped between brands, read reviews from strangers in Colorado, signed up for a discount, and came back days later to browse backpacks she didn’t need yet.

And then she moved on.

Not because her interest was fake, but because attention rarely behaves the way systems expect it to. Work got busy. She got a flat tire. Her sister needed help planning a birthday. The version of Sarah who hikes on weekends gave way to everything else competing for her attention, even as AI continued reinforcing that moment and turning it into something that looked like ongoing intent.

Which raises an uncomfortable possibility: intent isn’t a stable signal. It’s fragile, contextual, and becoming harder to trust as AI produces the same behaviors we’ve been using as evidence of it.

The system still wants intent to be real

For years, marketing relied on a simple assumption: behavior stands in for motivation.

Someone is planning a hiking trip.
Or at least, someone looks like they are.

We don’t just browse products. We step into temporary versions of ourselves, and increasingly, those moments are shaped as much by machines as by us.

If someone keeps clicking, browsing, and returning, it must mean something. Engagement looks like commitment, so confidence follows. That assumption worked when behavior felt more linear. Today, there’s more noise: boredom, distraction, stress, and fleeting reinvention.

Now AI is amplifying those signals and generating more of them.

What intent was really meant to capture

Intent was always a proxy for something more concrete: whether interest persists long enough to turn into action.

The difficulty is knowing whether that activity belongs to a consistent, reachable person and whether it continues beyond the moment it appears. Most systems evaluate behavior as a series of loosely connected events, which makes it easy to assign meaning unnecessarily.

Email-based identity introduces continuity by linking individual events to a persistent identifier, allowing behavior to be evaluated on whether it connects to prior activity, reappears over time, and forms a consistent pattern tied to a reachable individual.

A single interaction no longer carries inherent weight. Its value depends on whether it aligns with a broader sequence of behavior:

• Does interest reappear after attention shifts?
• Does it reinforce an existing pattern rather than create one out of nothing?
• Does it belong to an identifiable, active profile with a history of engagement?

This is the underlying shift: from interpreting events in the moment to evaluating behavior as part of a continuous pattern tied to identity, where signals are weighted based on persistence, consistency, and reachability.

The goal isn’t to perfectly label intent.
It’s to build confidence in whether observed behavior reflects something durable enough to act on.

Signals get louder as meaning gets thinner

The problem is, it’s getting harder to tell the difference.

Recommendation engines don’t just observe behavior, they shape it. What gets surfaced, revisited, and reinforced is already filtered through layers of interpretation. AI accelerates this, assigning confidence before attention has had time to settle into anything durable.

So, that same hiking boots session gets recorded and then amplified:

• Products reappear across channels
• Suggestions become more targeted
• Engagement loops tighten

Meanwhile, Sarah has already moved on.

Increasingly, AI agents, automated journeys, and synthetic interactions are generating the same signals we’ve historically treated as evidence of intent: visits, clicks, comparisons, and engagement.

In Sarah’s case, a moment becomes a pattern. In other cases, those patterns may not come from a person at all.

The issue isn’t that we have too little data, it’s what we do with it. When every click, visit, and interaction is taken at face value instead of grounded in a persistent identity, budget gets misallocated, messaging misses the mark, and confidence builds around the wrong signals.

The signal gets louder. The meaning gets thinner.

Intent was never the problem

Sarah’s behavior wasn’t misleading. The interpretation was.

Intent didn’t collapse because people got less sincere. It collapsed because we expected consistency from behavior, and kept trusting those signals, even as AI systems began amplifying, shaping, and increasingly generating that behavior on their own.

People are still curious, inconsistent, distracted, and aspirational. What’s changed is the environment: signals are no longer just human: they’re reinforced, manufactured, and made to look more meaningful than they are.

So, the question isn’t how to measure intent more precisely.

It’s whether behavior can still stand in for intent, or whether the real challenge now is deciding which signals actually reflect something real.

If you’re only working with part of the signal, you’re only getting part of the value. The rest is still out there.

Take the next step toward better data.