The Problem With Every Metric You Are Currently Using
Time on page tells you a user stayed. It does not tell you whether they absorbed anything. Scroll depth tells you a user reached the bottom of an article. It does not tell you whether they read it or just scrolled past. Click-through rate tells you someone moved. It says nothing about what moved them to act.
These metrics share a common flaw: they measure proximity to content, not cognitive engagement with it. And in a media environment where a brand's survival depends on whether an AI model decides to cite its content, proximity is no longer enough.
The metric that actually predicts whether content changes behavior, drives recall, and earns citation from large language models is attention decay.
What Attention Decay Is
Attention decay describes the rate at which human cognitive engagement with a piece of content diminishes over time and position. It is not a metaphor. It is a mathematically defined curve that can be measured, calibrated, and applied across every content format and distribution channel.
The concept derives from Hermann Ebbinghaus's forgetting curve, established in 1885, which showed that memory retention follows a predictable exponential pattern after initial encoding. Attention decay applies the same mathematical logic not to memory, but to the moment of content consumption itself: how fast does engagement fall as a reader progresses through an article, a video, or an AI-generated response?
At BAXindex, this principle is operationalized through the Behavioral Biometrics Engine (BBE), calibrated on more than five million URLs across nine content formats. The decay function takes the form:
Where d is position or distance from the attention peak, lambda controls the primary exponential drop, mu governs a secondary Gaussian component for formats with midpoint recovery, and alpha balances the two components. The result is a decay curve unique to each content format, grounded in behavioral signals rather than declared engagement.
Why Lambda Is the Number That Changes Everything
Lambda is the single most important parameter in the decay function. It defines how steeply attention falls off per unit of content depth or position.
BAXindex has calibrated lambda values across nine major content formats, based on behavioral data collected at scale. The full dataset is reserved for enterprise partners under NDA, but the cross-channel pattern reveals something structurally important:
| Format | Lambda | What it means |
|---|---|---|
| YouTube Shorts | 0.31 | Slow decay — short format, full commitment from start |
| Web editorial | 1.40 | Sustained engagement across long reads |
| YouTube long-form | 1.44 | Near-identical to web editorial — cross-channel proof |
| AI responses | 1.56 | Steep positional decay — position 1 vs position 5 gap is exponential |
| Web news | 1.80 | Faster decay — inverted pyramid format |
| Web short post | 2.20 | Limited depth engagement |
| Facebook feed | 5.40 | Very steep — attention window measured in seconds |
| Instagram feed | 8.70 | Extreme decay — visual-first, minimal text retention |
| TikTok | ~10–12 | Highest decay — attention resets with every video |
Two values being nearly identical — Web editorial (1.40) and YouTube long-form (1.44) — is not a coincidence: it is a validation proof. Two completely different channels, consumed on different devices, in different contexts, produce mathematically equivalent attention decay curves. This confirms that BBE is measuring a real cognitive phenomenon, not a channel artifact.
AI responses carry a lambda of 1.56, calibrated against Perplexity citation click-through data showing that position one receives 100 index points of engagement while position five receives only 21. The decay from first citation to fifth is exponential, not linear. A brand that appears fifth in an AI response does not receive one-fifth of the engagement of a first-placed brand. It receives roughly one-fifth of an already-diminished fraction.
What This Means for Content ROI
Content ROI is conventionally measured as a function of reach, engagement, and conversion. The problem is that reach metrics — impressions, views, sessions — treat all exposures as equivalent. They do not account for the fact that a brand appearing at position three in a Perplexity response is delivering approximately half the cognitive impact of a position-one appearance, and that this gap compounds with every piece of content that fails to earn deep attention.
When attention decay is incorporated into ROI modeling, the calculation changes substantially. It is no longer sufficient to ask how many people saw the content. The relevant questions become: at what position in the attention curve did they encounter it, how steeply did engagement fall before they reached the key message, and what behavioral signals confirm genuine cognitive contact rather than passive proximity?
The BAX Index incorporates attention decay as a weighted component within a four-factor formula:
Decay-weighted reach carries the largest weight because reach without attention weighting systematically overvalues low-quality exposures and undervalues deep-engagement placements.
Why AI Changed the Stakes
Before large language models became a primary interface for information retrieval, attention decay affected conversion rates and brand recall. The consequences were real but manageable: underperforming content could be replaced, amplified, or retargeted.
In the AI era, the stakes are structurally different. LLMs do not index all content equally. They build corpus-level representations of authority, trust, and relevance. Content that consistently generates high-attention behavioral signals becomes structurally more likely to be incorporated into model outputs. Content that accumulates low-attention signals, regardless of its volume or its reach, is progressively deprioritized.
This means that attention decay is no longer only a performance metric for marketing teams. It is a determinant of whether a brand earns organic presence in AI responses at all. A brand whose content consistently occupies the low end of the attention curve — generating skimming rather than reading, proximity rather than comprehension — is a brand that is actively training LLMs to treat it as low-authority.
The inverse is also true. Brands whose content generates deep-reader behavioral patterns, re-read loops, sustained attention retention, and low drop-off rates are building the kind of corpus signal that earns citation. The attention curve is, in this sense, an AI visibility strategy.
How Attention Decay Is Measured at Scale
Eye-tracking panels measure where attention lands. They do not measure how long it stays, how it moves through a piece of content, or whether engagement deepens or collapses as the reader progresses. Panel-based attention measurement also operates at sample scale, not census scale.
Behavioral biometrics operates differently. Rather than observing eye position, it reads the micro-patterns of how a user physically interacts with content: scroll velocity and rhythm, pause signals that indicate re-reading or reflection, cursor hesitation at specific content sections, interaction frequency, and the shape of the scroll path through a full article.
These signals, aggregated across millions of sessions, produce attention decay curves that are both population-level and content-specific. The BBE does not produce a single number representing overall engagement. It produces a curve showing how engagement evolves from the first paragraph to the last, which sections accelerate decay and which slow it, and what structural features of content correlate with sustained attention versus rapid abandonment.
The output of this measurement is a format-specific lambda value for each piece of content, which can then be used to weight reach figures, inform editorial decisions, benchmark content across publishers and channels, and ultimately calculate an attention-adjusted ROI that reflects what the content actually delivered cognitively.
The Practical Implication for CMOs
If your current content measurement stack does not incorporate attention decay, you are making budget allocation decisions based on reach and engagement figures that systematically misrepresent the cognitive value delivered to your audience.
You are likely overspending on channels with steep decay curves, where only the first seconds of contact carry weight and where even strong viewability scores produce no memory trace. You are likely undervaluing long-form editorial placements where lambda values indicate sustained cognitive engagement and where behavioral signals are generating positive corpus signals for AI models.
The first step toward attention-adjusted investment is not to abandon existing metrics but to weight them. Reach multiplied by a decay coefficient calibrated for its channel and format produces a figure that actually correlates with recall, preference shift, and AI citation probability. That number is the BAX Index, and it is the only metric currently available that connects AI visibility to human attention to source authority to actionable editorial fixes in a single, auditable workflow.