GEO Glossary

Adstock

Adstock is the MMM transform that captures how media exposure carries over into future periods. Definition, common functional forms, and AI search application.

By Ramanath, CTO & Co-Founder at Presenc AI · Last updated: April 23, 2026

What Is Adstock?

Adstock is the mathematical transformation in marketing mix modeling that captures the carryover effect of media exposure into future periods. A TV ad seen this week influences purchase decisions for weeks afterward; a PR placement keeps generating consideration for months. The adstock function distributes that influence over time so the model can fit it.

Without adstock, the regression would assume that media spend in week one has zero effect in week two, which contradicts every consumer behavior study ever published. Adstock is the operational fix.

Why Adstock Matters

The choice of adstock function and its parameters is one of the highest-leverage decisions in an MMM build. A half-life that is too short understates the channel's effect; a half-life that is too long overstates it and pulls credit from other channels. Modelers spend more time tuning adstock priors than almost any other aspect of model specification.

For AI search specifically, adstock matters because AI assistant exposure has a meaningful carryover: a brand recommendation surfaced this week influences purchase decisions for several subsequent weeks. Borrowing the paid search adstock prior (essentially zero carryover) for the AI variable produces an undervalued AI channel.

Common Functional Forms

Geometric adstock: The simplest and most widely used. Each period's effect decays by a constant factor (theta). Parameterized by the decay rate or, equivalently, the half-life. Standard choice when no strong prior knowledge suggests otherwise.

Weibull adstock: More flexible than geometric. Allows the effect to peak some periods after exposure rather than immediately, which is realistic for upper-funnel channels including TV and AI search. Parameterized by shape and scale.

Delayed adstock: A simpler peak-delayed form than Weibull. Useful when the analyst wants intuitive control over peak timing without the full Weibull flexibility.

In Practice

Typical half-life priors by channel: paid search 0-1 weeks, paid social 1-2 weeks, display 2-4 weeks, AI search 2-4 weeks, TV 3-6 weeks, OOH 4-8 weeks, PR 4-12 weeks. These are starting points; Bayesian MMM frameworks refine them from data when the time series is informative.

For AI visibility, geometric adstock with two- to four-week half-life is the standard starting prior. Categories with long consideration cycles (B2B SaaS, financial services, healthcare) skew to the upper end; short consideration cycles (snacks, fast fashion) skew to the lower end.

How Presenc AI Helps

Presenc AI provides AI visibility data in the right format for adstock-transformed MMM input. The platform also publishes adstock prior guidance by category so that the AI variable enters new MMMs with informative priors rather than vague defaults that underweight the channel.

Frequently Asked Questions

Two to four weeks for most categories. Long-consideration B2B may extend to eight weeks; short-cycle consumer may compress to one to two weeks. The single most common error is borrowing the paid search prior (zero to one week), which produces an undervalued AI coefficient.
Geometric for most channels and as a starting point for AI search. Weibull is better when the peak effect is known to lag exposure (TV, podcast, upper-funnel video). For AI search the choice usually does not matter materially because the carryover is short enough that geometric and Weibull produce similar estimates.
Robyn applies adstock as a preprocessing step on media variables with bounds on the decay rate. LightweightMMM and PyMC-Marketing model adstock parameters jointly with the rest of the model in the Bayesian inference. Both approaches work; Bayesian joint inference produces tighter posteriors when the time series is informative.
No. Adstock represents positive carryover; the parameter is bounded to produce non-negative decay. Negative effects (advertising fatigue, brand backlash) are modeled through other mechanisms, typically saturation curves or explicit fatigue terms rather than the adstock function itself.

Track Your AI Visibility

See how your brand appears across ChatGPT, Claude, Perplexity, and other AI platforms. Start monitoring today.