+## Advertising to AI Agents Has Reached the Inevitable Question
Publishers and startups experimenting with advertisements designed to be consumed by AI crawlers are struggling to establish whether those ads were retrieved, actually processed by the model, incorporated into a response and ultimately influenced a human purchase.
Current experiments range from AI-specific referral codes to dynamically injecting sponsored text into pages when AI crawlers arrive. But as Digiday reports, no industry-wide attribution standard exists yet.
The IAB is now drafting an attribution framework for agentic advertising. The core problem is fundamental.
The Seven-Step Causal Chain
Traditional digital advertising had a relatively simple attribution path:
Impression → Click → Conversion
And even that produced two decades of attribution arguments.
Agent advertising is orders of magnitude harder. The proposed causal chain looks something like:
- Ad exists on a webpage
- Agent crawls the page
- Sponsored information survives retrieval and processing
- Model incorporates the commercial content
- Recommendation changes because of the ad
- Human sees the AI-generated recommendation
- Human acts on the recommendation
There are at least seven causal steps where the chain can break.
A Crawler Hit Is Not an Impression
This is the conceptual mistake I would warn against in both agent advertising and GEO measurement [blocked]:
Observable presence is not attributable influence is not incremental commercial value.
That gives you a clean measurement hierarchy:
| Level | Question | Current Measurability |
|---|---|---|
| Exposure | Did the agent access the sponsored evidence? | Partially measurable via crawler logs |
| Utilisation | Did the model use it? | Very difficult to measure |
| Influence | Did it materially change the answer? | Requires controlled experiments |
| Outcome | Did user behaviour change? | Requires end-to-end tracking |
| Incrementality | Would the transaction otherwise have happened? | Requires holdout testing |
Unless agentic advertising solves those layers, "GEO lift" could become a beautifully packaged vanity metric.
Why This Matters for B2B Marketers
The agent advertising attribution problem is not just a media measurement issue. It is a preview of the measurement challenges that will affect every agentic AI deployment [blocked].
When an AI agent recommends your product, how do you know:
- Whether your content influenced the recommendation?
- Whether the recommendation influenced the purchase?
- Whether the purchase would have happened anyway?
These are the same incrementality questions that have challenged PPC and performance marketing [blocked] for years — but with far more causal steps and far less observable data.
The Agent Influence Measurement Framework
I would propose separating agent advertising measurement into five distinct layers, each requiring different methodology:
- Exposure measurement: Crawler log analysis, page-level access tracking
- Retrieval verification: Synthetic query testing to confirm content appears in AI responses
- Recommendation impact: A/B testing with and without sponsored content to measure answer changes
- User behaviour tracking: End-to-end attribution from AI recommendation to conversion
- Incrementality testing: Holdout groups and geographic experiments to isolate causal impact
No single metric captures the full chain. Any vendor claiming simple agent advertising ROI is measuring only one layer and calling it the whole story.
Agentic advertising has a causality problem, not just an attribution problem. Until advertisers can distinguish crawler exposure from model utilisation, recommendation influence and incremental revenue, agent ads risk becoming the most sophisticated vanity metric in marketing history.
Understand your AI visibility before investing in agent advertising. Try Manus free →
