We Tested Marketing Incentives to AI Agents. Here's What Happened.
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
Ramp ran a 5-week experiment embedding tracked referral incentives into pages on ramp.com, served conditionally to AI bots via Cloudflare Workers, to test whether AI agents would relay the offer to human users. Three content formats were tested: markdown, stripped HTML, and schema markup. Markdown was the only format that reliably surfaced in LLM responses. Claude eventually became the dominant relay channel, going from zero mentions to ~370 agent relays with specific offer details and booking URLs. ChatGPT never surfaced the offer once across 32 days. Perplexity picked it up quickly but kept mentions vague. Key findings include: Anthropic's crawler was more aggressive than all other AI crawlers combined; pages with existing high AI citation volume were far more likely to surface embedded content; and bot detection via Cloudflare categories misses major LLM platforms. The experiment suggests per-model content strategy is necessary, and that 'agent trust' may function analogously to domain authority in traditional SEO.
