AI Slop Is the New Bot Traffic — and We Just Shipped the Defense

Bot traffic flooded display advertising in the 2010s. AI-generated slop is doing the same to agent commerce — and Adobe Q2 2026 data shows +393% YoY growth. Nexbid just shipped layers 1 and 2 of the defense.

Display advertising in the 2010s lost a measurable portion of every campaign budget to bot traffic. Bots clicked ads. Bots viewed pages. Bots inflated impression counts. The industry's response — IAB-accredited filters, ads.txt, sellers.json — took the better part of a decade and is still incomplete. Agent commerce in 2026 faces an analogous threat at a different layer of the stack. The threat is AI slop: machine-generated content with no provenance, no editorial gate, no human accountability — produced at zero marginal cost, designed to compete for the same agent attention as legitimate publisher content. If a marketplace cannot distinguish slop from substance, the marketplace cannot exist. Yesterday Nexbid shipped the first two layers of that distinction. Why this is a 2026 problem, not a 2027 problem Adobe's Q2 2026 AI Traffic Report, published this quarter, makes the timing concrete. Traffic from AI sources to U.S. retail sites grew 393% year-over-year in Q1 2026. The 2025 holiday season alone showed +693% YoY. About a quarter of consumers now cite AI platforms — ChatGPT, Gemini, Perplexity — as their primary research tool, surpassing brand sites and traditional review sources. The conversion data is more striking. In March 2026, traffic from AI sources converted 42% better than non-AI traffic (a new record), with shoppers spending 48% longer per visit and browsing 13% more pages. These are not exploratory clicks. These are intent-loaded sessions — the kind that command high CPMs and high CPAs in a healthy auction, and the kind that justify high-volume slop generation in an unhealthy one. Volume is the precondition for the slop problem to bite. We just hit the volume. What slop looks like in agent commerce A publisher uploads a feed. The feed contains 10'000 product descriptions, each 80 to 120 tokens, each "enriched" by a generic LLM prompt. The descriptions are grammatical. They contain category-correct keywords. They look like content. They are not content. They are noise pretending to be signal. An agent that ingests this feed and surfaces it to a user has been weaponised — the publisher has converted the agent's trust into reach without earning the trust. The pattern matches bot traffic in display: low-cost generation, mass-produced volume, indistinguishable from legitimate engagement at the surface level, distinguishable only with structural verification. Layer 1 — provenance and length gates Every piece of content entering the Nexbid catalog now carries a provenance hash. The hash …

Author

Holger von Ellerts

Published

2026-04-29

Topics

Agentic, Enrichment, Ad Tech, Protocol Commerce