Research: Traditional Marketing Doesn’t Work on AI Shopping Agents

Why It Matters

As AI agents capture an increasing share of online purchases, relying on human‑focused marketing cues risks wasted spend and even backfiring, forcing brands to redesign their conversion strategies for a non‑human audience.

Key Takeaways

• Star ratings boost AI agent selections across models and categories
• Scarcity, countdown, and strike‑through cues show inconsistent or negative effects
• Non‑reasoning models respond more to promotional badges than reasoning ones
• Pricing remains a strong negative driver; higher prices reduce AI selections
• Tailoring product data to specific AI models can improve conversion

Pulse Analysis

The rise of AI‑driven shopping assistants is reshaping digital commerce. OpenAI’s deeper integration of ChatGPT into product discovery, Google’s universal commerce protocol (UCP), and Amazon’s tools that let its agents shop rival retailers have turned algorithms into active buyers. These agents parse product pages, compare specs, and complete transactions based on user prompts, effectively becoming a new class of shopper that bypasses traditional human decision‑making pathways. Brands that once optimized for human psychology now face a parallel audience that evaluates offers through a computational lens.

In a large‑scale simulation covering four AI models—GPT‑4.1‑mini, GPT‑5, Gemini 2.5 Pro, and Gemini 2.5 Flash Lite—researchers tested eight promotional badges across four product categories. Star ratings emerged as the sole cue that consistently increased selection, mirroring human reliance on quality signals. Price behaved predictably, with higher costs reducing picks. By contrast, scarcity notices, countdown timers, and strike‑through discounts produced erratic or even negative effects, especially in more advanced reasoning models. Non‑reasoning models showed slightly higher sensitivity to these cues, but the variability underscores that human‑centric heuristics no longer guarantee conversion.

For marketers, the implication is clear: prioritize fundamentals—competitive pricing and authentic review profiles—before investing in AI‑specific persuasion. Treat each AI model as a distinct market segment, tailoring product feeds and badge displays to the agent’s response profile. Building simulation environments that replay product pages against multiple AI versions can surface shifting dynamics as models evolve. Early adopters who develop real‑time detection of agent types and dynamic content adaptation will capture the emerging AI‑driven purchase flow, while those clinging to outdated human‑only tactics risk losing relevance in the next wave of e‑commerce.