---
id: "claim-fixed-strategies-expire"
type: "claim"
source_timestamps: ["¶4", "\\\"§ Build a testing infrastructure", "not a one-off strategy.\\\""]
tags: ["strategy", "model-updates"]
related: ["concept-continuous-ai-simulation-infrastructure", "open-question-model-update-volatility"]
confidence: "high"
testable: true
speakers: ["Jafar Sabbah", "Oguz A. Acar"]
sources: ["geo"]
sourceVaultSlug: "hbr-seg-geo"
originDay: 3
articleStem: "hbr-tier2-06-ai-shopping-agents"
sourceUrl: "https://hbr.org/2026/05/research-traditional-marketing-doesnt-work-on-ai-shopping-agents"
sourceTitle: "Research: Traditional Marketing Doesn’t Work on AI Shopping Agents"
---
# Fixed AI optimization strategies have a short shelf life

**Claim (confidence: high · testable: true):** Because AI behaviors are tied to **specific model versions**, any finding about how an agent responds to a marketing cue may **expire with the next model update**. Every major release, fine-tuning adjustment, or safety alignment can fundamentally shift how an agent interprets pricing frames or urgency cues — rendering static optimization strategies useless.

This is the driving rationale for [[concept-continuous-ai-simulation-infrastructure|continuous simulation infrastructure]] and the action item [[action-build-simulation-environment|build continuous AI simulation environments]]. It also frames the [[open-question-model-update-volatility|open question]] about which alignment changes cause the most drastic commercial shifts.

> "[[quote-hypotheses-to-test|...findings may expire with every model update.]]"

**Enrichment / external corroboration:** ACES/ACE longitudinal analysis across model generations (e.g., GPT-4.1 → GPT-5.1, Gemini 2.5 Flash → 3 Pro Preview) shows updates can produce **almost opposite position biases** and distinct product preferences, functioning as **"exogenous demand shocks."** Basic rationality improves over time, but market shares, choice homogeneity, and position biases remain highly model-dependent and shift with new releases. Well grounded.

**Related:** [[concept-continuous-ai-simulation-infrastructure]] · [[open-question-model-update-volatility]] · [[action-build-simulation-environment]]


## Related across articles
- [[concept-continuous-ai-simulation-infrastructure]]
- [[question-som-volatility]]
- [[open-question-model-update-volatility]]