Brandtech

Emploi, productivité et créativité : le marketing face à la révolution IA

Financial Times Special

Marketers Seek Fix for AI's Creative 'Sameness Trap'

Marketing strategists and agency planners report that large language models (LLMs) such as ChatGPT and Claude are producing increasingly homogeneous, predictable outputs that limit creative novelty. Practitioners use LLMs for research, critique and ideation (examples: internal bots like a “Banging Brief Bot”, persona agents via Pencil), but some are intentionally avoiding the tools for tasks requiring originality. Academic work (Carnegie Mellon researchers' 2025 NoveltyBench) finds larger models yield less diverse responses. Startups are pursuing technical alternatives: Sydney-based Springboards built a smaller, “divergence” model called Flint (based on Alibaba’s Qwen family / ~30B parameters) that scored higher on NoveltyBench and plans an API release later in the year. Agencies debate whether technical fixes or human curation are the right response to AI-driven sameness.

Marketers See AI Benefits; Trust Limits Agentic AI

Digiday+ Research (excerpt) summarizes a survey of 142 brand and agency professionals (conducted Q4 2025) showing widespread embedding of AI across marketing workflows but persistent barriers to broader adoption. Generative AI has higher adoption than predictive AI, with generative tools most used for creative production (82%), marketing (81%) and external/internal communications (75%/56%). Predictive AI is most commonly applied to measurement and KPI analysis (48%). More than half of respondents (54%) said their companies do not use agentic AI; interviewees and case studies (Unilever, Kroger, Monks) illustrate benefits but highlight trust, governance, data access and technical complexity as obstacles to agentic AI adoption.