The daily briefing for AI marketers, growth hackers, and operators.

TL;DR — Bain survey of 1,100 US consumers: 80% rely on zero-click AI results for 40%+ of searches. Adobe reports 1,200% increase in AI-referred retail traffic (Feb 2025 vs July 2024). AI-referred conversion rates now 23% below traditional search, down from 49% gap in January. Discovery, evaluation, and shortlisting now happen inside AI tools before brands see the customer. HubSpot reports up to 30% traffic decline to company sites. Optimize for machine-readable content, structured data, and third-party validation.

TL;DR — Google's third 2025 core update rolled out December 11-29. E-E-A-T signals and AI content quality remain central evaluation criteria. New documentation confirms Google rolls out smaller, unannounced core changes continuously. Sites hit by September 2023 HCU saw partial recoveries during June update. No specific recovery actions; focus on helpful, people-first content. Health, finance, news, and shopping sectors experienced highest volatility.

TL;DR — Stop freestyle prompting Claude Code. Three-phase system: Spec (define marketing outcome, launch milestones, create project spec with marketing and technical requirements), Stack (seven-step config including claude.md with brand guidelines, tracking standards, integration patterns, MCPs for analytics/CRM/deployment), Ship (three workflows: general for single pages, campaign-based for multi-asset launches, multi-agent for parallel development). Key insight: 15 minutes of speccing saves weekend debugging. Always verify tracking before considering anything done. 'Page is live' is not done. 'Conversions recording correctly' is done.

TL;DR — OpenAI co-founder admits feeling behind despite building these systems. Marketing parallel is direct: the gap between marketers using AI as a feature and marketers orchestrating AI workflows is widening fast. His vocabulary (agents, prompts, contexts, memory, tools, plugins, workflows) maps to marketing ops. The 10X productivity claim requires stringing tools together correctly. Key insight: failure to capture AI leverage is now a skill issue, not access issue. Same tools available to everyone. Competitive advantage shifts to those who build mental models for 'stochastic, fallible' systems. Marketers face identical challenge: learn to orchestrate unreliable-but-powerful AI across content, analytics, automation.

TL;DR — LLMs use the same probability system to explain themselves as they do to answer questions. When you ask 'why did you recommend that?', you get another statistical lottery, not truth. SparkToro tested 100 people asking ChatGPT identical knife recommendation prompts. Almost no two got the same brand list. When asked to explain, ChatGPT fabricated reasoning. Marketers making decisions based on LLM self-explanations are building on false foundations. The only honest answer: 'most likely token based on training data.' Applies directly to anyone using AI for brand tracking, competitor analysis, or content recommendations.