A Google rep's $250/day AI Max build on a trade-tools store returned $0.67 sales per $1 over three weeks vs ~$3.05 on the rest of the account. Forty-five percent of spend hid from search terms, and under 2% came from keywordless matching.
Why it matters: Treat every AI Max pitch as a test on a winning Search campaign with a pre-written stop rule, not a cold campaign Google designed to look like a win.
Kristina Frunze's B2B SaaS shop pipes Peec visibility through a Claude MCP skill that splits share of voice by engine, tiers listicle gaps (25+ retrievals first), and traces each objection to the source URL.
Why it matters: If AI-search work is still a CSV export, the operator advantage is the MCP loop plus evals, not another visibility dashboard.
MATG Ep 461 has Kieran walking a researcher-to-draft Grok Bot lineup with five prompts, Monit API access, and model-per-step routing across Claude, Codex, and Opus.
Why it matters: Content ops is splitting into specialist agents with a human gate. Copy the role split, not a single chatbot prompt.
Goodway is running incrementality inside Arima's synthetic population before small live activations. Aveeno cut strategy briefs from 30 days to 3 with Vurvey, and Sacheu is cloning personas from its creator roster.
Why it matters: Planning agents will ship faster briefs. Keep a live-holdout check so the simulation does not become the media plan.