Wyndham's $565 million marketing bench is not waiting on Unilever-scale AI spend. Five people plus Adora cut production time 75%, shipped 15 times the Instagram assets, and lifted site-visit rate 6x over benchmark, while Opella's 85-person team hit 20,000 pieces and BetMGM still will not let the model write the odds.
Why it matters: Steal the loop, not the vendor: pair a small in-house bench with platform performance data before you scale volume.
DemandSphere watched branded AI Overviews jump from about 26% of tracked queries at the start of September to more than 80% by month-end, peaking at 90.48% on Sept. 27. Chris Long hit 93 of 100 enterprise brands in a SerpApi pass, and neither cut measured the CTR.
Why it matters: Audit what the branded Overview says and cites this week, and treat the brand query as an AEO surface, not a guaranteed homepage click.
Reddit operators spotted UTMs on Gemini outbound links and Mueller confirmed he sees them too. CrawlWarden has the Gemini mobile-app referrer surviving about 9% of the time and Android assistant invocations stripping it entirely, which is why a UTM that lives through the in-app webview and server logs matters more than the desktop referrer.
Why it matters: Rebuild the Gemini versus Direct split in GA4 and raw logs this week before you decide the assistant sends no traffic.
On blu.AI the brand owns the underlying data and any model trained on it, and Bob Lord called agencies treating those tools as agency IP "not acceptable." WPP is already seeing DSP usage go agent-to-agent, and the panel's answer to the FTE hole is to license the login rather than cut the bench.
Why it matters: Put data-and-model ownership in the SOW before the next agent build, and price the interface, not the headcount.