Budget-limited Target CPA and ROAS campaigns now climb toward the stated target instead of clearing under it, so the same spend buys fewer, pricier conversions. The gradual rollout started August 17 across Search, Shopping, PMax, Demand Gen and Travel, and it lands on top of language-targeting removal and the AI Max conversion, which makes any performance drop hard to attribute.
Why it matters: If you run tCPA or tROAS, log a pre-change baseline this week. Without it the next 60 days of reporting is unreadable.
Google is using generative AI to extend horizontal Performance Max video into vertical and square formats, on by default. It buys more inventory and quietly moves creative control to the platform.
Why it matters: Creative governance is now a real line item. Decide your opt-out position before the September 4 deadline rather than after.
DataBeat's May data has agentic demand clearing at $6.13 CPM against $6.95 for conventional buyers, a 13.4% discount, with 11.5% better fill while sitting out 86% of auctions. Agents aren't optimizing the auction, they're opting out of it toward guaranteed-style deals.
Why it matters: Treat agentic buying as a change in deal structure, not a tooling upgrade. The negotiation moves before the auction does.
Only 11% of brands appeared across ChatGPT, Gemini and Claude, 77% surfaced in just one, and roughly 5% with strong domain ratings and real organic traffic drew almost no model references at all, Aetna, Cigna, Sephora and Samsung among them. Third-party mentions move the needle, owned content doesn't.
Why it matters: Search rank and model visibility are separate metrics. Measure them separately, and per model, or you will assume coverage you don't have.
RESONEO's teardown of 1,200 answers and 26,900 pages found three retrieval layers, a shared cache that ignores your Cache-Control and noindex headers, and only 5% of retrieved URLs becoming citations. That rises to 74% for pages actually opened versus 7% for those that aren't.
Why it matters: These are testable mechanics you can audit against, which beats generic advice to structure your content better.
Trace Cohen went from near-zero to 4.79M impressions and 18.1K clicks running weekly striking-distance, cannibalization and orphan-page passes, then open-sourced the 14 scripts behind it. Once content production is cheap, measurement and site architecture become the moat.
Why it matters: The system produced the result, not the publishing volume. Worth copying as a weekly operating cadence.