1 Meta retired the off-platform data opt-out — your retargeting pool just got bigger
Meta is retiring the “Your activity off Meta technologies” setting and replacing it with a broader control called “Activity from other businesses.” The data it governs now feeds three systems instead of one: ad targeting, Feed recommendations, and Meta AI responses. Announced June 9, the rollout starts this month across the US, UK, Brazil, and other markets, with the EU handled separately under GDPR. Users who previously disconnected their off-platform activity re-enter targeting pools.
The practical effect: retargeting audiences quietly expand, and your frequency and CPM math shifts with them — watch audience sizes on website custom audiences over the next few weeks. The bigger shift is philosophical: your pixel events now shape what users see organically and what Meta AI says, which makes event hygiene a brand-experience input, not just an ads one.
2 Google flips Demand Gen on Discover to CPM billing — July 15, automatically
Google began notifying advertisers on June 16 that Demand Gen campaigns serving on Discover with view-through conversion optimization enabled will move from CPC to CPM billing on July 15. The change applies automatically, with no advertiser action required. Google's rationale: view-through conversions are triggered by impressions seen, not clicks taken, so the billing unit should match. Only campaigns that explicitly enabled VTC optimization after its April launch are in scope.
If that's you, your budget math changes next Wednesday — spend pacing, impression volume, and effective CPA will all move. The choice is binary: rebuild your forecasts on CPM, or opt out by disabling VTC optimization. Note the pattern — the default is Google's preference, and silence equals consent.
3 Amazon will cap product titles at 75 characters — and its AI rewrites the overage
Starting July 27, Amazon caps product titles at 75 characters. Titles over the limit get automatically rewritten using Amazon's AI recommendation — in search results and on the product detail page. Stibo Systems and CommerceIQ announced a partnership on July 10 that gives Stibo's PIM customers complimentary access to CommerceIQ's Content Agent to rewrite titles before the deadline, with PIM compliance, SEO, and answer engine optimization named as the next areas of collaboration.
If you sell on Amazon, your brand voice on the digital shelf now has a two-week deadline: rewrite your titles yourself, or the machine does it in Amazon's voice, not yours. Expect this to become the template — retailers enforcing AI-normalized content at the platform level, with brands given a compliance window instead of a choice.
4 Mastercard added a sandbox to the agent era — test before you trust
Mastercard expanded its Agent Suite with Proto, a sandbox environment for exploring, testing, and validating agentic AI use cases before they touch production workflows, plus three new agents covering merchant management, onboarding, and dispute management. The pitch: organizations need to evaluate agent performance, find failure points, and refine controls before moving from pilots to trusted deployment.
A payments giant just productized the pilot-to-production gap — the exact place where most enterprise agent projects stall. The signal for marketers: buyers now expect a place to watch agents fail safely before letting them near money or customers. Ask every agent vendor in your stack where their sandbox is; if the answer is “production,” that's your answer.
5 Workato shipped Agent Guardrails — every agent action gets a name attached
Workato announced two Agent Studio capabilities: Headless API, which lets its agents run on any surface, respond to business events, and coordinate with other agents without a human triggering each step; and Agent Guardrails, which blocks or tokenizes PII before it reaches the model, binds every action to a real user identity or approved service account, routes high-stakes actions to human approval in Slack or Teams, and logs everything in an auto-redacted history inheriting SOC 2 Type II, ISO 27001, HIPAA, and PCI-DSS 4.0 certification.
Most organizations have proven agents work in a pilot — production stalls when security and compliance get renegotiated for every new surface an agent touches. Governance delivered as native infrastructure removes that rework. Budget for guardrails the way you budget for seats: they're not a nice-to-have line item, they're what unblocks scale.
6 IBM put a cost meter on agents — meet Return on Token Investment
IBM updated Bob, its agentic software development platform, with Bobalytics for consumption and cost visibility, multi-agent coordination, and subagents that isolate exploratory work so context-window inflation stops driving up spend. The release cites a survey finding 85% of DevSecOps professionals say AI shifted the bottleneck from writing code to reviewing it — and one modernization program projected at nine months with fourteen engineers that finished in three days.
The discipline IBM is naming applies directly to marketing AI budgets: the unit economics of agent execution — not model access — determine whether a workflow is worth automating. Marketing ops teams should start tracking cost per automated workflow now, before finance starts asking. “Return on Token Investment” is coming to a QBR near you.
7 DoubleVerify exposed “DreamScheme” — AI slop built an industrial ad-fraud funnel
DoubleVerify's Fraud Lab identified DreamScheme, a mobile-first fraud operation that uses AI-generated “dream girl” ads to lure users into networks of made-for-advertising sites and search-arbitrage funnels. The networks drew over 2 million visits per month in Q1 2026. A parallel scheme uses deepfaked pop-culture characters to promote AI services through the same arbitrage plumbing, and DV has flagged thousands of AI slop sites across multiple languages this year.
Generative AI dropped the cost of manufacturing fraud inventory to roughly zero, which means MFA regenerates faster than blocklists update. If you buy programmatic, category blocks are no longer enough — move to inclusion lists, audit them quarterly, and ask your verification partner specifically about AI-generated MFA coverage.
8 Multiply launched “10 Minute ABM” — and ChatGPT Ads is already on the channel list
Multiply announced 10 Minute ABM: define target accounts, messaging, and objectives, and the system generates hundreds of personalized account-based ads, launches them in under ten minutes, and runs continuous per-account experiments. Early customers report up to 700% improvement in sales meetings booked and pipeline — self-reported numbers. Coverage today spans Google Ads, LinkedIn Ads, and ChatGPT Ads, with Meta, Reddit, and Bing planned.
Two tells in one launch. First, ABM production is collapsing from weeks to minutes, which moves the paid-media job toward account understanding and away from campaign assembly. Second — and quieter — ChatGPT Ads now appears as a standard line item next to Google and LinkedIn in a B2B launch stack. Treat the 700% as a pitch; treat the channel list as the news.
9 SAS shipped 360 Marketing AI — ML models without waiting for data science
SAS announced SAS 360 Marketing AI on July 8, built to let marketers create, deploy, and scale machine-learning models without leaning on overstretched data science teams. The platform uses purpose-built guided workflows and customizable recipe templates for common marketing use cases, compressing the path from data to decision.
The “AI for marketers” fight is moving past content generation into modeling — propensity, churn, and LTV built inside the marketing department. That changes the hiring spec: the valuable skill becomes knowing which recipe fits which question and how to sanity-check the output, not writing the model. If your team can't articulate what a propensity score should drive, no template will save you.
10 LinkedIn is enforcing the pay rules the EU couldn't — most member states missed their own deadline
The EU Pay Transparency Directive's transposition deadline passed on June 7, and as of July only four countries — Italy, Slovakia, Lithuania, and Malta — have complete legislation in force, with Germany, France, Spain, and others running late. LinkedIn isn't waiting: it requires salary ranges on job postings targeting EU/EEA candidates, algorithmically demotes non-compliant listings, and badges them with a “Salary not disclosed” warning that also blunts paid job promotion.
The platform became the regulator — compliance is now a distribution factor, enforced by ranking rather than by fines. For employer-brand and recruitment marketers: post the range or pay more for less reach. And watch this pattern, because platform enforcement outrunning national law is exactly how AI-disclosure rules are likely to arrive too.