Platform Digest: Gemini 4 Carbon, AI Mode Goes Global & Meta's GEM

Three separate moves this week — one from Google's model lab, one from Google's ads team, one from Meta's engineering floor — share a common thread: the underlying AI deciding what gets shown, to whom, and when is getting substantially more powerful, and marketers have less visibility into it than ever.

Google's Gemini 4 Carbon: The Most Capable Model It Has Ever Built

On October 28, Google announced Gemini 4 Carbon, describing it as its strongest model to date. Built on top of Gemini 4 Argon — which only launched two weeks earlier — Carbon is designed for cybersecurity work and what Google calls "long-horizon" engineering tasks: sustained, multi-step problems that can take days to resolve. Google's own internal benchmarks put Carbon at 88.2% on vulnerability discovery, 54.8% on exploit testing, and 71.0% on code patching, making it more capable than any previous Gemini release on those dimensions.

Carbon is not publicly available yet. Access begins with a narrow group of vetted cyber defenders through Google DeepMind's Fairwind Program, with no confirmed date for general API access and no published context window limit. Standard Gemini Pro and Ultra subscriptions still show Gemini 4 Argon. Google has said developers, businesses, and consumers will get access "as soon as possible" — which is, pointedly, not a date.

For most marketing teams, Carbon offers nothing actionable this week. What it does signal is harder to ignore: Google is shipping frontier models at a pace that was genuinely unthinkable twelve months ago. Teams building AI-assisted workflows — content, competitive research, audience analysis — should keep a close eye on Gemini API access announcements, because Carbon's capabilities will eventually reach those surfaces, and the window between announcement and widespread adoption is shrinking with every release.

Google AI Mode Ads Land in India — and Ten Other Markets

Also on October 28, Google's VP of Search Ads Shashi Thakur announced that ads in AI Mode are expanding to ten new countries: Australia, Canada, India, Kenya, Malaysia, New Zealand, Nigeria, Pakistan, the Philippines, and Singapore. The US already had them. Brazil, Mexico, and the UK are queued for Q4 2026.

Google also launched Multiturn Ads in the United States — a format that places sponsored results inside follow-up queries within an ongoing AI Mode conversation. Queries in AI Mode average significantly longer than standard searches. A user asking three follow-up questions about home loan options or a financial product gives the ad system multiple points of entry, not just one. Google's early data suggests users find the ads relevant, though independent verification of that claim is thin.

India's inclusion is the one to act on immediately. AI Mode is already seeing significant usage in Indian markets, and advertisers running Search and Performance Max campaigns will start seeing impressions on this surface without any separate opt-in. Pull your placement reports, check where impressions are being served, and audit whether your ad copy and landing pages actually hold up in a conversational context — a two-line text ad written for a traditional SERP often reads poorly inside a multi-sentence AI-generated answer. BFSI and real estate brands feel this most acutely, given how many comparison queries live in those categories.

Meta's GEM Rewires Every Ad Auction — and Threads Is Now a Default Placement

Two significant Meta changes landed in the same week, and both point the same direction: more automation, less manual control.

Meta published results from its Generative Ads Model (GEM), which the company calls its largest AI recommendation system ever built. GEM now operates as a single central foundation model across Facebook, Instagram, Reels, and Stories, replacing what were previously separate prediction systems per surface — and no advertiser opt-in was required. Meta's reported figures are 18.9% more conversions and 12% higher CTR across its ad surfaces. Independent analysts have flagged that the 18.9% figure is a platform-wide average across billions of impressions, not a per-account guarantee, and Meta hasn't published the comparison baseline. Take it as a directional signal rather than a promised uplift for your specific account.

Threads ads, meanwhile, became the default placement in all campaigns globally, with full rollout completing the week of October 26. Every active Meta campaign now automatically allocates some budget to Threads. CPMs there are running 20–40% below Instagram feed, which looks attractive — but check your placement breakdowns before crediting Threads with results that may be coming from elsewhere. D2C brands with tight ROAS targets should pull Threads out separately in reporting and make a deliberate call on whether to keep or exclude it, rather than letting the default run unchecked.

One more thing worth flagging: Meta's updated Commercial Terms — originally due October 30 — have been pushed to November 20 after complaints were filed with EU and UK regulators. If you haven't reviewed what those terms say about AI-generated content liability and Meta's right to crawl your website for model training, November 20 is now your hard deadline.

The Week's Throughline

Google shipped its most powerful model yet and put ads inside conversational search across a dozen markets in a single day. Meta's unified AI engine is now running every auction on the platform without surfacing much of its logic to advertisers. The platforms are consolidating control over delivery, placement, and optimisation into AI systems that operate largely out of view — and that trend is accelerating, not plateauing. Auditing placements, tracking new surfaces, and keeping measurement clean is less glamorous than chasing the latest model release, but it's the work that actually protects campaign performance when the ground keeps shifting.

If you're running paid media in BFSI, real estate, or ecommerce and want a hard look at how this week's changes are showing up in your specific campaigns — placements, auction dynamics, AI Mode impression share — that's exactly the kind of audit Transformics runs. Get in touch and we'll dig into the data with you.