Three major model releases in a single week, ads arriving in AI search results, and a landmark acquisition that signals where SEO is heading. Paul and Martin cut through the noise.

This week saw an unprecedented flurry of model releases: Google's Gemini 3 Pro stormed the benchmarks, only for Anthropic's Opus 4.5 to arrive days later with superior coding performance. We unpack what these advances mean for marketers, from Nano Banana Pro's near-production-ready image generation to Opus's orchestration capabilities for multi-agent workflows. Beyond the model wars, we examine Gartner's 2026 predictions—including a warning about critical thinking atrophy—and debate the implications of ads appearing in AI overviews. Adobe's $1.9 billion acquisition of SEMrush closes the episode, raising questions about the future of generative engine optimisation.

Key Takeaways

• Gemini 3 Pro benchmarks: Scored 45.8% on Humanity's Last Exam versus the previous high of 26.5%, and doubled ScreenSpot Pro accuracy to 72%—critical for agent-based computer use.
• Opus 4.5 efficiency: Achieved 81% accuracy on software engineering tasks using 12,000 tokens, while Sonnet 4.5 needed 22,000 tokens for 77% accuracy.
• Nano Banana Pro: Now handles up to 5 consistent characters and 14 objects in a scene, making product imagery workflows increasingly viable.
• Gartner's lazy thinking warning: Predicts 50% of organisations will require AI-free skills assessments by 2026 as critical thinking atrophies.
• B2B procurement shift: Gartner forecasts 90% of B2B buying will be AI-agent intermediated by 2028, pushing $15 trillion through agent exchanges.
• Ads in AI search: Google has begun placing sponsored links in AI overviews; reports suggest ChatGPT may follow.
• Adobe acquires SEMrush: The $1.9 billion deal validates generative engine optimisation as a strategic priority.

What to Do Now

• Test your own use cases: Benchmarks are directional; run Gemini 3 Pro and Opus 4.5 against your specific workflows before switching daily drivers.
• Experiment with Nano Banana Pro: If you need consistent product imagery, trial multi-character and multi-object prompts to assess production readiness.
• Protect critical thinking: Build review processes where humans form independent judgments before consulting AI—especially for strategic decisions.
• Prepare for agent-readable content: Convert key product documentation from PDFs to markdown to ensure AI procurement agents can parse your information.
• Monitor AI ad placements: Track emerging ad formats in Google AI overviews and prepare to test early for potential first-mover advantages.

Mentioned in This Episode

• Models: Gemini 3 Pro, Nano Banana Pro, Opus 4.5, Sonnet 4.5, Haiku 4.5, GPT 5.1, DeepSeek 3.2
• Platforms/Features: AI Overviews, AI Mode, Veo 3.1, Claude Code
• Companies: Google, Anthropic, OpenAI, Adobe, SEMrush, Microsoft, HSBC, McKinsey, Gartner
• Benchmarks: Humanity's Last Exam, ScreenSpot Pro, VendingBench, SWE-bench, GPQA Diamond

Subscribe and Share

New episodes of Artificially Intelligent Marketing drop weekly. Subscribe on your preferred platform and share your questions or AI experiments with us on LinkedIn—we may feature them in a future episode.

What's driving Google and Meta's record earnings? Hint: it's AI, not traditional ads.

In this episode, Paul and Martin unpack the latest earnings calls, HubSpot's strategic acquisition of X Funnel, and then venture into scenario planning for marketing's future. From business-as-usual to AI agent swarms running entire organisations, we map out what you should watch for and what you should do right now.

Key Takeaways

• Meta's AI-powered advertising suite is now handling $60 billion in annualised ad spend. That's real money, real results: 5% more time on Facebook, 10% on Threads, 30% growth on Instagram video.
• Google posted its first $100 billion quarter, driven by AI features in Search and YouTube. YouTube Shorts now earn more per watch hour than traditional in-stream ads.
• HubSpot acquired X Funnel to double down on Generative Engine Optimisation (GEO) — ensuring your brand shows up in AI chatbot answers. But don't chase FOMO. Focus on brand-building and content quality; those work across all search channels.
• Five plausible futures for marketing:
  1. Business as usual
  2. Niche AI agents handling specific tasks
  3. Hyper-personalisation at scale
  4. AI agents driving strategy
  5. Fully autonomous agent swarms Which scenario emerges depends on technology progress, regulation, and how quickly organisations can rethink workflows.
• The real bottleneck isn't technology — it's human change. AI-first startups will outcompete legacy organisations unless incumbents commit to wholesale rethinking. Disruption ahead.
• Customer preference trumps efficiency. If your audience values human interaction and human-crafted outputs, that changes everything. Personalisation won't win if customers prefer the human touch.

What to Do Now

• Invest in AI skills training for yourself and your team. The saying holds: marketers aren't replaced by AI; they're replaced by marketers who use AI.
• Get your customer data in order. Clean, structured, first-party data is essential — especially if privacy regulations tighten. If an AI agent needs to draw insights, it needs good data to work with.
• Establish an AI ethics and usage charter. Give customers confidence. Guide employees on responsible AI use.
• Build human-in-the-loop QA processes. Identify where you can safely automate and where humans must review. As niche agents emerge, you'll know exactly where to deploy them.
• Run scenario planning for your business. Use ChatGPT or Claude with your business context to model which scenario matters most to you. Don't just accept our five; build your own.

Mentioned in This Episode

• Meta: AI recommendations; Vibes (image generation); Advantage+.
• Google: AI Overviews; Performance Max; YouTube Shorts.
• HubSpot: X Funnel acquisition; Loop Marketing; GEO.
• Tools: Zapier; Claude; ChatGPT; Canva; HubSpot; Salesforce.
• Reference: The Last Economy by Emad Mauck.

New episodes every week on Artificially Intelligent Marketing. For questions, use cases, or scenario disagreements, find us on LinkedIn or reply to our latest episode post.

Share this with a colleague who needs to hear it.

What actually matters now in AI for marketers—beyond the hype. Paul and Martin unpack the week’s launches from Anthropic and Google, new Microsoft–Claude tie-ins, and the UK government’s civil-service AI literacy push. They also tackle the practical question: should you chase “AI SEO” today, or double-down on timeless content and distribution?

Top takeaways

• AI visibility ≠ old SEO. Generative search is highly personalised; “rank tracking” for LLM answers is shaky. Invest in high-quality, well-structured content and robust distribution over gimmicky “get featured in ChatGPT” checklists.

• Anthropic’s direction: Haiku 4.5 offers fast, low-cost coding and general tasks; Claude Skills hint at SOP-driven, tool-aware assistants and a future marketplace.

• Enterprise shift: Claude is deepening its Microsoft 365 integration (SharePoint, Outlook, OneDrive, Teams), making business context more accessible to assistants.

• Video creation reality: Google’s Veo 3.1 and Flow expand control (reference frames, storyboard timelines). Quality still depends on human editing judgement—AI is an ingredient, not the whole recipe.

• AI Studio cleanup: Google has unified its playground and simplified keys/settings, making it easier for newcomers to explore models and tune temperature/controls.

• Public sector skills: The UK’s “One Big Thing: AI for All” aims to boost civil-service confidence via short modules and department case studies—baseline literacy first, then role-specific adoption.

• Training that sticks: Start with common foundations (what LLMs are/aren’t), then move to function-level workflows. Create “champion” groups and a light governance rhythm so updates translate into process change.

Chapters

00:00 Harrison recap; AI in sales needs thoughtful application, not blind scale.

03:00 Brand presence in LLM answers: content + distribution over hacks.

10:00 Why LLM personalisation breaks classic rank tracking.

15:00 Anthropic: Haiku 4.5 and Claude Skills; SOPs for agentic work.

21:00 Claude ↔ Microsoft 365: the enterprise context advantage.

25:00 Google Veo 3.1 & Flow: creative control vs. “one-click” video.

34:00 Google AI Studio: unified playground; why marketers should tune temperature.

38:00 “Zero-to-magic” tease; what a no-code app path might mean.

40:00 UK “One Big Thing—AI for All”: literacy at scale; early gov use cases.

42:00 Good AI training design: baseline, then function-specific change.

55:00 Reality check: useful today, but still needs steering and review.

Practical guidance

• If you’re a marketer: Treat AI outputs as drafts. Edit for voice, accuracy, and distinctiveness. Measure distribution as much as production speed.

• If you’re a leader: Provide one sanctioned assistant (Copilot/Gemini/Claude), set guardrails, and fund a small “AI council” to turn releases into workflow changes.

• If you’re experimenting with video: Use Flow (or your editor) to storyboard, trim, and mix assets. Don’t ship raw generations.

Mentioned in this episode

• Anthropic: Haiku 4.5; Claude Skills; Microsoft 365 integrations

• Google: Veo 3.1; Flow storyboard editor; AI Studio updates

• UK Government: “One Big Thing 2025—AI for All” initiative

• Discussion: training that works vs. one-and-done workshops

Subscribe & follow:

New episodes weekly. Share your questions or use-cases for a chance to be featured.