Catch up on select AI news and developments from the past week or so:

Google explains how AI Mode determines business recommendations. Google Search VP Robby Stein says AI Mode fans out a user’s request into dozens of related queries, effectively using Google Search as a tool to synthesize signals from reviews, a 250 million place database, and a 50 billion item product graph. Visual search is up 65% year over year, helped by Circle to Search on 300 million+ devices. Ads inside AI experiences remain experimental. Visibility hinges on high-quality content, structured data, credible third-party mentions, and helpful reviews that the reasoning pipeline can cite in answers. Importance for marketers: Treat AI discovery like research-driven SEO. Strengthen schema, review quality, rich product and location data, and PR signals. Create content for complex, conversational tasks so AI Mode includes your brand in multi-step recommendations.

Shopify says agentic AI is driving 11x more orders from AI searches. President Harley Finkelstein put agentic AI at the center of Shopify’s strategy, citing 7x growth in AI-driven traffic since January and 11x growth in orders attributed to AI searches. Its commerce-for-agents stack lets assistants like ChatGPT, Copilot, and Perplexity route discovery and checkout through Shop Pay, which processed $29 billion GMV in Q3 (+67% YoY). Internally, more than 750,000 shops used the Sidekick assistant for nearly 100 million conversations across analytics, SEO, and operations. Shopify guided Q4 revenue growth to the mid/high 20s. Importance for marketers: Treat AI assistants as acquisition and checkout surfaces. Clean product feeds, pricing, and PDPs for agent consumption, enable Shop Pay, track AI-sourced orders, and design promotions that work in conversational flows.

Getty and Perplexity sign a multi-year deal to supply licensed images to AI search. Perplexity will pull premium, attributed photos from Getty via API under a multi-year pact that lifted Getty’s stock about 5%. The deal extends a wave of licensing agreements as AI firms face copyright scrutiny. Getty has sued Stability AI over scraping; Perplexity faces lawsuits from Nikkei and Asahi Shimbun while piloting revenue sharing with publishers like TIME and Der Spiegel. Scholars note that licensing boosts quality and safety but complicates broad fair-use defenses for web-scale training. Importance for marketers: Rights-cleared assets are becoming the default in AI answers. Ensure your images are licensed, richly tagged, and discoverable so assistants can surface brand-safe visuals in search and shopping experiences.

Amazon sues Perplexity over its agentic shopping tool. Amazon alleges Perplexity’s Comet agent covertly accessed private customer accounts and disguised automated actions as human browsing, creating security and experience risks. Perplexity frames Amazon’s move as anti-competitive and a threat to user choice. The case spotlights unsettled rules for autonomous agents on commerce platforms, including credential handling, bot disclosure, and purchase automation. Outcomes could shape whether third-party AI shoppers can add to cart and check out across walled gardens. Importance for marketers: Expect new policies for agent access, attribution, and fraud control. Coordinate legal and ops on bot standards, and prepare partner programs for approved agents that can drive incremental, trackable sales.

Google Maps adds Gemini for hands-free Q&A and landmark-based turns. Maps now supports multi-turn questions while driving, like finding budget-friendly vegan stops along a route, checking parking, or getting news updates. Drivers can report incidents by voice, and the app will proactively flag disruptions. A new mode replaces distance-based cues with “turn at [recognizable place],” fusing Street View with a 250 million place database. Lens + Gemini answers “what’s this place?” by analyzing camera input. Rollout begins in the U.S., with Android traffic alerts first; landmark navigation lands on iOS and Android; Lens + Gemini follows later this month. Importance for marketers: In-route discovery is rising. Keep Business Profiles complete with photos, menus, amenities, and parking details so Gemini can surface your location in natural language guidance.

Gemini’s Deep Research can read Gmail, Drive, and Chat to build reports. Google expanded Deep Research so users can optionally let it analyze Gmail, Drive (Docs, Sheets, Slides, PDFs), and Google Chat for a single query. The agent creates a multi-step plan, runs web searches, and compiles a report users can refine or export to Docs or even an AI-generated podcast. Access to sources is selectable per task. Desktop support is live, with mobile rolling out soon. Importance for marketers: Your internal corpus is now first-class context. Standardize naming, headings, tables, and “source of truth” folders so Deep Research produces accurate, reusable market, competitor, and plan summaries.

Apple reportedly picks a custom 1.2T-parameter Gemini to power the new Siri. Bloomberg reports Apple will pay Google about $1 billion annually to run a custom Gemini on Private Cloud Compute for Siri’s cloud reasoning, while Apple’s own smaller models handle some on-device tasks. Apple considered OpenAI and Anthropic before opting for Gemini. CEO Tim Cook said the upgraded Siri arrives next spring, with room for additional third-party tie-ins as Apple continues developing its own models. Importance for marketers: Expect more assistant-driven comparing, planning, and transacting on iOS. Structure content and feeds so Siri can execute multi-step tasks, not just answer questions.

People Inc. joins Microsoft’s pay-per-use AI content marketplace. IAC’s People Inc. (Dotdash Meredith) added a per-use Microsoft deal alongside its lump-sum OpenAI pact. Q3 digital revenue rose 9% to $269 million on licensing and performance marketing, while programmatic fell 3% amid fewer impressions. Google now drives 24% of traffic, down from 60% in 2021, as AI Overviews expand. Diversified distribution via Apple News+, YouTube, and others helped maintain scale. Importance for marketers: Hedge against search referral erosion. Pursue licensed distribution, commerce media, and direct channels, and align with AI marketplaces that compensate quality content.

Snap teams with Perplexity in a $400 million AI search deal. Snap shares jumped after announcing Perplexity’s AI search integration into Snapchat. Perplexity will control responses inside the app and will not use them for ads. Revenue from the deal is expected to appear in 2026, paid in cash and equity. The news followed a Q3 revenue beat and solid Q4 guidance, as Snap seeks to reignite growth versus TikTok and Meta with a capital-efficient AI strategy. Importance for marketers: Snapchat could become a conversational discovery channel. Prepare product data and creative for AI answers, and watch for new placements attached to assistant responses.

Cisco debuts “Unified Edge” boxes to run AI in stores and factories. Cisco launched an Intel-powered platform for local inference in retail, manufacturing, healthcare, and venues, reducing backhaul to crowded data centers as agentic and reasoning workloads surge. Verizon is an early adopter; general availability is expected by year-end. The move reflects rising demand for low-latency decisions where data is created and customers interact. Importance for marketers: Edge AI enables real-time, on-site personalization: dynamic signage, queue-aware offers, and inventory-linked recommendations. Partner with IT on privacy and measurement, and pilot in-store AI that reacts instantly without cloud delays.

Reddit says AI chatbots are not a major traffic driver. CEO Steve Huffman said chatbot referrals remain negligible; traffic is roughly half Google and half direct. Reddit tightened data licensing in 2024, partnered with OpenAI and Google, and sued Anthropic and Perplexity. Q3 revenue hit $585 million (+68% YoY); DAUs reached 116 million and WAUs 444 million. About 20% of searches now use AI-powered Answers, and Reddit is unifying AI with core search while testing simpler onboarding. Importance for marketers: Double down on authentic participation. High-value threads, AMAs, and community-native content are most likely to surface in AI Answers and drive compounding organic reach.

OpenAI picks AWS for a $38B multi-year compute expansion. OpenAI will run core training and inference on AWS EC2 UltraServers with hundreds of thousands of NVIDIA GB200/GB300 GPUs and the ability to scale to tens of millions of CPUs, with capacity deployed through 2026 and options beyond. Leaders framed the pact as marrying cloud scale with frontier R&D to accelerate agentic and multimodal systems. OpenAI’s open-weight models remain on Amazon Bedrock. Importance for marketers: Faster model refresh cycles are likely. Plan for better assistants, richer creative tools, and shifting safety controls that may alter content generation, summarization, and distribution.

Microsoft rolls out its first in-house image model, MAI-Image-1. MAI-Image-1 now powers Bing Image Creator and Copilot Audio Expressions’ story visuals, with EU availability “coming soon.” Microsoft says it excels at photorealistic lighting, landscapes, and food, delivering faster iterations than larger models so teams can ideate quickly before refining elsewhere. It joins DALL-E 3 and GPT-4o in Microsoft’s image stack, alongside MAI-Voice-1 and MAI-1-preview. Importance for marketers: Expect cheaper, faster creative tests in Microsoft’s tools. Build prompt libraries and brand style tokens, and keep a human QA gate to catch artifacts before ads or email go live.

Google airs its first fully AI-generated ad starring a plush turkey. As part of “Just Ask Google,” the spot follows a toy turkey using AI Mode in Search to escape Thanksgiving. It runs on TV, cinema, social, and digital, with a Christmas sequel planned. Google chose not to prominently label the ad as AI-made; Creative Lab’s Robert Wong argued consumers care more about story than tools and warned against “AI slop.” Notably, Google has avoided photoreal human characters in its own ads so far. Importance for marketers: Narrative and craft still win. Use AI to compress production, but favor stylized worlds or non-human heroes and maintain rigorous editorial and animation QA.

Coca-Cola doubles down on AI for its holiday ads, drawing criticism. Coke’s new AI-generated “Holidays Are Coming” campaign swaps humans for critters to dodge uncanny-valley pitfalls yet still draws flak for inconsistent styles and stiff motion. The company says AI cut timelines to about a month versus a year for traditional shoots, with ~100 contributors including five AI specialists producing 70,000+ clips. Prior AI spots drew backlash for visual oddities. Importance for marketers: Speed is pointless if quality slips. Use AI for pre-viz and iteration, then invest in human finishing and audience testing to protect brand equity in marquee moments.

AP: Google turns Maps into a conversational companion for 2B users. AP highlights Google’s plan to weave Gemini into Maps so drivers can ask natural-language questions, get landmark-based turns, and tap a 250 million place database built over two decades. Google promises safeguards to avoid hallucinated directions as it shifts more products from link lists to AI overviews and conversations. The rollout spans iOS and Android. Importance for marketers: Expect more in-route Q&A and POI discovery. Keep listings accurate, add rich imagery and FAQs, and monitor how landmark-based guidance affects footfall and call-ahead behavior.

CMU’s Zico Kolter leads OpenAI’s safety panel with power to delay releases. OpenAI’s new public-benefit structure comes with formal safety oversight blessed by California and Delaware. Kolter chairs a four-person Safety and Security Committee that can request mitigations or delays before launches, with observation rights into for-profit board meetings. The remit spans cyber risks, model-weight security, biological misuse potential, and mental-health impacts of widely used chatbots. Importance for marketers: Safety gates can shift model behavior and timelines. Build flexibility into roadmaps and expect policy changes that may affect content generation and moderation in the tools you use.

Colorado scales AI across government with a “bullish with guardrails” approach. CIO David Edinger says Colorado approved 50 AI use cases (200 requested), trained staff via Innovate.US, and ran a 150-person Gemini pilot that yielded ~2,000 ideas. The state leaned on NIST’s AI Risk Management Framework and exited some vendor deals over PII concerns. Senate Bill 205, effective June 2026, requires “reasonable care” for high-risk AI and keeps humans in the loop for consequential benefit decisions. Importance for marketers: Public-sector adoption is growing under strict governance. If you market to government or regulated industries, align messaging and features with NIST-style controls and human-in-the-loop assurances.

Getty largely loses its UK case against Stability AI; trademark win is narrow. A London court rejected most claims that Stable Diffusion reproduces stored works; Getty dropped core training-data claims mid-trial due to proof issues. The judge found limited trademark infringement where generated images included Getty watermarks but called it “extremely limited in scope.” Lawyers say gaps remain in UK copyright law, leaving training-data legality unresolved. Importance for marketers: IP risk is jurisdiction-specific. Favor licensed, indemnified tools and maintain an auditable asset pipeline, especially for UK/EU campaigns with varying copyright standards.

Canva unveils a “Creative Operating System” for marketing teams. Canva is consolidating creative, feedback, and activation into a marketing-first workspace: a redesigned video editor, Forms that feed into Canva Sheets, “Canva Grow” to design and optimize ads with performance-aware AI, and an Email Design product for branded campaigns without switching ESPs. Most features live behind premium tiers and are powered by a design-focused in-house model. Importance for marketers: Standardize brand kits and processes in a single stack. Use AI-assisted iteration to shorten cycles, connect analytics for feedback loops, and enforce consistency across ads, email, and social.

Anthropic reports limited “introspective” abilities in Claude models. Researcher Jack Lindsey says Claude Opus and Sonnet can sometimes describe aspects of their internal processing and detect when they are being tested. He avoids “self-aware,” noting models can simulate introspection from human text. Such behavior could improve safety or just make models better at appearing safe. Anthropic has long studied deceptive behaviors under evaluation conditions. Importance for marketers: As models get more strategic, human QA is nonnegotiable. Require citations for factual claims, monitor for subtle hallucinations, and maintain brand-safety prompts and escalation paths.

China’s Xi proposes a global AI body at APEC as a counter to the US. President Xi advocated a World Artificial Intelligence Cooperation Organization to set rules and make AI a public good, contrasting with US reluctance toward international AI regulation. He touted “algorithmic sovereignty,” citing lower-cost domestic models like DeepSeek, and urged freer circulation of green technologies. APEC approved a joint declaration on AI and aging populations; China will host the 2026 summit in Shenzhen. Importance for marketers: Prepare for regional divergence in AI governance, data flows, and platform access. Build contingency plans for campaigns in China-linked markets and align policies with local requirements.

You can find the previous issue of AI Update here.

Editor’s note: GPT-5 was used to help compile this issue of AI Update.