| |

This Week in AI in Marketing (23rd Mar 26)

ByAnn Stanley
By Ann Stanley | 23 March 2026 | 21 min read

Claude Launches 3 New Products, GPT-5.4 Gets Hands, Meta Bans Claude Users, & 66% of CEOs Freeze Hiring

This was a week of power moves and unintended consequences. Anthropic shipped three major products in seven days — Claude Marketplace, Claude Dispatch, and Claude Code Channels — with Channels already being called an “OpenClaw killer.” OpenAI launched GPT-5.4 with native computer control, and Nvidia entered the AI agent race with NemoClaw. Meanwhile, Meta started banning ad accounts that use Claude via MCP — weeks after embedding its own $2 billion Manus AI acquisition into Ads Manager. In marketing, Google quietly added AI voiceovers to Performance Max videos (opt-out deadline already passed), and GEO is now officially the fastest-growing discipline in digital marketing. On the management side, Fortune reports that 66% of CEOs are freezing hiring while pouring billions into AI — and Harvard Business Review asks whether that strategy can actually work.

Table of Contents

AI News, Tech & Tools AI in Marketing AI in Management AI for Sectors & Industries Key TakeawaysFrequently Asked Questions Conclusion

AI News, Tech & Tools

GPT-5.4 Launches with Native Computer Use

Source: openai.com | 5 March 2026OpenAI launched GPT-5.4 on 5 March with a headline feature that changes the game: native computer use. The model can now read your screen, click, type, scroll, and navigate applications autonomously — turning it from a text generator into a genuine agentic tool. Three variants ship at launch: Standard, Pro, and Thinking. The model brings a 1M token context window, 33% fewer factual errors across benchmarks, and 47% token efficiency gains. On OSWorld-Verified, it achieves a 75.0% success rate — surpassing human performance benchmarked at 72.4%.Why it matters Computer use is the bridge between “AI that advises” and “AI that executes.” For agencies and marketing teams, this means GPT-5.4 can now fill forms, run reports, navigate ad platforms, and execute multi-step workflows that previously required human hands. The competitive pressure on every other AI provider just intensified dramatically.

Anthropic Launches Claude Marketplace for Enterprise

Source: venturebeat.com | digitalcommerce360.com | 16 March 2026Anthropic launched Claude Marketplace, allowing enterprise customers with existing API spend commitments to purchase third-party applications built on Claude from partners including GitLab, Harvey, Lovable, Replit, Rogo, and Snowflake. The model is simple: enterprises can redirect part of their existing Anthropic budget toward these tools without Anthropic taking a commission on transactions. The marketplace is currently in limited preview.Why it matters This is Anthropic’s play to lock enterprise customers into its ecosystem. By making it easier to buy Claude-powered tools with money already committed to Anthropic, they are building a walled garden that competes directly with Microsoft’s Copilot marketplace and Salesforce’s Agentforce. For agencies evaluating AI stacks, the platform you commit budget to increasingly determines which tools you can access.

Claude Dispatch: Control Your Desktop AI from Your Phone

Source: mlq.ai | geeky-gadgets.com | 17 March 2026Anthropic launched Claude Dispatch on 17 March as a Cowork feature that lets Max subscribers control a Mac-based AI session from their mobile device. You can assign complex tasks to Claude running on your desktop, then monitor progress and retrieve completed results from your phone. Pro plan access followed within days.Why it matters Dispatch turns Claude into an always-on digital worker. You brief it from your phone over coffee, and it executes on your desktop — processing files, running analysis, building documents. Early testing found it works roughly half the time, which is honest but highlights that agentic reliability remains the real bottleneck across the industry.

Claude Code Channels: Anthropic Ships an OpenClaw Killer

Source: venturebeat.com | techzine.eu | 20 March 2026Anthropic launched Claude Code Channels on 20 March, a feature widely described as an “OpenClaw killer.” Channels lets you connect Claude Code to Telegram or Discord, so you can message instructions from your phone and have Claude execute them on your local machine. Everything runs locally — nothing in the cloud — and Claude replies back through the same chat with progress updates and results. It can reply, react with emoji, edit its own messages, and on Discord, fetch channel history and download attachments. The feature is built on Anthropic’s Model Context Protocol (MCP) and ships with official plugins for both platforms, with community connectors possible via the open plugin architecture.Why it matters This is Anthropic’s direct answer to OpenClaw, and it arrives just days after Nvidia launched NemoClaw with the same goal. The difference: Claude Code Channels works within an existing commercial product with enterprise-grade security, while OpenClaw and NemoClaw are open-source frameworks requiring more setup. For developers and technical marketers already using Claude Code, Channels removes the friction of switching to a separate agent platform. Combined with Dispatch (phone-to-desktop control) and the Claude Marketplace announced the same week, Anthropic shipped three major features in seven days — a clear signal they are racing to own the agentic workflow space.

OpenAI Admits Prompt Injection May Never Be Fully Solved

Source: openai.com | venturebeat.com | March 2026OpenAI shipped a security update to ChatGPT Atlas after internal red-teaming uncovered a new class of prompt-injection attacks. The update includes a newly adversarially trained model and strengthened safeguards. But here is the real story: OpenAI officially acknowledged that prompt injection — where malicious instructions hidden in web content trick AI agents into unauthorised actions — is “unlikely to ever be fully solved.”Why it matters As AI agents gain the ability to browse the web, send emails, and control computers, the attack surface grows exponentially. An attacker could embed instructions in an email that trick your AI agent into forwarding sensitive documents. Any business deploying agentic AI needs a security review that accounts for this structural risk — not a patch, but a permanent constraint.

Microsoft Copilot Cowork (Wave 3): AI That Works for Hours, Not Seconds

Source: microsoft.com | venturebeat.com | 9 March 2026Microsoft announced Wave 3 of Microsoft 365 Copilot on 9 March, introducing Copilot Cowork — built in partnership with Anthropic. Unlike standard Copilot prompts that return instant answers, Cowork handles long-running, multi-step tasks: breaking down complex requests, reasoning across tools and files, and carrying work forward for minutes or hours with visible progress. Agent 365, the governance layer, will be generally available from 1 May at $15 per user per month.Why it matters The Anthropic partnership is significant — Microsoft is using the technology behind Claude Cowork inside its own flagship product. For the millions of businesses already paying for Microsoft 365, this brings enterprise-grade AI agents into their existing workflow without switching platforms. The $15/user Agent 365 pricing will force every competing vendor to justify their agent governance costs.

Nvidia Launches NemoClaw: Open-Source Enterprise AI Agent Platform

Source: cnbc.com | techcrunch.com | 10-16 March 2026Nvidia announced NemoClaw, an enterprise-ready, open-source AI agent platform that adds security, privacy controls, and policy enforcement to the popular OpenClaw framework. It installs NVIDIA Nemotron models and the OpenShell runtime in a single command, and is hardware-agnostic — running on chips from Nvidia, Intel, AMD, and others. CEO Jensen Huang has reportedly been pitching NemoClaw to Salesforce, Cisco, Google, Adobe, and CrowdStrike.Why it matters OpenClaw’s biggest problem has always been security. NemoClaw solves that with enterprise-grade controls baked in, which could make it the default platform for companies that want locally-controlled AI agents without exposing sensitive data. Nvidia is positioning itself not just as the chip supplier for AI, but as the infrastructure layer for the entire agentic ecosystem.

Perplexity Launches “Computer” — Local and Cloud AI Agents

Source: venturebeat.com | axios.com | 11 March 2026Perplexity announced two AI agent products at its Ask 2026 developer conference: “Personal Computer” runs locally on a dedicated device like a Mac mini with full file and app access, while “Perplexity Computer” operates in the cloud. The enterprise version orchestrates 20 AI models with Slack integration and Snowflake connectors, taking direct aim at Microsoft Copilot and Salesforce Agentforce.Why it matters Perplexity — a company without its own frontier models — is betting that orchestration beats ownership. If they can make 20 models work together better than any single provider, the “best model” race becomes less relevant than the “best integration” race. This is a direct challenge to the platform lock-in strategies of OpenAI, Anthropic, and Google.

Meta Acquires Manus for $2B, Bans Accounts Using Claude via MCP

Source: almcorp.com | socialmediatoday.com | adage.com | March 2026Meta acquired Manus AI in December 2025 for more than $2 billion — its third-largest deal ever — and by February had embedded it directly into Ads Manager. Manus can automate tasks including report building, audience research, and campaign management within the ad workflow. But the flip side is causing alarm: reports surfaced across LinkedIn of media buyers having their ad accounts banned for using Claude via MCP (Model Context Protocol) to manage their Meta campaigns. The irony is that Manus itself is powered by Anthropic’s Claude models under the hood.Why it matters This is classic platform lock-in. Meta acquires an AI tool, integrates it into its walled garden, then punishes advertisers who use competing AI tools to access the same platform. If you are currently using third-party AI automation to manage Meta ads — via Claude, custom scripts, or MCP integrations — you need to audit your access methods immediately. The risk of account suspension is real and the appeals process is not fast.

AI in Marketing

Google Expands Ads in AI Overviews to More Markets

Source: searchengineland.com | March 2026Google is expanding ads within AI Overviews from mobile-only US availability to desktop and multiple global markets. These placements integrate ads directly into the AI-generated summaries that appear at the top of search results, fundamentally changing where paid visibility sits in the search experience.Why it matters AI Overviews now appear on approximately 48% of search queries and have been shown to reduce click-through rates for top-ranking organic results by up to 58%. As Google monetises this space with ads, the economics of both SEO and PPC shift. Brands that are not visible in AI Overviews — either through citations or ads — are losing ground to those that are.

Google Auto-Adds AI Voiceover to Performance Max Videos — Opt-Out Deadline Has Passed

Source: almcorp.com | ppc.land | March 2026Google rolled out automatic AI-generated voiceovers for Performance Max video ads that do not already contain a voice track. The feature uses Google’s AI voice models to create spoken audio from headlines and descriptions in the campaign’s asset group. It was enabled by default, with the opt-out deadline set for 20 March — which has now passed. Advertisers in regulated categories (financial services, healthcare, pharmaceuticals) face meaningful compliance risks, as Google’s AI has no awareness of regulatory disclosure requirements.Why it matters If you run Performance Max campaigns and missed the 20 March deadline, check your video assets now. Google is adding AI-generated narration to your ads without explicit consent. For regulated industries, this is not just an inconvenience — it is a compliance risk. Check your Video Enhancement Controls settings immediately. Source: almcorp.com | March 2026Google Ads Editor version 2.12 arrived in March with fifteen new features spanning Performance Max, Demand Gen, Search, Shopping, and cross-account campaign management. Updates include improved PMax asset group management, new Demand Gen creative controls, and streamlined total budget workflows.Why it matters For PPC managers handling multiple accounts, this is a practical quality-of-life update. The cross-account improvements and Demand Gen controls in particular reduce the manual overhead of managing campaigns at scale.

GEO Set to Eclipse SEO as the Fastest-Growing Digital Marketing Discipline

Source: aimagazine.com | searchengineland.com | financialcontent.com | 19 March 2026Generative Engine Optimisation (GEO) has officially emerged as the fastest-growing discipline in digital marketing. Multiple sources this week confirmed that GEO — optimising content for citations within AI-generated responses rather than traditional search rankings — is poised to overtake traditional SEO in 2026. The key difference: SEO optimises for clicks from search engine results pages, while GEO optimises for citations within AI-generated answers. By 2028, an estimated $750 billion in US revenue will flow through AI-powered search.Why it matters The ranking number one for a keyword no longer guarantees traffic if an AI Overview answers the query first. Content extractability — structuring every section so AI models can isolate and quote specific claims — is now a core competency. If your content strategy still centres on keyword density and backlinks alone, you are optimising for a shrinking channel.

AI Influencers and Digital Twins: What It Means for Marketers

Source: marketingweek.com | March 2026Meta and TikTok are pushing AI creators hard. Meta now enables Instagram influencers to create AI versions of themselves, and 85% of creators surveyed say they are open to creating a digital twin with a brand for marketing purposes. The appeal is clear: digital twins can produce content around the clock, respond to followers, and scale brand partnerships without the creator being physically present.Why it matters Influencer marketing is about to bifurcate. Brands will need to decide whether they are partnering with a human creator or licensing their digital twin — and the pricing, authenticity, and legal frameworks for each are completely different. Expect disclosure rules and audience trust to become the defining battleground. Source: socialbee.com | 13 March 2026TikTok announced three new AI-powered features for TikTok Shop: AI Dubbing for videos, AI Fashion Video Maker, and List with AI. LinkedIn now allows users to search for jobs by describing the role they want in natural language rather than using keyword filters. Both platforms continue to embed AI deeper into their core experiences.Why it matters TikTok’s AI tools lower the barrier to entry for Shop sellers — particularly the AI Dubbing feature, which opens cross-border selling to creators who previously could not localise content. LinkedIn’s natural language job search is another signal that traditional keyword-based interfaces are being replaced by conversational AI across every platform.

EU Council Agrees to Streamline AI Act Rules

Source: consilium.europa.eu | 13 March 2026The EU Council proposed adjusting the timeline for high-risk AI system rules by up to 16 months, delaying full application until the Commission confirms standards and tools are available. A new provision was added prohibiting AI practices related to generating non-consensual sexual and intimate content or child sexual abuse material. The AI Act becomes fully applicable on 2 August 2026, with a second draft code expected in March and the final code by June 2026.Why it matters The timeline extension gives businesses more breathing room, but the direction is clear: regulation is coming. UK businesses serving EU customers need to start mapping their AI usage against the Act’s requirements now, not when enforcement begins.

AI in Management

66% of CEOs Are Freezing Hiring While Betting Billions on AI

Source: fortune.com | 18 March 2026A survey of more than 350 public-company CEOs and investors managing $19 trillion in assets found that 66% plan to freeze or cut hiring through the rest of 2026. Corporate America has eliminated more than 1.17 million jobs under the logic that excess labour must be cut to fund the future of AI. But Fortune calls this a costly miscalculation: CEOs are buying powerful computational engines while cutting the middle-management and HR functions required to implement, govern, and scale them.Why it matters You cannot automate what you have not first understood. The companies cutting management layers to fund AI are often removing the very people who understand the workflows AI is supposed to improve. A 2026 Gallup survey found that only 12% of workers report using AI daily despite widespread enterprise deployment. The gap between buying AI and actually using it remains enormous.

The “Last Mile” Problem Slowing AI Transformation

Source: hbr.org | March 2026Harvard Business Review reports that few companies have been able to fundamentally change their operating and business models around AI, despite having initiated hundreds of pilots and provided widespread access to tools like Copilot and ChatGPT. The primary obstacle is rarely model quality or data availability — it is the “last mile” where technical capability must meet organisational design. Gartner research finds that only one in 50 AI investments delivers transformational value, and only one in five delivers any measurable return.Why it matters This validates what many agency leaders already suspect: the AI tool is rarely the bottleneck. The bottleneck is redesigning processes, roles, and incentives around AI capabilities. Companies that focus exclusively on tool procurement without investing equally in change management will continue to see disappointing returns.

Who in the C-Suite Should Own AI?

Source: hbr.org | March 2026HBR tackles the governance question that most leadership teams are still debating: who actually owns AI strategy? The article argues that AI ownership cannot sit solely with the CTO or CIO because AI’s impact spans marketing, operations, HR, and finance. Nor can it be delegated to a standalone Chief AI Officer without cross-functional authority.Why it matters For mid-sized businesses and agencies, this is a practical question. If nobody owns AI, everybody experiments and nobody scales. The emerging consensus is that AI needs a cross-functional steering group with executive sponsorship — not another silo.

How Deep Industry Research Agents Can Change Your Organisation

Source: hbr.org | March 2026Corporate leaders often chase breakthrough innovations from AI, but HBR argues the faster and more reliable payoff lies in using research agents to radically improve services productivity by eliminating low-value work. Deep industry research agents can synthesise competitive intelligence, regulatory changes, and market data in hours rather than weeks.Why it matters This is directly relevant to agencies and consultancies. The first use case for AI agents is not replacing client-facing work — it is eliminating the research drudgery behind it. Teams that deploy research agents now will deliver faster, more data-rich recommendations while competitors are still compiling spreadsheets manually.

The Quiet Panic in HR: Chief People Officers Are Losing Sleep Over AI

Source: webpronews.com | March 2026Chief People Officers are caught between CEO demands to deploy AI rapidly and the reality that most workforces are unprepared. Gartner predicts that through 2026, 20% of organisations will use AI to flatten their structures, eliminating more than half of current middle management positions. Yet the education gap remains the number one barrier: companies say training — not role redesign — is their primary talent adjustment for AI.Why it matters The management layer being cut is often the same layer responsible for implementing change. Organisations that flatten too aggressively risk creating an “execution gap” where senior leaders set AI strategy and junior staff are expected to deliver it, with nobody in between to translate, train, or troubleshoot.

AI for Sectors & Industries

The Agentic Shopping Race: Shopify, OpenAI, and Google All Launch AI Commerce Tools

Source: techcrunch.com | cnbc.com | blog.google | 16-20 March 2026Three major moves in agentic commerce this week. Shopify president Harley Finkelstein said the company is preparing for AI shopping agents to serve as a new entry point for ecommerce. OpenAI is pivoting away from instant checkout in ChatGPT, instead working with retailers including Etsy, Walmart, and Amazon to create dedicated apps that reroute users to the retailer’s own site. And Google announced the Universal Commerce Protocol (UCP), a new open standard that will power checkout within AI Mode in Search and the Gemini app.Why it matters The question is no longer whether AI agents will shop on behalf of consumers — it is whose agent becomes the default. Google’s UCP is an open standard, which could become the HTTP of agentic commerce. OpenAI’s retreat from direct checkout suggests that retailers want control of the transaction, not just the discovery. Ecommerce brands need to start optimising product data for agent readability now, not when these features go mainstream.

Retail’s Risky AI Commerce Bet

Source: retaildive.com | March 2026Retail Dive warns that the rush toward AI-powered commerce carries significant data and market share risks. As retailers integrate AI agents into their shopping experiences, they are sharing product data, pricing, and inventory information with platforms that could ultimately disintermediate them. The article examines how AI intermediaries could capture customer relationships that retailers have spent decades building.Why it matters This is the tension at the heart of agentic commerce. Retailers need AI agents to reach customers, but every agent integration hands more data and customer access to the platform provider. Brands that rely entirely on third-party AI agents for discovery risk becoming invisible when those agents change their algorithms or priorities.

NHS Launches £6M AI Screening Platform Across 64 Trusts

Source: gov.uk | nihr.ac.uk | March 2026The UK government announced a new AI research screening platform called AIR-SP, backed by nearly £6 million in funding. It will enable NHS trusts to join trials of AI in screening to speed up diagnosis, with Annalise.ai’s chest X-ray platform already deployed across 64 trusts checking for more than 120 conditions including cancers, pneumonia, and collapsed lungs. The platform is expected to be operational for research purposes in 2027.Why it matters This is one of the largest public-sector AI deployments in the UK. Healthcare is where AI’s impact on human outcomes is most direct and measurable. The £6 million investment signals that the NHS is moving past pilot stages into systematic AI integration — and the regulatory framework being developed alongside it will likely influence AI governance across other UK public services. Source: lawnext.com | artificiallawyer.com | dentons.com | 3-16 March 2026AI adoption among legal professionals has more than doubled in a year, with approximately 70% now using AI tools. But there is a striking gap: individual practitioners are adopting far faster than firms. Law firms are held back by governance concerns, billing model conflicts, and partnership structures that reward hours billed rather than efficiency gained. Meanwhile, landmark AI rulings are beginning to establish legal precedents for AI-generated content and liability.Why it matters The legal sector mirrors what is happening across professional services: individuals adopt AI fast, but organisations lag. For any business that relies on legal services, this creates an arbitrage opportunity — smaller, AI-enabled practitioners can now deliver work faster and cheaper than large firms still debating their AI policy.

UK Government Hires Google to Build AI Planning Tool, but Ranks Just 6th in Public Sector AI Adoption

Source: dezeen.com | openaccessgovernment.org | 11 March 2026The UK government hired Google to develop an AI planning tool aimed at halving the processing time for planning applications. Yet the Public Sector AI Adoption Index 2026 ranks the UK just 6th globally. Meanwhile, the government backed away from an opt-out copyright law for AI training, a decision that will affect how AI companies can use published content for model development.Why it matters The UK is investing in specific AI applications (planning, NHS screening) while struggling with broader adoption and regulatory clarity. The copyright decision is particularly significant for publishers, content creators, and marketing agencies — it determines whether your published content can be used to train AI models without explicit permission.

Key Takeaways

  • GPT-5.4’s native computer use turns AI from an adviser into an executor — every AI provider now faces pressure to match this capability.
  • Meta is banning ad accounts that use Claude via MCP while pushing its own Manus AI — audit your Meta automation tools immediately.
  • Google’s AI voiceover is now auto-enabled on Performance Max videos — check your Video Enhancement Controls if you missed the 20 March opt-out deadline.
  • GEO is officially the fastest-growing digital marketing discipline — optimise content for AI citations, not just search rankings.
  • 66% of CEOs are freezing hiring to fund AI, but only 12% of workers use AI daily — the implementation gap is the real crisis.
  • The agentic shopping race is accelerating: Google’s Universal Commerce Protocol could become the open standard for AI-powered checkout.
  • NHS is deploying AI screening across 64 trusts — the UK’s largest public-sector AI rollout signals systematic adoption beyond pilots.

Frequently Asked Questions

Should I be worried about Meta banning accounts for using AI automation?

Yes, if you are using third-party AI tools (particularly Claude via MCP) to manage Meta ad accounts. Reports from media buyers confirm account suspensions. Meta wants you to use its own Manus AI integration. Review your automation setup and consider switching to Meta’s native tools for any account-level management.

What is the difference between SEO and GEO?

SEO optimises for clicks from traditional search results pages. GEO (Generative Engine Optimisation) optimises for citations within AI-generated responses — the summaries produced by Google AI Overviews, ChatGPT, Perplexity, and other AI search tools. Both matter, but GEO is growing faster because AI-generated answers are replacing clicks for an increasing number of queries.

How should businesses respond to the CEO hiring freeze trend?

Do not cut the people who understand your workflows before you have documented and automated those workflows. The companies seeing AI ROI are investing in training and process redesign alongside the technology — not instead of the workforce. Harvard Business Review’s “last mile” research shows that organisational design, not model quality, is the real bottleneck.

Conclusion

This week’s theme is convergence: AI agents are arriving across every platform (GPT-5.4, Copilot Cowork, NemoClaw, Perplexity Computer), AI is reshaping every marketing channel (GEO, Performance Max voiceovers, agentic shopping, AI influencers), and management teams are grappling with the human side of AI transformation (hiring freezes, last-mile problems, governance gaps). The businesses that will thrive are those treating AI as an organisational change programme, not just a technology purchase.Three things to do this week: check your Performance Max video settings for unwanted AI voiceovers, audit any third-party AI tools accessing your Meta ad accounts, and start a conversation with your leadership team about who actually owns AI strategy in your organisation.Need help adapting your AI strategy? Contact the Anicca team for expert guidance.This roundup is compiled from publicly available sources using AI-assisted research. While we review every article for accuracy, our analysis reflects our interpretation of the original reporting. We strongly encourage readers to click through to the original sources linked throughout this post for full context and detail. If you spot anything that needs correcting, please let us know.
FREE GUIDE

Free download

The Complete Claude Code Guide

67 pages for marketers, managers and business leaders. No coding required.

Ann Stanley
Ann StanleyFounder & CTO, Anicca Digital
Why it matters Understanding Claude Code Getting set up Working with Claude Skills Business use cases Rollout & governance
Download → anicca.co.uk/claude-code-guide
The Complete Claude Code Implementation Guide ebook cover
The Thursday AI Club logo

Thursday AI Club

A hands-on AI club for marketers and managers. Fortnightly 3-hour sessions: Hour 1 is an open Q&A on any AI question; Hours 2-3 split into a workshop track (for newbies) and an advanced track (live demos and deeper questions).

Led by Ann Stanley, James Allen.

Workshop recordings library
Skills and prompts to copy
Private community channel
Secret Agents community access
Full-day hackathon every quarter
Cancel any time

Membership: £40/month or £400/year

Join the Thursday AI Club →

Similar Posts