How Companies Run Marketing on AI in 2026

Picture a marketing team’s Monday in 2026. The month’s campaign drafts are already written, the week’s ad variants already built, last week’s performance already summarized in plain language, and the lapsing-customer email already drafted and waiting for a human to approve. None of that is a forecast. The pieces are live right now at Unilever in London, at Dentsu in Tokyo, and at agencies in Oslo and São Paulo.

Marketing was the first business function AI rebuilt, because so much of it is the production of language and images, which is exactly what these models do. The result is a function that looks unchanged from the outside — campaigns still ship, emails still send — while nearly every task inside it has been rewired.

The short version: Companies now run marketing on AI across the whole stack — content, SEO and AI-search, paid media, lifecycle, creative production, and analytics — and the instructive examples are global. This briefing tours real, sourced campaigns from Unilever, L’Oréal, Mondelez, Dentsu, Mercado Libre, and agencies in Norway and Brazil; it includes a “backlash file” of the public failures, because the failures are where the real lessons are; and it ends with the operating system a marketing team needs — the system behind our forthcoming Daily Marketing Leverage System.

The new production line: faster and cheaper

The cost of producing a brand asset has collapsed, and the largest consumer companies rebuilt their content supply chains around that collapse. Unilever, the Anglo-Dutch group behind Dove and Vaseline, built photorealistic 3D “digital twins” of its products with NVIDIA’s tools and reports product imagery created twice as fast and roughly fifty percent cheaper, with content duplication cut five to one (NVIDIA / Unilever, vendor-reported). Mondelez, the maker of Oreo and Cadbury, told Reuters it is targeting a thirty-to-fifty-percent cut in marketing-content costs through its in-house generative tool (CNBC / Reuters). L’Oréal runs a generative content platform it calls CreAItech across its brands and markets (Glossy).

Speed is the other half. In Japan, the agency Dentsu reported accelerating creative workflows roughly sixfold with generative AI (Microsoft Advertising, vendor-reported). And the most efficient campaign of the early wave cost almost nothing to make: Kraft Heinz asked an image model to draw “ketchup,” it kept drawing Heinz-shaped bottles, and the resulting campaign reached more than a billion people with engagement reported about thirty-eight percent above past work (Adweek).

The reinvented agency

The service layer of marketing is adapting fastest, and three agency cases — all built on Claude — show the shape of it. In Norway, the communications group TRY put Claude in the hands of more than four hundred creatives and reports a thirty-percent reduction in time on routine tasks and forty-percent-faster proposal development (Anthropic, vendor-reported). In the UK, the performance agency Brainlabs gave more than a thousand marketers an “AI coworker” for audits, reporting, and client work (Anthropic, vendor-reported).

The sharpest number comes from Brazil. The adtech firm Advolve uses Claude as the engine for autonomous ad creation, deployment, and budget optimization for clients including iFood and Cogna, and reports a ninety-percent reduction in operational time and a fifteen-percent lift in client return on ad spend while steering budgets toward nine figures (Anthropic, vendor-reported). These are vendor accounts and read as such, but the direction is unmistakable: the agency’s value is moving from production hours to judgment and orchestration.

The discovery shift no marketer can ignore

The deepest change is on the demand side, in how customers find things, and most marketers underweight it. Gartner has forecast that traditional search volume will fall twenty-five percent by 2026 as buyers move to AI assistants (Gartner), and Pew Research found that when Google shows an AI summary, the share of users clicking any link falls to roughly eight percent from fifteen (Pew Research). The channel marketing optimized for two decades is being replaced by one where the answer often contains no link at all.

The brands moving early are optimizing to be the source those answers cite — the discipline now called generative engine optimization. Webflow, which did this deliberately, reports that roughly eight-to-ten percent of its signups now come from AI sources and that its ChatGPT traffic converts at around twenty-four percent, several times its rate from Google search (Webflow). On the commerce side, Shopify reported AI-attributed orders rising roughly elevenfold over 2025 (Shopify). The customer journey increasingly begins, and sometimes ends, inside an assistant.

Lifecycle and the agentic turn

Email and CRM are being rebuilt around AI that reads the customer data and acts on it in plain language. The marquee example: in May 2026, Klaviyo and Anthropic launched an integration that lets a marketer pull live Klaviyo data into Claude and ask, conversationally, to audit flows, build reports, and draft re-engagement campaigns — work that, in Klaviyo’s words, “used to take hours” (Klaviyo, vendor-reported). The longer track record belongs to AI copy optimization: JPMorgan Chase’s multi-year work with Persado reported click-through lifts as high as 450% in testing (Marketing Dive). And in Latin America, Mercado Libre’s generative ad tools produced more than ninety thousand creatives and lifted ad impressions 45%, contributing to 67% ad-revenue growth disclosed in earnings (PYMNTS).

The backlash file

The failures are as instructive as the wins, and a marketer who studies only the wins will repeat the failures. Four belong on the wall.

Speed without taste. Coca-Cola compressed a holiday campaign from roughly a year to a month using thousands of AI-generated clips across two consecutive years, and both years drew heavy criticism for work many viewers called soulless (The Wrap). Automation oversold. Klarna celebrated an AI assistant doing the work of 700 agents, then in 2025 walked back its AI-only posture and re-hired people to guarantee a human option (Customer Experience Dive). Authenticity backlash. Mango and Levi’s drew “false advertising” and “artificial diversity” criticism for AI-generated models, and Levi’s walked its program back (NBC News). The common thread is not that AI marketing fails; it is that it fails when judgment is removed from the loop. Every one of these is a governance failure, not a technology failure.

The Claude layer in marketing

A pattern runs through the agency and lifecycle cases above: the teams that care about brand voice and a reviewable workflow build on Claude. Anthropic’s own marketing organization reports the sharpest internal numbers — responsive search-ad creation cut from thirty minutes to thirty seconds, roughly ten times the creative output, and a hundred-plus hours a month freed on the influencer team — built largely by small, non-technical teams (Anthropic, vendor-reported). Canva’s AI design features, used across countless small-business campaigns, are built on Claude (Anthropic, vendor-reported). The reason is consistent with everything The Leverage Years teaches: marketing rewards a capable drafter that defers to human judgment and produces output a person can review and own.

The whole stack, in one small brand

What the conglomerates do with departments, a small brand can now do alone. EZILDINHA, a Brazilian fashion label, runs content and SEO drafted against its brand voice, Google Ads structured and optimized with AI, Meta creative built and iterated the same way, and an email calendar that mirrors the Klaviyo-and-Claude pattern — all with a team of a few people and a human reviewing before anything ships. The marketing department did not disappear. It was compressed into an operating system one operator can run, which is the entire premise of what follows.

The operating system to build

The teams that get durable value share a four-part structure, and it is the same idea The Leverage Years teaches applied to marketing. A brand-voice layer — how the brand sounds, what it claims, what it never says — so AI content sounds like the brand, not the model’s default. A data and confidentiality boundary for what customer data never enters a model. A review gate run before anything publishes, sends, or goes live — the thing the backlash file shows is missing when campaigns go wrong. And a measurement spine of a few honest numbers — acquisition cost, conversion, retention, share of AI-discovery — rather than vanity reach. With those four in place, the eight functions become a system rather than scattered experiments, and the tools become interchangeable.