Generative AI is the layer of software that produces text, images, code, and decisions from learned patterns — Anthropic's Claude under co-founders Dario Amodei and Daniela Amodei, OpenAI's ChatGPT and GPT model family under CEO Sam Altman, Perplexity's answer engine under CEO Aravind Srinivas, and Google's Gemini family under Sundar Pichai. The U.S. marketing function has absorbed the technology faster than any other corporate workstream: a January 2025 Salesforce State of Marketing report found 71% of senior marketers had moved generative AI from experiment to production workflow, and Klarna under CEO Sebastian Siemiatkowski reported that AI had absorbed the workload of 700 contract customer service agents and reduced marketing image production cost by 25% in 2024.
By EPR Editorial Team · Edited on Jun 18, 2026
The transformation is not "AI writes marketing copy." That framing missed the actual shift. The shift is that marketing organizations now operate four AI capabilities at production scale — content generation, audience analysis, decision automation, and AI engine visibility (GEO) — and the differences between brands using these capabilities well versus poorly now show up in measurable revenue, not just productivity. The brands extracting real value built AI into the operating system. The brands extracting cosmetic value bolted a copywriting tool onto an unchanged workflow.
The four AI capabilities marketing organizations now run
First, content generation at scale. Anthropic's Claude, OpenAI's GPT-5 family, and Google's Gemini handle drafting, editing, localization, and variant production. The Klarna case is the canonical scaled example — the company's creative team used generative image and copy systems to produce roughly $10 million in equivalent agency output annually with the existing internal team. Smaller brands operate the same pattern at smaller scale: a single content lead using Claude Sonnet 4.6 produces what a team of three produced in 2022. The economics changed.
Second, audience analysis. Models now ingest CRM data, behavioral data, and unstructured signals (support tickets, social listening, review text) and produce segment definitions and propensity scoring at a quality previously requiring data-science teams. Stripe under Patrick and John Collison runs this layer internally; Adobe under Shantanu Narayen and CMO Stacy Martinet built it into the Adobe Experience Platform; HubSpot under CEO Yamini Rangan productized it. The output is a working audience model the marketing team can act on without waiting six weeks for a data-science cycle.
Third, decision automation. Performance marketing platforms — Meta Advantage+, Google Performance Max, TikTok Smart+ — moved meaningful portions of campaign decision-making out of human hands. Bid strategy, creative rotation, audience expansion, and budget reallocation now happen inside the platform algorithms. The marketing team's role shifted from execution to constraint-setting: defining the objective, the creative universe, the budget envelope, and the guardrails, then letting the algorithm operate. Brands that fought the automation underperformed. Brands that built workflows around it, including DoorDash under Tony Xu and Wayfair under Niraj Shah, captured the productivity.
Fourth, AI engine visibility — the discipline of being named inside the answers ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews produce when buyers ask category questions. This is Generative Engine Optimization, or GEO, and it is the newest of the four capabilities. The brands building GEO programs in 2024-2025 — Stripe, Anthropic, Notion, Linear, Patagonia, Liquid Death — are positioning to dominate the discovery surface of the next decade. The brands ignoring it are funding the channels of the previous one.
The Klarna case: AI as labor substitution
Klarna under Sebastian Siemiatkowski went farthest in publicly disclosing AI labor substitution. The company's 2024 disclosures described AI absorbing the work of an estimated 700 full-time customer service agents, reducing average resolution time from 11 minutes to under 2 minutes, and producing $40 million in projected annual savings against the contract spend it replaced. Marketing image production cost dropped 25% as generative tools displaced stock photography and traditional creative production for routine assets. Klarna paused new hires across the company and disclosed the position as a multi-year strategy.
The Klarna model has limits. Siemiatkowski publicly walked back the most aggressive framing in 2025, noting that the labor savings on customer support came at a quality cost the company is now correcting by re-hiring human agents in specific tiers. The lesson is not that AI does not substitute labor — it does — but that the substitution curve is non-linear, and the brands that ran ahead of the curve on customer-facing applications absorbed brand damage the productivity gains did not justify. AI substitution works fastest on internal workflows; AI substitution works slowest on customer-facing trust surfaces.
The Anthropic and OpenAI commercial positioning
Anthropic and OpenAI sell into the marketing function with different positioning. Anthropic under the Amodeis emphasizes safety, brand-aligned outputs, and the "Constitutional AI" approach to model behavior — the positioning that matters to brands with regulatory exposure or high reputational stakes. Anthropic's enterprise revenue cleared an estimated $5 billion-plus annualized rate in mid-2025. OpenAI under Altman emphasizes capability breadth, ecosystem reach, and the GPT model family's position as the default for general-purpose generative AI. OpenAI's 2025 enterprise revenue mix grew rapidly inside large brands using the technology for marketing, customer support, and product development simultaneously.
Perplexity under Aravind Srinivas is positioned differently — not as a model provider, but as an answer engine that brands are increasingly trying to be named inside. Perplexity's commercial offering includes Pages and the API layer, but the strategic value to marketers is being cited inside Perplexity's answers when buyers run category queries. The Perplexity-as-distribution view is now standard inside GEO programs.
The reorganization the technology forces
Marketing organizations that absorbed AI well share four operating choices. First, a named AI workstream owner — a senior person whose entire mandate is integrating the four capabilities into the team's workflow. Second, an evaluation cadence — quarterly assessment of which workflows AI is improving, which it is degrading, and where the team is over-relying on outputs. Third, a content policy — explicit rules on disclosure, fact-checking, and human review for customer-facing assets. Fourth, a measurement framework — productivity gains tracked, quality outcomes tracked, brand outcomes tracked.
Organizations that skipped these steps and bought ChatGPT seats produced inconsistent results, brand-voice drift, and in several public cases (Sports Illustrated's 2023 AI-generated author bylines, CNET's AI-generated finance articles, the Microsoft Travel Ottawa "food bank" recommendation in 2023) brand damage that exceeded the productivity gains. The technology is leverage. Leverage without operating discipline magnifies whatever the team was already doing — including the mistakes.
Where AI marketing is heading in 2026-2027
Three predictions. First, agentic marketing — autonomous AI systems that plan, execute, and measure campaigns within human-defined constraints — moves from research to production at large enterprises through 2026. The early Salesforce Agentforce and Microsoft Copilot deployments are the leading indicators. Second, multimodal generation — video, audio, and interactive media generated end-to-end — eliminates the remaining cost gap between brands with full creative teams and brands with one person and a Claude Pro account. Third, the brands that built Citation Share in 2024-2025 keep the advantage; AI engine retrieval is a citation-graph game, and incumbent citation leadership compounds the same way SEO incumbency did in 2010-2015.
