In January 2026, Anthropic published something most companies would keep quiet: exactly how its own marketing team uses AI, with the numbers. Case studies that took two and a half hours now take thirty minutes. Responsive search ads that took half an hour now take thirty seconds. The growth team’s creative output rose roughly tenfold. The influencer team freed up more than a hundred hours a month. None of this was done by a department of specialists with a large budget; much of it was built by small, non-technical teams using Claude.
That disclosure is a useful mirror, because it shows what has quietly happened to marketing everywhere. The function looks the same from the outside — campaigns still ship, emails still send, ads still run. But inside, nearly every task that makes up a marketing operation has been rebuilt around AI, one function at a time, and the gap between the teams that rebuilt early and the ones still working the old way is widening fast.
The short version: Marketing in 2026 runs on AI across eight functions — content, SEO and AI-search, paid media, email and lifecycle, creative production, social and influencer, analytics, and personalization. This briefing tours each with real, sourced examples (Unilever, L’Oréal, Klaviyo, HubSpot, Webflow, Mondelez, Heinz, and Anthropic’s own team), names the failures that should keep you honest (Coca-Cola, the Gemini Super Bowl error, rising ad-acquisition costs), shows how a small Brazilian label called EZILDINHA runs the whole stack alone, and lays out the operating system a marketing team should build — the system at the heart of our forthcoming Daily Marketing Leverage System.
This is not a tools roundup. It is an operating map — what AI now does well in each part of marketing, where the judgment line sits, and what separates leverage from expensive noise.
The scale: marketing is the function AI changed first
Of all the places AI has landed in business, marketing moved first and fastest — because so much of marketing is the production of language and images, exactly what generative models do. The adoption data is unambiguous.
McKinsey’s 2025 State of AI survey found that 88% of organizations now use AI in at least one function, and that marketing and sales saw the single biggest adoption surge — more than doubling since 2023. Salesforce’s tenth State of Marketing report, surveying 4,450 marketers, found 75% have adopted AI, and a related figure that 87% now use generative AI in at least one recurring workflow, up from 51% in early 2024. In roughly two years, AI went from a curiosity to a standard part of the marketing week.
The deeper shift is on the demand side, in how customers find things — and it is the part most marketers underestimate. Gartner has forecast that traditional search volume will drop 25% by 2026 and organic search traffic could fall 50% by 2028 as buyers move to AI answers. Pew Research found that when Google shows an AI summary, the share of users clicking any link falls to roughly 8% from 15%. Similarweb data shows zero-click searches rising from 56% to 69% in a year. The channel marketing has optimized for two decades is being replaced by one where the answer often contains no link at all.
And the new channel is already producing revenue. Shopify reported AI-referred traffic to its merchants up roughly seven-fold and AI-attributed orders eleven-fold year over year, with ChatGPT the dominant source. Adobe measured retail traffic from generative-AI sources up more than 1,300% over the holidays. The traffic that arrives from AI tends to convert unusually well: Webflow, which optimized deliberately for it, reports ChatGPT traffic converting at around 24% — roughly six times its Google-search rate. Marketing is being rebuilt on both sides at once: how the work gets made, and how customers arrive.
What follows is the operation, function by function.
The data, in depth
The headline adoption numbers understate how uneven the picture is, and the detail is where the strategy lives. Salesforce’s survey of 4,450 marketers found that while three-quarters have adopted AI, only about a third have fully implemented it, and a striking 84% admitted to running generic campaigns — a candid acknowledgment that adoption and skill are not the same thing. McKinsey’s 2025 work makes the same point from the value side: even as adoption approaches universality, only a small share of organizations, on the order of 6%, are capturing outsized value. The gap between using AI and using it well is the entire game.
On the demand-side shift, the figures compound. Pew’s study of nearly 69,000 searches found AI summaries appearing on roughly 18% of searches as of early 2025 and cutting link clicks by nearly half when present. Similarweb put zero-click searches at 69%. Shopify reported AI-attributed orders up elevenfold and AI-referred traffic up sevenfold year over year, and Adobe measured generative-AI retail traffic up more than 1,300% over the holidays, roughly doubling every couple of months. Whatever the precise number, the direction is not in dispute, and it is steep.
On effectiveness, the personalization band is wide and worth reading carefully. Industry analysis puts conversion lifts from AI recommendation systems anywhere from 15% to several hundred percent, with sites running AI chat and recommendations converting several times better than static ones — figures that are vendor-aggregated and should be treated as a range, not a promise. Amazon’s recommendation engine remains the benchmark, long estimated to drive on the order of a third of sales. And the copy-optimization track record is real and old: JPMorgan Chase’s work with Persado reported click-through lifts as high as 450% in testing years before the current wave. The takeaway is not any single number; it is that the effect is real, large, and entirely dependent on execution.
The international advantage
One of the most underrated effects of AI marketing is what it does to borders. Localization — the translation and cultural adaptation of marketing across markets — was historically expensive enough that only large brands did it well, which kept smaller brands penned into their home market. That constraint is dissolving.
The evidence is concrete. Adore Me cut a localized product launch from months to ten days. L’Oréal’s CreAItech produces brand-consistent creative across twenty markets in hours. And at the small end, the Claude-built app Reversia translates an entire store — product copy, SEO tags, structured data — into more than 110 languages at 99% accuracy, turning what was a five-figure, multi-week project into a setting. International SEO and multilingual marketing, once a moat that favored the largest players, are becoming available to any disciplined operator.
For a brand like EZILDINHA, this is not a hypothetical. A small label can now present itself credibly to customers in markets it could never previously afford to enter, with marketing that reads as native rather than machine-translated, and with the SEO and AI-search footprint to be found there. The brands that grasp this early will expand their addressable market faster than their headcount, which is precisely the kind of leverage this entire series is about. The border that used to protect incumbents is becoming a line a small brand can simply step across.
The budget, the team, and the skills are changing with it
Before the functions, a word on what the adoption is doing to the shape of a marketing organization, because the second-order effects are larger than the productivity numbers suggest.
The budget is moving. As production costs collapse, the money that once paid for making assets is migrating toward strategy, distribution, and the data layer. Mondelez did not invest more than $40 million in generative tooling to make the same number of ads more cheaply; it did so to make far more, tuned to far more contexts, and to redirect the savings into media and measurement. The pattern repeats wherever AI lands in marketing: the cost of making the thing falls, and the value concentrates in deciding what to make, where to put it, and how to know if it worked.
The team is changing with it. The traditional marketing org chart was built around production specialties: a copywriter, a designer, a media buyer, an email manager. The emerging shape is fewer, more senior generalists who orchestrate AI across several functions, supported by a smaller number of deep specialists for the work that still demands craft. Anthropic’s disclosure is instructive. The tenfold creative output and the thirty-second ad creation came from small, often one-person teams, not from added headcount. The leverage did not make the team bigger. It made each person’s reach wider.
The valued skills are shifting accordingly. The premium is no longer on producing a single polished asset by hand; the model does that. It is on briefing precisely, judging output against a standard, catching the error a model will confidently make, and deciding what is worth producing at all. Those are editorial and strategic skills, not production skills, and they are the ones a seasoned marketer already owns. The marketer most exposed is the one whose value was production speed. The marketer most rewarded is the one whose value was judgment, who now has a tireless production team to direct.
This is the quiet through-line of the whole shift. AI does not replace the marketer with taste. It clears the production drudgery that kept that marketer from spending enough hours on the work only taste can do.
Function 1 — Content & copywriting at scale
Content is where most marketing teams start, because it is the highest-volume, most repetitive output and the most obvious fit for a fast drafter. The results are real and, in places, dramatic.
Anthropic’s own customer-marketing team reports cutting case-study drafting from two and a half hours to thirty minutes — roughly ten hours a week saved — and its product-marketing team saving five to ten hours per launch on briefs, both using Claude with structured projects and skills. The lingerie and apparel brand Adore Me, using a content tool, cut product-description generation from twenty hours to twenty minutes per batch and compressed a localized launch from months to ten days. A B2B agency reported producing SEO blog drafts 50% faster. And in the splashiest early example, Kraft Heinz’s “AI Ketchup” campaign — built on generative imagery — reportedly reached more than a billion people and earned roughly 25 times its ad spend in earned media.
The judgment line. Content is the safest function to scale and the easiest to ruin. A team that feeds the model a clear brand voice and reviews every piece sounds like itself, faster. A team that ships the default floods its own channels with generic material that erodes the brand it was meant to build. The discipline is “draft at scale, edit by judgment” — volume is a means, never the goal.
Function 2 — SEO & the move to AI search (GEO)
Search engine optimization is being split in two. The classic discipline — ranking in Google’s blue links — still matters, but a second discipline has emerged beside it: getting cited by AI answers, now called generative engine optimization, or GEO. The brands moving early are treating both as one job.
Webflow is the clearest documented case: by optimizing deliberately to be cited and to convert from AI discovery, it reports that roughly 8–10% of its signups now come from LLM sources, growing about four-fold year over year, with ChatGPT traffic converting at around six times its Google rate. HubSpot has published its own playbook for becoming the most-cited CRM in AI search, and reports a customer, Docebo, where AI sources now drive around 15% of leads. At the structural end, programmatic SEO — AI-assisted generation of thousands of templated, data-fed pages — underpins giants like Zapier (millions of monthly organic visits across 50,000+ integration pages) and Wise (tens of millions of visits from currency-pair pages), though those traffic figures are third-party estimates.
The judgment line. AI-assisted SEO scales content production, but the failure mode is thin, near-duplicate pages that AI search and Google both increasingly ignore. The teams winning use AI to produce genuinely useful, structured, citable content — clear answers, real data, proper schema — and a human to ensure every page earns its place. The goal is to be the source an AI quotes, which rewards quality and structure, not volume for its own sake.
Function 3 — Paid media automation
Paid media has quietly become one of the most automated functions in marketing, as the platforms themselves moved to AI-driven buying and as teams used AI to produce the creative and copy that feed it.
The flagship example is Anthropic’s own growth team, which built a custom workflow in Claude Code — a slash command plus brand-voice and best-practice skills — that turns a brief into fifteen headlines and four descriptions in an upload-ready file. The reported result: responsive-search-ad creation cut from thirty minutes to thirty seconds, copy creation from two hours to fifteen minutes, and roughly ten times the creative output, built by a non-technical, one-person team. On the platform side, L’Oréal reported that Google’s AI Max for Search lifted click-through rate 67% and cut cost-per-conversion 31% versus other match types. Meta states its Advantage+ automation returns roughly $4.52 per dollar spent, about 22% above manual, and brands like Joybird (a 95% revenue lift on Performance Max) and Ray-Ban (a 9% ROAS gain on Advantage+) report concrete results.
The judgment line. Paid media is where automation can quietly mask a rising bill. The platforms optimize for what you tell them to, and AI creative can scale spend faster than judgment can check it. The discipline is to own the objective and watch the true cost — not just ROAS, but customer-acquisition cost and incrementality — because, as the cautionary section will show, automated buying can lift a vanity metric while the real cost of a new customer climbs.
Function 4 — Email, lifecycle & CRM
Lifecycle marketing — the email, the segment, the re-engagement flow — is being rebuilt around AI that can read the customer data and act on it in plain language. This is also where one of the marquee Claude integrations of 2026 landed.
In May 2026, Klaviyo and Anthropic launched an expanded integration that lets a marketer pull live Klaviyo data into Claude and ask, conversationally, to generate performance reports, audit flows, and draft re-engagement campaigns — work that, in Klaviyo’s words, “used to take hours,” now drafted in minutes, for a customer base that includes Glossier, Liquid Death, and Mattel. The longer track record belongs to AI copy optimization: JPMorgan Chase’s multi-year deal with Persado reported click-through lifts of up to 450% in testing, and Vanguard’s institutional arm reported a roughly 15% conversion lift on AI-optimized LinkedIn messaging. The sales-training firm Sandler, using HubSpot’s AI, reported a 25% email click-through lift and a sales cycle cut from 90 days to 45.
The judgment line. Lifecycle is where personalization and privacy meet, so the failure mode is a message that is wrong, tone-deaf, or sent to someone for whom it is inappropriate. The discipline is to let AI draft the report, the segment, and the copy — and to keep a human approving the send, especially around sensitive moments. The relationship is the asset; speed must never compromise it.
Function 5 — Creative & ad production
Creative production is the function where the cost curve broke most violently. Generating a brand image, a localized ad, or a short video used to mean a studio, a shoot, and weeks of lead time. The largest consumer companies have rebuilt that supply chain, and the numbers they report are the most quoted in the industry.
Unilever runs what it calls a Beauty AI Studio across brands including Dove and TRESemmé, producing roughly 400 assets per product against the 20 a traditional campaign yielded, with production about 30% faster and both video-completion and click-through rates roughly doubling. Mondelez invested more than $40 million in an in-house generative tool with Accenture, targeting a 30 to 50% cut in content costs, and reported early personalization tests lifting return on investment 20 to 30%. Zalando compressed image production from six-to-eight weeks to three-to-four days and reported campaign costs down roughly 90%, with around 70% of its editorial imagery now AI-generated. L’Oréal’s CreAItech lab, built on Google’s models, took campaign turnaround from weeks to hours across 20 markets. Ferrero generated seven million unique Nutella jar labels, and the run sold out.
The judgment line. Creative is where brand equity lives, and where the “more is better” instinct does the most damage. The companies that win treat AI as the production line and a brand team as quality control with veto power. Unilever governs its output with a framework it calls Brand DNAi and a public pledge not to use AI to depict real women. The discipline is a human who can look at faster, cheaper work and still say it is not good enough to carry the name.
Function 6 — Social & influencer
Social and influencer marketing has absorbed AI in two ways: the production of organic social content, and the rise of synthetic personas. Anthropic’s influencer team reports freeing more than a hundred hours a month using Claude to script and prepare influencer and podcast work. On the synthetic side, virtual influencers like Spain’s Aitana López have landed real brand deals, and Lil Miquela has worked with Samsung, Calvin Klein, and Prada.
The cautionary note here is built into the data. Brand willingness to use virtual influencers fell from 86% to 60% between late 2024 and late 2025, even as the category’s market value grew, a sign that the novelty is wearing off faster than the trust is building. The durable use of AI in social is not the synthetic face; it is the production leverage that lets a real brand voice show up more often and more consistently.
The judgment line. Social rewards authenticity, which is precisely what AI threatens when it is used to fake a person rather than to amplify a real one. The teams getting it right use AI to produce more of their genuine point of view, and reserve synthetic media for clearly labeled, low-stakes uses.
Function 7 — Analytics, mix modeling & agentic reporting
The least visible and fastest-maturing function is analysis. Marketing reporting has long been a tax on the team’s week; AI is turning it into a conversation. Anthropic’s digital-marketing team reports productivity up roughly fivefold year over year using Claude Code for reporting and web work, with trade-show preparation cut 40%. A new category, “agentic marketing-mix modeling,” is emerging: in one documented life-sciences example, an agentic system reallocated 15% of digital spend and reported an 11% lift in prescription volume within six weeks, and Gartner’s 2025 evaluation of mix-modeling vendors confirmed the leaders are investing heavily in agentic and generative capability.
The judgment line. Analysis is where AI’s confidence is most dangerous, because a clean-looking chart invites trust. The discipline is to treat AI-generated analysis as a fast first read that a human validates against the source data before any budget moves. The model surfaces the pattern; the marketer owns the decision and the dollars.
Function 8 — Personalization & conversion
The oldest form of marketing AI, recommendation and personalization, remains the one most directly tied to revenue. Amazon’s recommendation engine has long been estimated to drive on the order of a third of company sales, and industry analysis puts conversion lifts from AI recommendation systems across a wide band, with chat-and-recommendation sites converting several times better than static ones. On the creation side, Canva’s AI design features, built on Claude, now sit inside one of the most-used creative platforms in the world, with the company reporting that roughly two-thirds of its team use AI daily.
The judgment line. Personalization is trusted to act with the least human review, which is the risk. A system optimizing the wrong objective will cheerfully push the lowest-margin bundle or the most-returned product. The discipline is to define what “good” means — margin, lifetime value, retention — and audit what the system actually optimizes against it.
The whole stack, run by one brand: EZILDINHA
The function-by-function tour describes what large companies do with departments. The more striking case is what a small brand does with none. EZILDINHA, a Brazilian fashion label selling linen and silk dresses, kaftans, and resort wear, runs all eight functions with an AI operating layer and a team of a few people.
Its content and SEO run on AI drafted against the brand voice and built to be cited in AI answers, not merely ranked. Its paid media — Google Ads structured and optimized with AI, and Meta creative and targeting built and iterated the same way — covers the function that larger competitors staff with media buyers. Its email and lifecycle work mirrors the Klaviyo-and-Claude pattern now used by far larger brands. Its creative production borrows the logic of the Unilever assembly line at a fraction of the scale. And its analysis turns store data into operating decisions without a dedicated analyst.
This is not an outlier; it is the small-brand version of a documented pattern. Anthropic’s ChatPlace customer story describes solo operators getting “an AI marketing team,” reporting 15 to 20 hours saved per week and 15 to 40% revenue increases. Brazil’s Sebrae found 44% of the country’s small businesses already using some form of AI. EZILDINHA is what the full stack looks like when one operator runs it deliberately, with judgment in the loop. The marketing department did not disappear. It was compressed into an operating system that one person can run.
Three transformations, up close
Functions and workflows are easier to grasp through whole operations. Three short portraits show how the pieces fit into a working marketing function at very different scales.
The enterprise creative assembly line
Unilever is the most complete picture of marketing rebuilt as a system rather than a series of campaigns. Its Beauty AI Studio runs like a factory floor: a product and a brief go in, and hundreds of on-brand assets come out, formatted for every channel, against a governing framework the company calls Brand DNAi. The reported results — roughly 400 assets per product against 20 before, production about 30% faster, completion and click-through rates roughly doubling — matter less than the structure behind them. Unilever did not buy a tool and hope. It defined inputs, brand rules, a review step, and outputs, and inserted AI as the engine inside that defined process. The lesson for a far smaller team is the structure, not the scale: the value came from the operating system, not the model.
The performance DTC brand
A direct-to-consumer brand running on paid media and lifecycle email shows the other end of the spectrum, where AI touches the revenue line directly. The pattern, visible in the Joybird and Ray-Ban platform results and the Klaviyo integration, is to let AI produce the ad-creative variants and the lifecycle copy at volume while the platform’s own AI handles bidding and delivery, and to keep the human focused on the one thing automation cannot be trusted with: the true cost of a customer. The brands that win here treat AI as a force multiplier on creative and lifecycle throughput, and treat acquisition cost and incrementality as the numbers a human watches like a hawk, because automated buying can lift headline return while the real cost of growth climbs.
The B2B software company
For a B2B SaaS company, the transformation is concentrated in content and discovery. Webflow’s decision to optimize deliberately for AI search, and HubSpot’s published playbook for becoming the most-cited CRM in AI answers, describe the same move: produce a deep library of genuinely useful, structured, citable content, and measure the share of signups and leads now arriving from AI sources rather than classic search. The reported outcomes — Webflow’s 8 to 10% of signups from AI, converting at several times the search rate — point to where B2B demand generation is heading. The company that becomes the source AI answers quote owns a channel its competitors cannot see in their analytics yet.
Three scales, one logic. The enterprise rebuilds production, the performance brand rebuilds throughput and watches cost, the B2B company rebuilds discovery. In every case AI carried the volume and a human owned the standard and the number that mattered.
The Claude layer in marketing
A pattern runs through the strongest examples above. The teams that care about brand voice, controllable workflows, and a reviewable process tend to build their marketing operations on Claude. Anthropic’s own marketing organization reports the sharpest published numbers — ten hours a week saved on case studies, thirty-second ad creation, tenfold creative output, fivefold reporting productivity. The Klaviyo integration runs lifecycle marketing through Claude. Canva’s AI design features are built on Claude. The reason is consistent with everything The Leverage Years teaches: the work of marketing rewards a capable drafter that defers to human judgment, keeps confidential data controllable, and produces output a person can review and own. That posture is the point, and it is what separates a marketing function that has leverage from one that merely has tools.
Nine mini-workflows: how the work actually gets done
The functions describe what AI touches. The workflows describe how a disciplined team actually runs them day to day. Nine that recur across the best operations, each a small, repeatable loop with a human gate at the end.
1. The monthly content engine
A team feeds the brand-voice file and a month’s themes to the model, generates a full calendar of posts, articles, and email drafts in one sitting, then edits each by hand before scheduling. The work that consumed a week of a content manager’s time becomes an afternoon of drafting and a day of editing. The discipline is that nothing publishes unread.
2. The listing-to-collateral pipeline
For commerce brands, one product brief becomes a description, a set of feature bullets, an email, three social variants, and the structured data for the product page, all drafted together and reviewed as a set. This is the small-brand version of the Stitch Fix and Zalando catalog engines, available to any operator with a voice file and a review gate.
3. The responsive-search-ad loop
The Anthropic growth team’s pattern: a brief plus a brand-voice and best-practice skill produces fifteen headlines and four descriptions in an upload-ready file in seconds. The marketer reviews for claims and compliance, uploads, and lets the platform test. Creation time falls from half an hour to under a minute; the marketer’s job shifts entirely to judgment.
4. The GEO answer page
For a target question a customer might ask an AI, the team drafts a clear, structured, sourced answer page with proper schema, designed to be the passage an answer engine quotes. This is the Webflow and HubSpot pattern: produce the citable source rather than the keyword-stuffed page, and measure the share of traffic and leads arriving from AI.
5. The lifecycle audit and re-engagement draft
The Klaviyo-and-Claude pattern: pull live flow and campaign data into the model, ask it to audit the flows and identify the lapsing segment, and have it draft the re-engagement campaign for review. Work that took hours of dashboard archaeology becomes a conversation, with the send still gated by a human.
6. The creative variant set
One approved concept becomes dozens of on-brand variants for different formats, audiences, and markets, in the Unilever assembly-line mold, with a brand team approving the set against the standard. The model produces the variations; the human guards the brand.
7. The weekly reporting conversation
Instead of building a deck, the marketer asks the model to turn the week’s data into a plain-language read of what moved and why, then validates the surprising numbers against the source before circulating. Anthropic’s digital team reports reporting productivity up roughly fivefold on exactly this pattern.
8. The repurposing chain
One substantial asset — a webinar, a report, a long article — is decomposed by the model into a dozen derivative pieces: social posts, an email series, short scripts, a slide outline. The marginal cost of a second format approaches zero, and the brand shows up in more places from the same core idea.
9. The competitive and market scan
The model synthesizes public competitor moves, category trends, and customer language into a structured brief the marketer can act on, replacing hours of manual scanning. The output is a starting point a human verifies, not a finished analysis, but it removes the blank page from strategic work.
None of these is exotic. Each is a small loop: brief precisely, draft at the model’s speed, review by judgment, ship. The teams that win run several of these every week until they are simply how the work is done.
Channel by channel: a 2026 field guide
The eight functions cut across channels. It is worth a quick channel-level read, because the practical question a marketer asks is usually “what changed on this surface?”
Organic search. The biggest structural change. Classic ranking still matters, but the AI summary now intercepts a large share of clicks, and the job becomes being the cited source rather than the tenth blue link. Win condition: structured, genuinely useful, citable content with proper schema.
Paid search and social. Largely automated at the bidding layer by Google’s and Meta’s own AI, with the human role shifting to creative volume, audience strategy, and cost discipline. Win condition: feed the machine strong, varied creative and watch acquisition cost, not just return-on-ad-spend.
Email and lifecycle. The most mature application, now moving from AI-assisted copy to AI that reads the data and drafts the campaign, as the Klaviyo integration shows. Win condition: let AI audit and draft; keep a human approving the send and guarding the relationship.
Content and blog. Production cost collapsed; the differentiator is editorial judgment and structure. Win condition: a brand-voice layer and a review gate that keep volume from becoming generic.
Social organic. AI multiplies a real brand voice across formats and platforms; synthetic personas remain a niche, high-risk play. Win condition: more of your genuine point of view, more consistently.
Creative and video. The cost curve broke hardest here, enabling volume and personalization that were previously impossible. Win condition: an assembly-line process with brand quality control holding veto power.
Web and conversion. Personalization and recommendation, the oldest marketing AI, remain tied most directly to revenue. Win condition: define what the system optimizes for and audit that it actually does.
The new channel: the AI assistant itself. Increasingly, the customer journey begins inside ChatGPT, Perplexity, or an AI Overview, and sometimes ends there. This is a channel with its own optimization discipline (GEO) and, soon, its own transaction layer. Win condition: be present, cited, and structured where the answer is formed. The marketer who treats the AI answer as a channel rather than a threat is the one who will own the introduction.
Who is pulling ahead, and who is being left behind
Across the examples in this briefing, the teams winning and the teams stalling sort along a few clear lines, and it is worth naming them because they are diagnostic.
The teams pulling ahead share four traits. They wrote down a brand voice before they scaled, so their output sounds like them. They treat AI as a production multiplier directed by human judgment, not as a replacement for it. They moved early on AI search, treating GEO as a real channel rather than a curiosity. And they manage by honest numbers — acquisition cost, conversion, retention, share of AI discovery — rather than by output volume. Anthropic’s own team, Unilever, Webflow, and the disciplined DTC operators all fit this profile.
The teams being left behind share the opposite traits. They generate at volume without a voice, and their channels fill with generic content that erodes trust and underperforms in both search and AI answers. They chase tools instead of building an operating standard, accumulating subscriptions without leverage. They treat the search shift as a problem to wait out rather than a channel to win. And they manage by vanity metrics, mistaking more output for more value. The Coca-Cola and Gemini episodes are the visible failures; the invisible one, far more common, is the team quietly drowning its own brand in mediocre AI content while wondering why the numbers are not moving.
The encouraging part is that the dividing line is not budget or size. A small brand with discipline beats a large one without it, which is exactly why a label like EZILDINHA can run a marketing operation that embarrasses competitors many times its size. The differentiator is the operating system, and the operating system is learnable.
The compliance and brand-safety layer
Few marketing leaders enjoy the legal conversation, but in an AI-driven function it is no longer optional, and the teams that treat it as part of the operating system avoid the failures that befall the teams that treat it as an afterthought.
Three exposures matter most. The first is factual accuracy: a model will state a wrong number or a fabricated claim with complete confidence, and in advertising that is not a quirk but a liability, as Google’s own Gemini Super Bowl spot demonstrated. Every customer-facing claim a model produces must be verified against a source before it ships. The second is intellectual property and likeness: AI-generated imagery that resembles a real person, a competitor’s protected work, or copyrighted material can create real legal exposure, which is part of why brands like Unilever pair their AI creative with explicit governance and pledges. The third is regulated-category rules: financial, health, and other regulated marketing carries disclosure and substantiation requirements that a model neither knows nor honors unless instructed, and the brand remains responsible regardless of how the copy was produced.
The remedy is not to slow down; it is to build the guardrails into the workflow. A review gate that explicitly checks claims, likeness, and regulatory lines turns these from lurking risks into routine checks. The principle is the same one that governs every function in this briefing: the model drafts, a human with the relevant judgment approves, and the brand owns the output. Compliance is not the enemy of speed. It is the thing that lets a brand move fast without eventually moving into a lawsuit.
What backfires: four cautionary cases
A briefing that only counted wins would be marketing for marketing AI. The failures are as instructive as the successes, and they cluster into four lessons.
Speed without taste. Coca-Cola compressed a holiday campaign from roughly a year to a month using 70,000 AI-generated clips across two consecutive years, and both years drew heavy criticism, with viewers and the trade press calling the work soulless and pointing to visible continuity errors. The speed was real. So was the brand damage. A faster path to a worse ad is not progress.
Confidence without fact-checking. Google’s own Gemini advertisement for the 2025 Super Bowl cited a false statistic about global cheese consumption. The ad was quietly edited. When the company building the model ships a flagship spot with a fabricated number, it is a reminder that AI-generated claims in marketing carry accuracy risk that a human must catch.
Authenticity backlash. Mango and Levi’s both faced criticism for AI-generated models, with consumers raising concerns about false advertising, misrepresented fit, and the displacement of human talent. Levi’s walked its program back. In categories where trust is the product, synthetic imagery of people is a trust liability before it is an efficiency.
Automation that hides the bill. The subtlest failure is financial. Independent analysis of tens of thousands of Meta Advantage+ campaigns found new-customer acquisition cost more than doubling over a year even as automated buying lifted headline return-on-ad-spend, and AI-referred traffic, while highly engaged, has been found to convert lower on a first visit. Automation can improve the metric you watch while the metric that matters quietly erodes.
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 marketing operating system to build
The teams that get durable value from AI in marketing share a structure, and it is the same idea The Leverage Years teaches across professions, applied to a marketing function. It has four parts.
A brand-voice and standards layer. A written definition of how the brand sounds, what it claims, what it never says, and the legal and factual lines it will not cross. This is what keeps AI-drafted content sounding like the brand instead of like the model’s default, and it is the difference between content that builds equity and content that dilutes it.
A data and confidentiality boundary. A clear rule for what customer and business data never enters a model, and a tool posture that keeps the team in control of what is shared. This is non-negotiable in a function that handles customer information.
A review gate. A short, enforced checklist run before anything publishes, sends, or goes live: is it accurate, is it on-brand, did the model invent anything, does it meet the legal line, would we put our name on it. The four cautionary cases above are what happens when this gate is missing.
A measurement spine. A small set of numbers that tell the truth — not vanity reach, but acquisition cost, conversion, retention, and the share of discovery now coming from AI answers. The spine is what keeps automation honest and stops the team from optimizing a metric while the business erodes.
With those four in place, the eight functions become a system rather than eight disconnected experiments, and the tools become interchangeable. This is precisely what The Daily Marketing Leverage System is built to install.
The economics of AI marketing
Strip the case studies away and the economic logic is simple, which is why it is reshaping the function so quickly. AI collapses the marginal cost of producing a marketing asset toward zero. A description, an ad variant, an email, a localized version, a derivative social post: each used to carry real cost in time or money, and each now costs a fraction of what it did. When the marginal cost of production falls that far, three things follow predictably.
First, volume and variety rise, because it becomes economical to produce versions that were never worth making before — the second language, the tenth ad variant, the niche segment’s email. Unilever’s jump from 20 assets to 400 per product is the visible form of this. Second, the bottleneck moves downstream. When making the asset is cheap, the scarce resources become attention, distribution, and judgment about what is worth making. The constraint shifts from the studio to the strategy. Third, the value migrates with the constraint. The budget and the prestige flow toward the work that is now scarce — positioning, brand, measurement, the decision of what to produce — and away from the production that used to absorb both.
For a small brand, this economic shift is the great equalizer. The production capability that once required an agency retainer now costs a subscription, which is why a label like EZILDINHA can run a full marketing stack with a few people. For a large brand, the same shift is a mandate to redeploy: the savings from cheaper production are wasted if they are simply pocketed, and captured only if they are reinvested into the scarce, high-value work the production was crowding out. The CFO’s instinct is to bank the cost reduction. The CMO’s opportunity is to redirect it. The companies pulling ahead are doing the second.
The honest counterweight, visible in the rising-acquisition-cost data, is that cheaper production does not automatically mean cheaper growth. If everyone can produce infinite creative, the scarce input — attention — gets more expensive, not less. The advantage does not come from producing the most. It comes from producing the most relevant, to the right audience, measured against the cost of the customer rather than the cost of the asset. That is a judgment problem, and judgment is the one input AI does not supply.
What a marketing leader should do this quarter
For a CMO or a founder wearing the marketing hat, the path through all of this is narrower than the noise suggests. Four moves, in order.
Write the brand-voice layer. Before any scaling, before any new tool, produce the written specification of how the brand sounds, what it claims, and what it never says. This is the cheapest and highest-leverage thing a leader can do, and it is the one most often skipped in the rush to generate.
Pick one function and prove the loop. Choose the function with the most repetitive output and run it on AI for a quarter, through a review gate, measuring one honest number. Resist the urge to transform everything at once; a single proven loop earns the credibility and the reclaimed hours to fund the next.
Set the measurement spine and the data boundary. Decide the four or five numbers the team will manage by, and write the one-page rule for what customer and business data never enters a model. These two artifacts are what keep automation honest and the brand safe.
Build the system, not the tool stack. Once one loop works, document it, connect the next function to the same brand-voice layer and review gate, and resist the temptation to solve every problem with another subscription. The durable advantage is the operating system, and it compounds; the tools are commodities, and they do not.
Done in that order, a marketing function can go from scattered AI experiments to a coherent operating standard in a single quarter — which is precisely the transformation The Daily Marketing Leverage System is built to guide.
The next 90 days for a marketing team
Weeks 1–2: Pick one function and write the brand-voice layer
Choose the single function with the most repetitive output, usually content or paid-media creative. Before generating anything at scale, write the brand-voice and standards document. This is the foundation; skipping it is why most teams produce generic AI content.
Weeks 3–5: Run that function in production, with the review gate
Use AI for that function every working day, drafting at the model’s speed and reviewing through the gate. Track one honest number against the measurement spine so you know whether it is working. Expect the brand-voice file to improve as you see what the model gets wrong.
Weeks 6–9: Add a second function and connect the data
Bring in the next function — often SEO and AI-search, or lifecycle — and connect it to the same brand-voice layer and review gate. The leverage compounds when the functions share one standard rather than each reinventing it.
Weeks 10–13: Document the system and set the measurement spine
Write the workflows down as an operating manual a new hire could run, and formalize the four-or-five numbers you will manage by. A marketing team with its AI operating system documented has built an asset; a team running it from memory has built a dependency.
The method is deliberately unglamorous, and that is the point. The teams pulling ahead in 2026 did not find a magic prompt. They built a brand-voice layer, a data boundary, a review gate, and a measurement spine, then ran the eight functions through them until it became how marketing gets done.
The objections marketers raise
Experienced marketers do not adopt this on faith, and the good objections deserve straight answers.
“If everyone uses the same models, doesn’t all marketing converge on the same voice?” It does for the teams that ship the default. It does not for the teams that encode a distinct brand voice and edit every output. The convergence is real and is already visible in the flood of generic AI content; it is also exactly why a genuine, specific brand voice is becoming more valuable, not less. Scarcity moved from production to point of view.
“Won’t AI content get penalized by Google?” Google’s stated position rewards helpful, high-quality content regardless of how it was produced, and penalizes thin, unhelpful content regardless of how it was produced. AI makes both easier to make. The teams that use it to produce genuinely useful, well-structured material do well in both classic search and AI answers; the teams that mass-produce thin pages get filtered. The tool is neutral; the output is judged.
“The vendor numbers are self-serving.” Many are, and should be read as directional. The most consequential figures, though, are better sourced: McKinsey and Salesforce on adoption, Gartner and Pew on the search shift, and Anthropic’s own per-task numbers, which are unusually specific for a vendor. The failures are independently reported too. The balanced read is that the capability is proven and the value is uneven, accruing to disciplined teams.
“Where is the durable advantage if the tools are commodities?” In the operating system, not the tool. Everyone can license the same models. Few build a brand-voice layer, a data boundary, a review gate, and a measurement spine, and run them consistently. That discipline is the moat, as it has always been with any general-purpose technology.
What the next two years hold for marketing
Three shifts are already in motion, and each favors the team that builds the operating system now.
Marketing becomes agentic. The move underway is from AI that drafts to AI that acts — pulling the data, building the report, drafting the campaign, and increasingly executing the routine steps under human approval. The Klaviyo integration is an early instance. Within two years, the routine operational layer of marketing will run as supervised automations, and the marketer’s job will be more orchestration and judgment than execution.
Discovery keeps migrating to AI answers. The search shift is not a blip. As more buying journeys begin and sometimes end inside an AI assistant, the brands that are the clean, citable source those assistants pull from will win the introduction, and the rails for AI-assisted purchase are already being built. GEO becomes as foundational as SEO was.
The advantage moves from access to discipline. As the tools get easier and more universal, having them stops being a differentiator. The teams that win will be the ones with the operating discipline to use them well and the brand judgment to stay distinct. That is a durable advantage precisely because it is hard, and it is the advantage The Daily Marketing Leverage System is designed to build.
A short glossary of the modern marketing stack
The vocabulary has moved fast enough that even experienced marketers can lose the thread. A plain-English glossary of the terms that matter in 2026.
- GEO (generative engine optimization). Optimizing content to be cited inside AI-generated answers, as distinct from ranking in search links. The new sibling of SEO.
- Agentic marketing. AI that does not just draft but takes multi-step action — pulling data, building a report, drafting and queuing a campaign — under human approval, rather than producing text and stopping.
- Advantage+ / Performance Max. Meta’s and Google’s AI-driven campaign automation, which handle bidding, placement, and delivery and reward the marketer who feeds them strong creative and clear objectives.
- Brand voice file. A written specification of how a brand sounds and what it will and will not say, fed to the model so output is on-brand rather than default. The single highest-leverage artifact in AI marketing.
- Agentic MMM (marketing-mix modeling). AI agents layered on classic mix modeling to recommend and, increasingly, adjust budget allocation across channels.
- Zero-click search. A search that ends without the user clicking through to a website, increasingly because an AI summary answered the query in place.
- Human-in-the-loop. The governing posture of responsible AI marketing: the model drafts and proposes; a human reviews and approves before anything publishes, sends, or pays.
- Synthetic media. Fully AI-generated images, video, or personas. Powerful for production, risky for authenticity, and best used clearly labeled and in low-stakes contexts.
The terms will keep changing. The underlying division will not: the machine produces and proposes, the human directs and decides.
What AI still cannot do in marketing
A clear-eyed account has to name the limits, because the failures in this briefing trace directly to teams that assumed the model could do things it cannot.
AI cannot originate a genuinely distinctive point of view. It is trained on what exists, which makes it excellent at producing the median of a category and poor at producing the thing that breaks from it. The positioning that makes a brand matter — the angle no competitor has taken — still comes from a human who understands the market and is willing to bet on a view. A model will give you competent; it will rarely give you original.
AI cannot be accountable. When a claim is wrong, a campaign offends, or a number is fabricated, the responsibility sits with the brand and the person who approved it, not the model. This is not a technicality; it is the reason the human review gate is non-negotiable. Accountability cannot be delegated to a system that has none.
AI cannot feel the room. The judgment about whether a message lands as confident or arrogant, warm or saccharine, timely or tone-deaf, depends on a read of culture and context that the model approximates but does not possess. The Coca-Cola and Mango episodes were failures of exactly this — technically competent output that misjudged how people would feel about it.
And AI cannot own the relationship. The trust between a brand and its customer, the reputation built over years, the instinct for what this audience will and will not accept — these are human assets that AI can support but not replace. The marketer’s relationship with the market is the thing the whole operation exists to build, and it is the one thing that cannot be generated.
The practical conclusion is not to use AI less. It is to use it precisely for what it is good at — production, drafting, analysis, scale — and to reserve the scarce human hours for the four things it cannot do: originate the point of view, hold the accountability, feel the room, and own the relationship.
The crafts that appreciate in value
If production skills are depreciating, a specific set of skills is appreciating fast, and a marketer planning a career or building a team should weight toward them.
The first is editorial judgment — the ability to look at competent output and know what is wrong with it, what is missing, and what would make it sing. In a world of infinite drafts, the editor is worth more than the writer. The second is strategic positioning — deciding what to say and to whom, the work that sits above production entirely and that AI cannot originate. The third is the ability to brief, which is a real and learnable skill: the marketer who can specify precisely what they want gets dramatically better output than the one who types a vague request. The fourth is taste, the hardest to teach and the most valuable, because in a market flooded with the median, the ability to recognize and insist on the exceptional is the scarcest thing there is.
None of these is new. They are the skills good marketers always had. What has changed is that they are no longer diluted by the hours of production that used to crowd them out, which means they now determine almost the entire value a marketer creates. The function did not get less human. The human part got concentrated.
The judgment-first marketing function
Step back from the functions, the workflows, and the cautionary tales, and the conclusion is simple. AI has not replaced marketing; it has redistributed where the value sits. The production that used to consume most of a marketing team’s hours has become cheap and fast, and the scarce, valuable work — the brand point of view, the editorial judgment, the decision about what is worth making and how to know if it worked — has become the whole job.
That is good news for the marketer with taste and bad news for the one whose value was production speed. The function that wins in 2026 is not the one with the most tools or the largest content output. It is the one that built an operating system — a brand-voice layer, a data boundary, a review gate, a measurement spine — and used it to direct a tireless production capability toward the work that actually builds a brand. The tools are available to everyone. The discipline is not, and that is exactly where the advantage now lives.
The teams pulling ahead understood this early and built the system. The ones being left behind are still mistaking volume for value, and drowning their own brands in mediocre AI content while the disciplined operators take their customers. The difference between the two is not budget or size. It is judgment, applied through an operating standard — which is the entire premise of The Leverage Years, and the thing The Daily Marketing Leverage System is built to install.
Frequently asked questions
How are companies using AI in marketing in 2026?
Across eight functions: content and copywriting at scale; SEO and AI-search optimization (GEO); paid-media automation on Google and Meta; email, lifecycle, and CRM; creative and ad production; social and influencer; analytics and marketing-mix modeling; and personalization. Real examples include Unilever producing roughly 400 assets per product, Anthropic's own team cutting ad creation from 30 minutes to 30 seconds, and Klaviyo letting marketers run lifecycle campaigns through Claude. The pattern is consistent: AI drafts and carries volume while a human owns brand voice, accuracy, and the final decision.
Does AI marketing content hurt SEO or brand quality?
It does when teams ship the model's default output at volume, which produces thin, generic content that AI search and Google increasingly ignore and that erodes the brand. It does not when teams feed the model a written brand voice, produce genuinely useful and well-structured content, and review every piece. The deciding factor is governance, not the tool. Quality and structure also determine whether AI answer engines cite the content, which is the new SEO frontier.
What is GEO, and how is it different from SEO?
SEO optimizes to rank in a search engine's links. GEO, or generative engine optimization, optimizes to be cited inside AI-generated answers from tools like Google AI Overviews, ChatGPT, and Perplexity. It matters because search behavior is shifting fast: Gartner forecasts traditional search volume falling 25% by 2026, and AI summaries sharply reduce link clicks. GEO rewards clear, structured, well-sourced content with definitional passages and proper schema — content built to be quoted, not just skimmed.
Which AI is best for marketing?
The capable models all work, and the right one depends on the task. Teams that prioritize brand voice, controllable workflows, and a reviewable process often standardize on Claude, made by Anthropic — which is why Anthropic's own marketing team, Klaviyo's lifecycle integration, and Canva's AI design features are built on it. The more important choice than the model is the operating standard: a brand-voice layer, a data boundary, a review gate, and a measurement spine.
What are the biggest risks of AI in marketing?
Four recur. Speed without taste, as in Coca-Cola's criticized AI holiday ads. Confidence without fact-checking, as in Google's Gemini Super Bowl ad citing a false statistic. Authenticity backlash, as Mango and Levi's faced over AI-generated models. And automation that hides the bill, where automated ad buying lifts headline return while customer-acquisition cost quietly climbs. Each is a governance failure, preventable with a human review gate and an honest measurement spine.
Can a small business run its whole marketing function on AI?
Increasingly, yes. A small brand can run content, SEO, paid media, email, and creative with an AI operating layer and a few people, as the Brazilian fashion label EZILDINHA does. Anthropic's ChatPlace customer story reports solo operators getting “an AI marketing team,” saving 15 to 20 hours a week with 15 to 40% revenue increases. The requirement is not budget; it is the operating discipline — a brand-voice layer, a data boundary, a review gate, and a measurement spine.