Agents whisper. Buyers listen.

Somewhere in your category right now, a buyer is asking an agent which brand they should pick, and the agent is answering. Not with a list of links, but with a recommendation: a brand, a comparison, a reason. By the time the buyer ever opens a tab, the decision has effectively been made.

The instinct, when most CMOs first hear this, is to file it under "AI search" and assign someone to think about it next quarter. AI referral traffic to retail grew 4,700% year over year in 2025, according to Adobe, and the channel is already bigger than most of the lines on the dashboard marketing teams check on Monday morning. It isn't coming. It's running. The only question worth asking is whether marketing is running it as a real channel.

How agents actually behave

Agents do not browse the way humans do. They do not click, they do not scroll, they do not bounce. Each time a buyer asks one of them a question about a category, the agent does four things in sequence: it discovers a candidate set of brands, filters that set against the buyer's stated intent, ranks the survivors against each other, and answers, handing one brand to the buyer and leaving every other brand on the floor.

Marketing has spent two decades building muscles for the first two moves. SEO does discovery. Content does filtering. That playbook is in every CMO's training and every quarterly review. But ranking and answering, the moves that actually drive the revenue, sit in a layer the old playbook does not reach. Better creative will not close that gap. Neither will a louder content calendar.

What makes the gap concrete is what agents actually read. They do not read homepages or watch video ads. They read what the web has already written about a brand: reviews, threads, forums, PR coverage, third-party comparisons, conversations on Reddit from 2022 that nobody at the brand has ever seen. Then they quote it to a buyer, as if it were the truth. When the web's story is sparse, the agent's answer is sparse. The brand doesn't so much lose the deal as never enter the conversation.

Every channel you run is now an input to AI search

The reflex, on hearing this, is to file the problem under "content" or "PR" and hand it to whichever function seems closest. That reflex misunderstands the shape of what's happened. AI search does not read one channel. It reads every channel a brand already runs: content, PR, social, web copy, third-party reviews, support threads, podcast transcripts, footnotes, press mentions years old. And it reads them together as one signal. Every one of those surfaces becomes an input the agent reads and remixes before it answers a buyer.

These are the brand's agent surfaces, and they are the unit of operation in the world we're about to describe. The "AI search channel" you cannot see isn't a new channel to add to the stack. It is a new reader of every channel already on it.

The shift has a name: B2A

Marketing is no longer aimed only at humans. It is aimed at the agents acting on their behalf. The buyer is still human; the channel that decides what the buyer sees, considers, and chooses is an agent. The cleanest name for that shift is B2A, Business to Agent, and it is the new ground every marketing team is standing on, whether they have noticed or not.

From Inbound to Agent-Led Growth

New ground needs a new motion. Twenty years ago, HubSpot saw that buyers had started researching before they ever spoke to a salesperson, quietly, online, on their own clock. They named the way of running marketing built around that reality, and they called it Inbound. Then Google made the motion stick. Search rankings became the channel, page one became the goal, and an entire industry organized itself around earning the click. That motion ran the last two decades. Twenty years later, the click is no longer the channel: buyers don't research the way they used to, their agents do, and the recommendation has taken the click's place.

The motion that responds to that shift is one we've named Agent-Led Growth. ALG is the way of running marketing when agents sit between every brand and every buyer's first impression, and it is the motion we built Limy to run. It is not a campaign, a content strategy. It is an operating model where every move comes from what agents are actually doing: on the brand's site, in its category, across its competitors. The single line I have been telling every CMO I've met this year holds the whole motion in eight words. Your marketing team just got bigger. You can't fire them. But you can brief them. The team didn't change. The audience did.

The discipline behind the motion has a name as well. We call it Agentic Marketing: marketing on agentic channels, where agents discover, evaluate, recommend, and act on behalf of buyers, and using agents to execute that marketing. It runs across three surfaces. Agentic Search is how agents discover and recommend across prompts and products. Agentic Ads is paid placement inside AI agent environments. Agentic Brand is how agents represent a brand unprompted: sentiment, share of voice, revenue impact without a query triggering any of it. Three channels, one discipline, one motion.

Your dashboard is blind

There is a hard sentence to write, and it is the one that makes every other claim in this piece land. The dashboard most CMOs trust on Monday morning is blind to the channel that is growing fastest.

The "direct traffic" number on that dashboard isn't direct. Across the brands running Limy, more than 58% of the AI agent traffic on their sites is invisible to their existing analytics: logged as direct, logged as unknown, or simply not logged at all. Buyers arrived because an agent recommended the brand, compared it, or quoted it to them before they ever typed the URL. The dashboard logs the buyer. The agent gets no credit. The channel gets no line on any report.

The dashboard didn't break. It was built for humans, who leave a referrer, carry a keyword, and arrive with a campaign tag. Agents do none of that. They arrive clean, deliver the buyer to the front door, and walk away. Another integration will not fix this, and a more clever attribution model will not fix it either. The only fix is a different stack, one that lives inside the agent layer instead of on top of the web layer. That is not a strategy problem. It is an infrastructure problem.

The open loop every CMO is running

Even with a clearer view of the dashboard, the channel does not sit still. Consider the last thirty days. The team shipped ten content pieces, good pieces, on brief, aligned to the category. Three questions are worth asking about those ten pieces, and most CMOs cannot answer any of them.

First, agent models update weekly and competitors move daily. The gaps in your category's answers open faster than any manual marketing team can close them, and running this as a series of separate, manual activities won't stick. When was the last time your stack told you a gap opened, in time to close it before a buyer saw it?

Second, of the ten pieces the team shipped, which ones are agents actually citing in answers, and which ones are being ignored?

Third, the one that matters most: how much revenue did those ten pieces produce? Not influenced. Sourced. Traced back to the specific prompt, the specific agent, the specific moment a buyer started the journey.

Ten pieces shipped, three questions, three "I don't knows." That is the open loop, and it is the loop almost every CMO is running today. The gap between what your stack shows and what AI Search is doing isn't a feature gap. It's a category gap, and no stack of separate manual activities will close it.

What is ALG in practice?

ALG is the motion we've named, and the Agentic Marketing Stack is the only stack that runs it end to end. The legacy stack — analytics, CRM, the SEO suite, the visibility scoreboard — was built for a web of humans clicking links, and no combination of those tools closes the loop the agent layer demands. ALG runs as a closed loop across five moves, and the loop closes only on a stack purpose-built for it.

SEE real agent signal across the full category landscape: live agent behavior on the brand's site and across every model its buyers are using, paired with a view of every prompt forming in the category and every competitor showing up in those answers. Not visibility scores. Not weekly snapshots. Not simulated prompts.

KNOW the top gaps in the brand's story, the biggest wins on the table, and the recommendations agents are making in the category right now. The first move produces data; this one turns the data into a decision about where to act.

BRIEF the agents on what to create for those biggest wins — the assets, positioning, and signals that will move the answer in the brand's favor. The brief is the unit of work, and it spans every channel the agent reads.

ACT on the brief by creating the optimized content assets and distributing them to every agent surface. Fast enough to move the agent's next read before the next model update moves it for the competition.

MEASURE the lift: visibility, conversion, and revenue per prompt. Then feed the data back into the next brief, so the loop runs again, sharper than the round before.

See, know, brief, act, measure. One closed loop, run on the Agentic Marketing Stack. The marketer stays in control. The agents do the work.