.svg)
.svg)
The persona reshaping discovery, decision, and purchase, and the marketing org most brands haven't built for yet.
Agentic AI traffic grew 7,851% in 2025. McKinsey projects AI agents will drive a trillion dollars in US transactions by 2030. Visa just told the market that millions of consumers will use AI agents to complete purchases by the 2026 holiday season.
Your marketing org has no plan for them.
Not a real one. You may have an AEO doc somewhere. Maybe an llms.txt sitting on the dev backlog. Probably a Schema.org cleanup your SEO consultant has been pitching since Q1. But ask the simple question I keep asking marketing leaders this year: who on your team owns the agent persona? Who designs for them, measures them, prepares your brand for the moment they show up?
Silence.
That's the problem.
Agents are a persona. Not a channel, not a platform, not a tactic. They are buyers, researchers, evaluators, and decision-makers, with their own behaviors, criteria, and bandwidth constraints. They visit your site, fetch your docs, parse your product feed, buy or skip in milliseconds, and leave behind almost no analytics trail.
If you wouldn't ignore a new buyer segment driving meaningful share of your future revenue, don't ignore this one.
Three things converged and quietly rewired the internet.
The surface changed. Bot traffic now represents 32% of all HTTP requests, and AI crawlers are the fastest-growing category. Training crawlers crossed half of all AI bot traffic in Q1 2026, a full quarter ahead of schedule. Vercel's AI Gateway saw tool-using agent requests jump from 11.4% to 22.2% of total in six months, and those requests now carry 58.9% of all tokens. The cost surface of AI has shifted from chat-shaped to agent-shaped. Software is increasingly the thing consuming the internet, not eyeballs.
Commerce caught up. In September 2025, Mastercard processed the first live agentic payment, and Visa launched its Trusted Agent Protocol with Cloudflare two weeks later. Stripe shipped the Agentic Commerce Protocol with OpenAI. Google announced AP2 with more than 60 partners, including Mastercard, American Express, PayPal, Adyen, Coinbase, Salesforce, and Etsy. PayPal connected to ChatGPT, Perplexity, and Mastercard Agent Pay in a single quarter. Microsoft launched Copilot Checkout, powered by Stripe, letting Copilot users buy from Etsy, Urban Outfitters, and Anthropologie without leaving the chat. The plumbing is largely done.
Discovery moved. ChatGPT reaches 700 million weekly active users, and shopping queries now return up to 36 curated products tailored to budget and context, with no ads or marketplace carousel. Its shopping experience runs on a specialized GPT-5 mini variant trained for shopping and has live merchant program access, and OpenAI is rolling out Instant Checkout for direct in-chat purchase. Tatcha reports 3x conversion lifts from getting their data right for AI-driven discovery.
This is not theoretical. The buyer is already in the room. Your team just hasn't met them.
Marketing teams build personas for everything. The skeptical CFO. The hands-on developer. The Gen Z scroller. Weeks at offsites mapping their day, their pains, their triggers.
We don't have one for agents. So here it is.
Name: Agent. Background: Software acting on behalf of a human. Sometimes one human. Sometimes a thousand. Sometimes a corporate procurement function. Goals: Solve a task within a budget of tokens, time, and money. Reduce uncertainty. Avoid hallucinating. Pick the option most likely to succeed. Pain points: Bloated pages that blow context windows. Inconsistent product data. Sites that block legitimate crawlers. Marketing copy with zero extractable facts. Authentication and payment flows built for humans. Decision criteria: Structured data over prose. Stated capabilities over clever taglines. Verified facts over brand reputation. Fast machine-readable responses over beautiful design. Recent and accurate over authoritative but stale. Channels: Your robots.txt, llms.txt, sitemap, product feed, skill.md, MCP server, ACP endpoint, API documentation, structured data. What persuades them: Tokens of accurate, extractable, current information at the moment of retrieval.
This persona does not look at your hero image. Does not watch your launch video. Does not read your founder's LinkedIn manifesto. Does not care about your Cannes Lion or your brand colors. Does not notice the witty headline your creative director defended for two weeks.
Harsh? It is what it is. Different audience, different rules.
Here is the part most marketers miss. Many of these agents are now buying through other agents. A user agent asks a merchant agent. A merchant agent negotiates with a credentials provider. A payments agent settles with the network. Google's AP2 spec lays this out explicitly, with merchant agents creating time-sensitive bundle offers and shopping agents coordinating multi-merchant carts inside a single signed transaction. Four agents, one human signature, often no UI at any step.
Gareth Cummings of eDesk put it well: in 2026, a meaningful share of customer interactions will happen agent-to-agent. Shoppers will use AI assistants to check stock, confirm delivery, or verify returns, and brands will respond with their own agents that read order data and act instantly. Welcome to agent-to-agent marketing. The buyer is not browsing. The buyer is delegated.
Your marketing org optimizes for attention and persuasion. That works when the buyer scrolls, clicks, and feels something. Agents do not scroll, click, or feel.
Attention is not the asset. Extractability is. Persuasion is not the lever. Verification is. Brand awareness gets you in the consideration set. Structured data gets you on the answer.
Three more uncomfortable truths.
Your beautiful homepage is sometimes invisible. Agents often skip the rendered page and fetch /index.md, your sitemap, or your API. Cloudflare audited the top 200,000 websites and found only 3.9% support Markdown content negotiation, only 4% have declared AI usage preferences, and emerging standards like MCP Server Cards appear on fewer than 15 sites globally. Your site is probably one of the 96% that is hard for an agent to read.
Your funnel is partially collapsed. A user asking ChatGPT for "the best CRM for small teams under fifty bucks a seat" does not visit your homepage, enter your nurture flow, or open your email sequence. The agent decides. The funnel that took months took 4 seconds.
Your KPIs are misleading. Time on page is zero. Scroll depth is zero. Click-through rate is N/A. None of it tells you whether the agent picked you. The things that actually matter, citation share, inclusion in the answer set, share of agent-driven revenue, are not in your dashboard.
If your org has roles for SEO, paid, content, growth, brand, lifecycle, and creative, but no role responsible for how machines see, evaluate, and act on your brand, the gap is not theoretical. The gap is now.
Marketing to agents is a stack. Each layer answers a different question agents ask before they choose you.
This is the boring infrastructure layer where most brands lose before the game starts.
Agents check three things to decide whether to bother with you: your robots.txt, your sitemap, and your IP-level protections. A misconfigured robots.txt blocking GPTBot, ClaudeBot, or Google-Extended is the equivalent of putting a "closed" sign on your storefront, then wondering why no one walks in. GPTBot is the most-blocked AI crawler, disallowed by 476 of 4,055 top domains analyzed.
I get the impulse. Crawlers consume bandwidth. They train models that may compete with you. Anthropic's crawl-to-refer ratio runs into the tens of thousands. Blocking blindly costs visibility, and most teams have not done the math on the trade.
Get explicit. Differentiate between training crawlers, AI search crawlers, and user-action crawlers. The first crawls to improve models. The second crawls to power AI search. The third fetches your content because a user is actively asking. User-action crawling increased by more than 15x in 2025. That is the bot you want to welcome.
This week: audit your robots.txt with the lens of who you are blocking and why. The default of "let it all through" is no longer wise. The default of "block everything" is suicidal.
Once an agent decides to read your site, it needs to find what is relevant fast.
Think of llms.txt as the brand handshake for agents. A flat Markdown index at /llms.txt telling the agent what you offer, what is on which page, and how many tokens each section will cost them to read. Cloudflare overhauled its own developer docs and benchmarked 31% fewer tokens consumed and 66% faster correct answers compared to an average unoptimized documentation site. That is your bar.
If you sell products, you need a clean, complete, frequent-refresh product feed mapped to the formats agents prefer. ChatGPT pulls 83% of its shopping data from Google Shopping. Stripe's Agentic Commerce Suite normalizes catalogs across multiple agents through a single endpoint, and leading brands like URBN, Etsy, Coach, Kate Spade, Ashley Furniture, and Revolve are already onboard.The brands winning here all did the same boring work: clean attributes, standardized taxonomy, real-time inventory sync, no marketing fluff in product fields. A fashion brand lost ChatGPT visibility because it used "Navy" and "Navy Blue" and "Navy (Dark)" as variant labels, until they standardized and saw recommendations return within two weeks. That is the level of detail that matters.
This is where most marketing copy quietly fails.
Hero headlines like "transform your revenue motion with AI-powered intelligence" read as confident positioning to a human. To an agent, that carries zero extractable facts, and it moves on to the competitor whose page actually says what the product does.
Agents need clear claims, specific numbers, named integrations, declared pricing, stated outcomes. Vague positioning is no longer just bad copy. It is an AX failure that costs you citation, recommendation, and revenue.
This is also where skill.md and MCP server cards earn their place. Instead of asking an agent to infer your capabilities from prose, you tell it directly. Here is what we do. Here is what we don't do. Here is the input we need. Here is the output we produce. Here is the rate limit. Here is the docs link. Mathias Biilmann at Netlify made the original case here: tools that are simple for agents to integrate with and well suited to the strengths and constraints of LLMs will become vastly more capable, while platforms that are hard for agents to use will start feeling less powerful.
For brands selling anything online, the final layer is whether agents can complete the action they were sent to complete.
The stack got built faster than anyone expected. Google's Universal Commerce Protocol launched in January 2026, working across discovery, buying, and post-purchase, and compatible with A2A, AP2, and MCP. Visa shipped Intelligent Commerce Connect, an on-ramp that accepts agent-initiated payments via Trusted Agent Protocol, Machine Payments Protocol, ACP, and UCP through a single integration. Stripe has shipped four ACP releases in six months, with payment handlers, scoped tokens, discount extensions, native MCP transport, and built-in buyer auth.
The plumbing is largely done. What is not done is your integration into it. The companies moving now will own the first 18 months of agent-led purchase while the rest are still arguing about whether this matters.
There is a deeper point here for brand. When agents can compare two similar products in seconds, brand story, fulfillment quality, trust signals, and customer service start mattering more than acquisition tricks. Google's Philipp Schindler put it directly: agents will handle the grunt work of shopping so consumers can focus on the enjoyable parts. Merchants will have to compete on the parts agents cannot replace. Shallow tactics get commoditized. Deep brand work gets more valuable.
You do not need a new department. You need a new owner.
Most B2B companies in the 200 to 2,000 employee band are quietly assigning AX to one of three places: web strategy, content operations, or product marketing. The center of gravity is settling on web strategy in 2026, with dedicated VP of Agent Experience titles emerging late this year, the way Head of UX appeared in the early 2010s.
The skills profile is a blend that did not exist a year ago: structured content thinking like a technical writer, measurement instinct like a growth marketer, and basic fluency in how LLMs ingest and cite content. Closer to a search engineer than a brand designer, but with the taste of both.
The KPIs change. Stop relying on the dashboard you have. Start measuring:
That last one is the most underrated. The single biggest source of agent failure is missing or stale information. The brand that has answered every reasonable question about its product, pricing, integrations, and outcomes is the brand the agent reaches for.
In 2023, I made a similar argument about AEO. The funnel was collapsing inside ChatGPT and Perplexity, and most marketing teams had not built their orgs to respond. The teams that moved early on AEO are now the ones with disproportionate share of citation, AI-driven traffic, and influenced pipeline. Goodie's own research across millions of citations and dozens of brands shows the same pattern. Early movers compound. Late movers chase.
Agent marketing is the next layer of that same shift, and it is louder.
The internet just gained its first new majority audience in 20 years. They do not scroll. They do not tolerate slop. They do not waste tokens on bad copy. They reward clarity, structure, accuracy, freshness, and trust signals. They punish vagueness, gating, and theater.
Marketing has always been about adapting your message to who is listening. The "who" got bigger this year. The smart move is to start adapting now, while the playbook is still being written, the citation graph is still forming, and most of your competitors are still treating their site as if every visitor has eyes.
Treat agents as a persona. Hire an owner. Pick your protocols. Clean your data. Get specific in your copy. Measure what matters.
You are no longer marketing only to humans. You are marketing to the systems that act on their behalf. The brands that internalize this in 2026 will be the brands recommended, transacted with, and trusted in 2027 and beyond.
The rest will spend the next year explaining to their boards why their share kept slipping in places they could not see.
