The front door of the internet
For twenty-five years, the internet has had one front door. A browser, a search bar, an index of pages with links to websites. Behind that door, every company built the same machine: the website as the place where business happens, with search, social and email as the channels that keep it filled with visitors. Webshops, news sites, insurers, travel platforms: the entire online economy rested on one unspoken assumption. A human being would walk through that door and land on your website.
Here's the thing most people get wrong about the platform shift we're living through. The front door of the internet hasn't actually changed. It's still the same open web: one big library we call the index, full of links to web pages. What has changed is that we, as humans, no longer need to use that door ourselves.
Instead, we go to a platform like ChatGPT, Gemini, or Google's AI Mode, and that platform sends a machine through the door for us. A bot, a crawler, an agent. It explores the open web, reads it, interprets it, weighs sources against each other, and walks back out to present the answer directly to us. We drop our problem with an AI agent, and that thing searches (and increasingly buys) on our behalf. Without anyone ever seeing your website.
Which means the internet is no longer a destination you visit. It's an assistant you talk to. In ten or twenty years we'll look back on the mid-2020s as the era of AI and superintelligence, the same way we look back on the rise of the personal computer in the late nineties, the web in the early 2000s, and the iPhone in 2007.
And that's what is killing the open web as we know it. Not the pages, not the links. Those will still be there. What's dying is the human visit. The channels you spent twenty years mastering still exist, but they no longer end at your website: your owned media stops being the destination. Bot traffic is exploding while human traffic flatlines. Humans are becoming a minority among the readers of the web. The open web isn't disappearing. It's becoming machine food.
So the question that should be at the top of every company's agenda is this: when that machine walks back through the front door with an answer, is it carrying your brand, your product, your service? Because in this world, it no longer matters whether you're B2C or B2B. We are all B2A now: Business to Agent. Your biggest customer is no longer a human. It's a bot, acting on a human's behalf.
The rise of AI search: it's not coming, it's here
Everyone is talking about AI search. It's the future. It's coming. Get ready for what's ahead. Here's the problem with that framing: it's not coming. By the numbers, it's already here. At Follo we've been tracking the rise of AI search for months. Not on gut feeling or LinkedIn hot takes, but on data: Google I/O, Alphabet earnings, OpenAI's own numbers, Sensor Tower, Semrush, Backlinko, DemandSage, Similarweb. We pulled it all together into one picture. Here's what we found.
Early 2025, this was the landscape: traditional search (Google, Bing) at roughly 5,500 million monthly active users. AI search? About 310 million. A rounding error. Bing was bigger than ChatGPT, and we didn't even take Bing seriously.
Then it went fast. October 2025: 3,088 million. January 2026: 3,760 million. February: 4,494 million. May 2026: 5,975 million monthly active users on AI search platforms. That's not a typo. From 310 million to almost 6 billion in roughly one year.
And traditional search in that same period? Still 5,500 million. Flat. Not growing. At least, Google doesn't actually share those numbers anymore, and that silence has to be telling. When growth stops being a number you brag about in earnings calls, it usually stops being growth. Which means May 2026 is the month the lines crossed: there are now more monthly active users on AI search than on traditional search. The takeover isn't coming. On this metric, it already happened.
Now, before the data police arrive: yes, we know you can't blindly add up active-user numbers. Public sources mix weekly and monthly actives. ChatGPT's headline figure, for example, is reported in weekly active users, while Gemini reports monthly. We're aware, and we handled it deliberately: we treated ChatGPT's weekly number as an absolute minimum baseline for its monthly reach. Because no, you can't just multiply WAU by four. A user who comes back every week still counts as one unique monthly user. But a platform's monthly actives are by definition at least its weekly actives. So if anything, the chart understates AI search, not the other way around. And if one platform's weekly volume already matches a competitor's monthly volume, that's not an unfair comparison. That's the story. We also rely on public snapshots, so individual figures will wobble per source and per month. The direction of travel won't.
Look at who's behind those 6 billion and you see something remarkable: it's mostly Google. AI Overviews sit at 2,500 million users. AI Mode at 1,000 million. Gemini at 900 million. ChatGPT, the brand everyone talks about, at 1,000 million. The biggest AI search players aren't challengers attacking Google. They are Google, rebuilding search as an AI product in front of our eyes.
Now, the fair debate: AI Overviews live inside Google Search, so you could argue they're still traditional search. But be honest. It's a completely different experience. A different machine, a different answer, a different way your brand does or doesn't show up. These aren't new users; they're the same Google users now getting AI answers. So run it both ways: count AI Overviews as AI search and you get +1,827% growth in a year. Exclude them entirely and it's still +1,021%. Both numbers are insane. The conclusion doesn't change.
AI search is not the future of search. It's the present. Fewer clicks, more influence, and increasingly an agent in the middle doing the searching for us.
Your biggest customer is a bot
Fewer clicks, more influence, an agent in the middle. And here's the uncomfortable part: that's not even the real story. Because while we're all staring at user numbers (humans choosing ChatGPT over Google), something bigger is happening in the background. The fastest-growing visitor of the web isn't human at all.
The bot numbers
Cloudflare, which sees roughly a fifth of all web traffic, reported that AI bot traffic on its network grew 187% in 2025. Human traffic? 3.1%. Agentic AI traffic, bots acting on behalf of a user, grew 7,851% year over year. And their big boss, Cloudflare CEO Matthew Prince, predicts that bot traffic will overtake human traffic on the web by 2027.
And these machines take far more than they give back. Cloudflare measures this with a metric called the crawl-to-refer ratio: how many pages a platform's bot crawls for every single visitor it sends back to your website. The numbers are brutal. Anthropic's ClaudeBot runs at roughly 24,000:1 (Cloudflare Radar, Q1 2026). Read that again: for every 24,000 pages ClaudeBot downloads and reads from a website, exactly one user clicks a link inside Claude and actually visits that site. OpenAI's GPTBot does “better” at 1,276:1. Still more than a thousand pages consumed per visitor returned. Now compare that to the old deal. Googlebot sits at roughly 5:1: for every 5 pages Google crawls, it sends one human visitor back. That deal, your content in exchange for traffic, is what kept the open web alive for 25 years. The machines just ended it.
The machines decide
But the volume isn't the scary part. The scary part is what those machines do. They don't just read the web. They decide. Where Google gave you 50 to 100 clickable results to choose from, an AI gives you 3 to 5 recommendations. The agent reads everything, weighs everything, and hands you a shortlist. It decides which brands you see, which products you compare, and increasingly (through agentic commerce) which one gets bought. The machine isn't just answering your question. It's shaping what you want.
And this is only the opening act. Play it forward a few years: an assistant in your phone, your car, your earbuds. One that knows your preferences, your history, your budget and your agenda. In that world, the web no longer reaches you raw. The agent reads it for you, weighs it against everything it knows about you, and serves the result: the answer, the recommendation, the booking, the purchase. Opening a website yourself becomes the exception, something you'll do about as often as you walk into a bank branch today. And your brand? If it shows up at all, it shows up as a snippet in a story someone else is telling, on a surface you don't own and can't design. The interface between you and your customer is no longer your website. It's their agent.
Notice what's actually happening here: a new layer has formed on top of the web. Between your company and your customers now lives a whole population of machines. Crawlers, bots, assistants, agents. And everything flows through it: search, social, shopping, support. For twenty years we optimized the channels that connected us directly to people. Those channels still exist, but they all end in the same place first now. The agent layer. And whatever lives in that layer decides what gets through to the human on the other side.
Gartner put it bluntly in 2025: companies and brands must prepare for the day when their biggest buyers are bots, not people. That day isn't a decade away. We drop our problem with an AI agent, and that thing searches, compares, and buys for us. Without anyone ever seeing your website.
And that's the thing. Everyone keeps saying the same comfortable line: “just keep optimizing for your customers.” No. That era is ending. You're optimizing for the machine that advises your customer. The machine that builds the shortlist, that frames the comparison, that closes the deal. Your customer still makes the final call, for now, but the agent decides what's on the menu. Business to Agent. If the machine doesn't know you, your customer never will.
What this means for search marketers
For the SEOs
Let's start with the SEOs. Is this the end of SEO? No. But let's be honest: SEO was already getting harder before the agents showed up. We track keyword sets where 30 to 95 percent of all queries now trigger an AI Overview. You can hold a steady market position and still be invisible in 70 percent of the AI answers above your hard-earned ranking. The old chain (keyword goes from position 6 to 3, traffic goes up, revenue goes up) doesn't run on rails anymore.
And I'll say the quiet part out loud: attribution is broken. It was broken before AI search, by the way. You can never measure all touchpoints; most are invisible or untraceable. The channels that are easiest to measure (your paid channels) get a disproportionate share of the credit. Which means you're often paying for sales that would have happened anyway. Remember when Airbnb switched off its performance ads and 90 to 95 percent of sales just kept coming? Now add AI search on top: a customer journey that fragments across ChatGPT, Gemini, AI Mode and AI Overviews, where you see nothing in your analytics. Fewer clicks, more influence. If you only look at GA4, you'll conclude AI search is small. It's not small. It's massive. It's just invisible to the tools you're used to.
So stop measuring AI search the traditional way. Sessions, clicks and last-click conversions tell you what's left of a dying funnel, not what's happening in the new one. The metrics that matter now: AI visibility, answer share, citation share, share of conversation, branded search lift, and the revenue that does come through LLM traffic. Which, by the way, converts noticeably better, because those visitors arrive pre-educated by the machine.
For the PPC marketers
And if you do PPC and you've been reading this thinking “glad this is an SEO problem”: it isn't. Yes, the ad money is following the users. ChatGPT started showing ads to free users in February 2026 and opened a self-serve Ads Manager in May. Google runs ads in AI Overviews and AI Mode. But look closer and you see this isn't Google Ads with a chat interface. There are no keywords to bid on. You give the platform “context hints” and it decides when your ad fits the conversation. And one major player went the other way entirely: Perplexity killed its ads program, arguing that the moment users see sponsored content in an AI answer, they start doubting every answer. They'd rather sell trust than impressions.
That trust problem is exactly why the platforms are eyeing a different model: not selling attention, but taking commission. Sam Altman said it out loud back in March 2025. He'd rather not do traditional ads at all, but if you buy something through Deep Research, OpenAI charging “like a 2% affiliate fee” would be fine by him. A year later that's no longer a musing: OpenAI's Agentic Commerce Protocol takes a cut of transactions completed inside ChatGPT, and Stripe and Google have built the payment rails for agents to buy on a user's behalf. The entire ad model rests on impressions, clicks and scarce inventory. All three lose meaning when there's a commission on the outcome instead of an auction for attention.
And then the real head-scratcher: what happens when the one reading your ad isn't a person? If an agent does the searching, comparing and buying, then the “audience” for your message is a machine. It sounds absurd, but B2A advertising (Business to Agent) may well proliferate: advertising built to target agents instead of people. And it's less far-fetched than it sounds, because browsing agents like OpenAI's Operator literally load web pages the way a human does, ads included. An agent doesn't feel brand warmth. It doesn't get retargeted into submission. It weighs data: price, specs, availability, reviews, policies. So maybe the future of “paid” in agentic search isn't a banner for human eyeballs, but content engineered to persuade a machine: placement in the data the agent trusts, or a take rate for being transactable. Nobody knows yet. But if your entire playbook is bidding on human attention, you should be deeply curious about a world with fewer humans paying attention.
What this means for CEOs, CFOs and CMOs
Now zoom out from the marketing department. One question for the boardroom: what if all your customers use AI search? Not some. All. A year ago that was a thought experiment. With almost 6 billion monthly active users on AI search platforms, it's a planning assumption.
For the CEO
This is not a marketing tactic. It's a distribution risk. The agents that will intermediate your market are being trained right now, and the set of brands they know, trust and recommend is being decided right now. The penalty for waiting isn't a lower ranking. It's absence. The agent won't punish you; it will simply pick someone else, and your next generation of customers will never know you existed.
For the CMO
The uncomfortable truth is structural: your org chart is built on channels. An SEO team, an SEA team, a social team, an email team. But search, social and email are being absorbed into one agent-mediated layer that doesn't care about your departmental boundaries. Somebody has to own how visible (and how legible) your brand is to machines, across every channel at once. That's a discipline, not a side project for the SEO team.
For the CFO
The numbers you steer on are changing shape. Attribution was already shaky; in an agent-mediated journey it collapses. Measure total impact instead of fake precision. And watch the cost side: as platforms shift from ads to take rates on completed transactions, marketing spend moves from media budgets to something new. Paying for outcomes, and investing in the data infrastructure that makes your company buyable by machines in the first place.
Do we still need a website?
Fair question at this point: if nobody visits, do we still need a website? Hell yes, you do. But you need to understand what it's becoming. Your website is no longer a shop window for people. It's a source of truth for machines. The place where the machine gets its facts about you: your products, your prices, your specs, your policies, your proof. Because here's the thing about agents: AI isn't that smart, and AI agents aren't either. They need information. Without input, they hallucinate. So everything an agent claims about your brand, it has to find somewhere. And you'd better make sure it finds it from you, in a format it can read, instead of from a competitor's comparison blog. The playground has changed: you're no longer optimizing a destination for human visitors. You're optimizing a data layer for the AI layer that sits in front of everything. The website still matters. It just has a new boss reading it.
How do you optimize for AI search?
Collect lottery tickets
Here's the mental model I use. Once a user drops a prompt and the model starts synthesizing its answer, you have zero control. None. The model draws its winners from a pool of candidate sources, like a lottery draw. You can't influence the draw. The only thing you can influence is how many lottery tickets you hold.
So what's a ticket? Behind every prompt, the model fires off a set of synthetic queries: the query fan-out. “Best running shoes for a marathon” becomes a dozen searches behind the scenes. Every time a passage of your content ranks as a highly relevant answer for one of those synthetic queries, you earn a ticket. Every relevant passage, on every relevant surface, is one more entry in the draw.
Collecting tickets means three things. One: map the synthetic queries behind your prompts and optimize relevance at the passage level. Not the page level, the passage level. Two: stop relying on your own website. Your information needs to live across the whole ecosystem of owned, shared and earned properties. Three: go omnimedia. YouTube, Reddit, industry forums, third-party blogs and comparison sites. The machines look for consensus across the web, and a brand that shows up consistently everywhere holds far more tickets than a brand with one perfect website.
Welcome to the era of protocols
Which brings me to the debate. I'm one of the people who does not believe GEO is just SEO. Sure, there's overlap. Good rankings still feed the machine, structured content still wins. But let me ask you: when was the last time you optimized for WebMCP to improve your visibility? Have you ever exposed your pricing through an MCP server, or built a product feed for an agent instead of for Google Shopping? Exactly. This is a different game with different rails.
Welcome to the era of protocols. MCP (Model Context Protocol) lets agents securely tap into your data and tools (inventory, pricing, availability) instead of scraping your HTML. WebMCP brings that same capability directly to your website, so agents can use your site's functionality. A2A (Agent2Agent) lets agents talk to each other: your customer's agent negotiating with your brand's agent. ACP (OpenAI's Agentic Commerce Protocol) lets users buy inside ChatGPT, with the platform taking a cut. AP2 (Google's Agent Payments Protocol) standardizes how agents pay on a user's behalf, with Stripe building the rails underneath. UCP (Universal Commerce Protocol) exposes your catalog and checkout to any agent. And A2UI/AG-UI define how agents render your brand inside their interface. None of this is optional for long. The models and agents being trained today will be answering your customers' questions for the next decade, and the brands they learn now are the brands they'll recommend later.
The playbook: feed the machine
So what do you do on Monday morning? Five moves.
1. Turn your website into a data layer. Remember that new layer where the agents live? It needs to eat. Your website is no longer the destination on top of everything; it's the foundation underneath. The single source of truth the agent layer draws its facts from. So stop hiding information for lead-gen reasons. Put your prices, specs, availability and policies out in the open, complete and current, in one place. If an agent can't find a fact about you on your own website, it will get that fact somewhere else. And you won't like the source.
2. Make your data machine-ready. Having the facts online isn't enough; machines have to be able to parse them. That means structured data and schema, clean page structure with clear headings, extractable passages instead of walls of text, product feeds built for agents and not just for Google Shopping, and APIs or MCP endpoints where it makes sense. Google's Developer's Guide to AI Agent Protocols is a good map of what's coming. The rule of thumb: if a machine can't read it, reason over it and cite it, it doesn't exist.
3. Cover the whole journey: In the Answer, In the Mind, In the Wallet. In the Answer is the entrance ticket: the orientation prompts where the shortlist gets made. If you're not mentioned, you simply don't exist. In the Mind is head-to-head: comparisons, reviews, validation, where trust is won or lost. In the Wallet is conversion: the deal hunt, still in Google and increasingly inside ChatGPT itself. Each phase has its own prompts, and each phase has its own KPIs.
4. Measure presence. Your traffic numbers tell you how the old channel is shrinking, not how the new one is growing. The new scoreboard: Mention Answer Share, Citation Answer Share, your market position against competitors, bot activity in your logs, and the LLM traffic and revenue that does come through. Build that dashboard now and set your baseline. You can't manage what you only discover in hindsight.
5. Make it a company discipline. Don't park it with the SEO team. This work spans content, data, product and brand. Your SEO people are the best positioned to lead it, but the mandate has to come from the top. Because the decisions involved (APIs, feeds, pricing transparency, who owns machine visibility) aren't theirs to make alone.
There is no second wave
The agents that will sit between you and your customers are being trained today. The brands they learn now are the brands they'll recommend for years. And there's no catching up later: latecomers aren't ranked lower. They're simply not in the answer.
So make sure the machines can find your products, your services and your content, and can choose them, with the right data. Because then I'm convinced that when that machine walks back through the front door of the internet, it will be carrying your brand, your company, your product.
Be there.
