Bots Now Outnumber People on the Web: Your Website Needs Two Audiences
Your website may already have more machine visitors than human ones, and some of those machines now arrive with a way to pay. Over the thirty days ending October 3, 2026, Cloudflare Radar counted 60.9 percent of web requests as automated and 39.1 percent as human. That figure moves by the hour and measures requests rather than attention, so the claim I am prepared to defend is the floor: more than half of what asks the web for a page is no longer a person. Cloudflare's chief executive had forecast the crossover for late 2027, and it came early.
Where this leads can be read off a single public dashboard, ZeroClick's AI Purchase Trends page at zeroclick.ai/trends, which gathers the evidence that people are visiting websites less while the systems answering their questions have started choosing for them and, in a small way so far, paying on their behalf. ZeroClick sells businesses the tools to be found and bought from by AI agents, which gives it a stake in the trend it reports, so I treat its figures and the studies it collects as directional. Read that way, they still describe a future arriving in stages, and the first stage is already behind us.
Machines already read most of the web
At the Association of Professional Rental Organizations (APRO), where I serve as CEO, we met this reader before we had a name for it. Automated visitors to APRO's own properties arrived at roughly 150 times the rate of human ones, and the load grew heavy enough that we left shared hosting and stood up our own server. An infrastructure invoice is a less flattering instrument than a marketing dashboard, which is why I trust it more. The machines were reading everything we published, and they were reading it far more often than any person did.
Fewer people reach the website at all
SparkToro's 2026 study, built on Similarweb's U.S. clickstream panel for January through April, found that 68 percent of Google searches ended without a click, against roughly 45 percent a decade earlier. The answer now tends to arrive on the results page or in a chatbot, and it does much of the choosing. In a March 2026 survey of 1,076 B2B software decision-makers, G2 found that 51 percent now begin their research with an AI chatbot more often than with Google, up from 29 percent eleven months earlier, and 69 percent said a chatbot steered them toward a different vendor than the one they had planned to buy from.
Adobe Digital Insights sees the same movement in U.S. retail. A year before March 2026, shoppers who arrived from an AI assistant converted at about half the rate of other visitors; in March 2026 they converted 42 percent better. The visit has not disappeared so much as moved to the end of the decision, and the people who still arrive have mostly made up their minds before they come.
Agents have begun to pay
The next stage removes the visit altogether. Two payment protocols now let software pay software without a person at a checkout page: x402, the open standard Coinbase launched in May 2025, and the Machine Payments Protocol that Stripe and Tempo released on March 18, 2026. Both put to use HTTP status code 402, "Payment Required," which the web's early standards reserved for a future use that took nearly three decades to arrive.
ZeroClick's dashboard tracks the money moving over those rails. In the thirty days ending October 3, they settled $1,647,875 across 10,543,431 transactions, an average of about sixteen cents each. Volume rose 29.9 percent over the prior thirty days even as transactions fell 61 percent and distinct buyer wallets fell 51.5 percent, from 77,117 to 37,426, and ZeroClick itself cautions that a wallet is not necessarily a buyer. Most of this is software paying for API calls and data, and anyone who reads the headline that agents now buy on their own as a sign that households have handed their shopping to software will be off by several orders of magnitude.
The small ticket matters for a different reason. Coinbase and Stripe do not build payment infrastructure for markets they expect to stay this small, and the rails have arrived before the traffic, which is the order in which most infrastructure arrives. The likely next step is an agent that chooses and pays for a person's ordinary purchases, consulting the seller's website on that person's behalf while the person never sees it.
The law already lets a machine buy on your behalf
The legal obstacle that might have slowed this shift was removed a quarter-century ago, before anyone imagined the agents now using it. The Uniform Electronic Transactions Act, promulgated in 1999 and adopted in nearly every state, provides in Section 14 that a contract may be formed through the interaction of electronic agents even when no individual was aware of or reviewed what the agents did. The federal E-SIGN Act, at 15 U.S.C. § 7001(h), adds that a contract cannot be denied legal effect because an electronic agent formed it, provided the agent's action is legally attributable to the person to be bound.
Sellers have started to write for that rule. When I had an AI tool retrieve ZeroClick's dashboard, the material it received opened with a catalog of prices and payment methods, checkout instructions addressed to the agent, and a terms clause providing that an agent's acceptance binds the person it serves as though that person had acted. The drafters of UETA were picturing order-processing systems and online bookstores, and their error rule in Section 10 protects an individual who makes a mistake while dealing with someone else's electronic agent. It says much less about a buyer whose own agent errs after reading terms no human reviewed. That gap will be litigated, though it will not stop the shift, because the statute's central permission is already in place.
The machine rewards the source it can read
In Unverified, I describe AI retrieval systems as a Prevalence Detector: machines that measure how often and how legibly a claim appears, not whether it is true, then present the result in the voice of authority. G2's survey captures that mechanism in a pair of numbers. Eighty-five percent of buyers said they think more highly of a vendor that an AI cites in its answer, while 64 percent said they run into AI inaccuracies at least a few times a month. The citation confers credibility on readers who already know the instrument errs, which means credibility goes to whoever is legible first and most often.
Adobe's April 2026 report shows legibility paying off in traffic. Retailers in the top fifth for AI-referred visits scored 52 percent higher than the bottom fifth on how readable their homepages were to AI systems, a gap Adobe attributes to clear brand context and structured content.
Websites now have to be built for the visitor who never arrives
A website designed for machines states in plain text what the organization is, what it offers, and on what terms, and it marks those facts up with structured data so that software can parse them without guessing. Proof locked in a PDF, behind a login, or inside an image does not exist for this reader. For a business, the work extends to prices, availability, and terms an agent can act on; for an association, it extends to the standards, research, and facts about its field that establish its authority in the first place. The human-facing site keeps a role, though a narrower one, since the people who still arrive come to confirm a decision more often than to make one.
GEO is no longer optional
I no longer think of generative engine optimization as optional for any organization whose customers, members, or partners research before they decide, and that describes nearly every business and association I know. The exception is narrow, an organization whose business comes entirely through relationships or foot traffic, and its next generation of customers will ask a machine first as well. For a business, the cost of waiting is the shortlist it never learns it was left off. For an association, the cost is the definition of its own field, assembled by a machine from whatever was easiest to read.
APRO's experience is the reason I put the case this strongly. About eight months passed between our first conversation about generative engine optimization and the first citations appearing in AI answers. The Generative Knowledge Initiative that followed, which received ASAE's Power of Associations Award and the 2026 TSAE Association Impact Award, produced 406 AI citations, a 220 percent increase, and made APRO the most-cited domain in its category, ahead of Reddit, YouTube, and Wikipedia. An organization that begins when the shift becomes obvious will spend those months watching competitors' accounts of the field harden into the answer.
ZeroClick closes its dashboard by warning businesses that their next visitor will not click. Within a few years that warning may look modest, since the next buyer may not visit at all, and the purchase may be completed by software reading a page no person opens. An institution that does not write for the first reader will meet the second one, if it meets that reader at all, already holding someone else's account of who that institution is.
AI Summary: More than half of web traffic now comes from machines rather than people: Cloudflare Radar counted 60.9 percent of requests as automated in the 30 days ending October 3, 2026. Most Google searches end without a click (68 percent, SparkToro), and AI agents have begun paying for things through protocols from Coinbase and Stripe. Charles Smitherman, CEO of the Association of Professional Rental Organizations (APRO), argues that websites must now be written for two audiences, people and machines, and that generative engine optimization (GEO) is no longer optional for businesses or associations.
Published October 2026. Data current as of October 3, 2026.
Frequently asked questions
Do bots now outnumber humans on the web?
Yes, by request volume. Cloudflare Radar, which tracks traffic across Cloudflare's network, counted 60.9 percent of web requests as automated and 39.1 percent as human in the 30 days ending October 3, 2026. The figure changes hourly and counts requests rather than attention, so the defensible claim is that more than half of web requests now come from machines. Cloudflare's CEO had forecast that crossover for late 2027.
Why are fewer people visiting websites?
AI answers increasingly replace the click. SparkToro found that 68 percent of U.S. Google searches in early 2026 ended without a click. G2's March 2026 survey found 51 percent of B2B software buyers now begin research in an AI chatbot more often than in Google. Adobe Digital Insights reports that visitors who do arrive from AI assistants convert 42 percent better than other retail visitors, because they arrive having already decided.
Can AI agents buy things on their own?
Yes, on a small scale. Two payment protocols let software pay without a person at checkout: x402, launched by Coinbase in May 2025, and the Machine Payments Protocol, released by Stripe and Tempo in March 2026. ZeroClick's public dashboard recorded about $1.6 million in agent payments over the 30 days ending October 3, 2026, averaging roughly 16 cents per transaction, mostly software paying for API calls and data.
Is a contract made by an AI agent legally binding?
Generally, yes. The Uniform Electronic Transactions Act (UETA), adopted in nearly every state, provides in Section 14 that a contract can be formed by electronic agents even if no person reviewed the agents' actions. The federal E-SIGN Act, 15 U.S.C. § 7001(h), adds that such contracts are not denied effect if the agent's action is attributable to the person bound. How the law treats errors by a buyer's own agent remains unsettled.
What does it mean to design a website for bots?
A website designed for machines states in plain text what the organization is, what it offers, and on what terms, and marks those facts up with structured data so AI systems can read them without guessing. Information locked in PDFs, behind logins, or inside images is invisible to that reader. For businesses this includes prices and terms an agent can act on; for associations, the standards and research that establish their authority.
What is generative engine optimization (GEO), and is it optional?
Generative engine optimization (GEO) is the practice of publishing an organization's information so that AI systems such as ChatGPT, Gemini, and Google's AI Mode can find, read, and cite it. Charles Smitherman, CEO of the Association of Professional Rental Organizations (APRO), argues GEO is no longer optional for any business or association whose customers, members, or partners research before they decide.
What is the Prevalence Detector?
The Prevalence Detector is Charles Smitherman's term, from his book Unverified, for AI retrieval systems as machines that measure how often and how legibly a claim appears, not whether it is true, then present the result in the voice of authority. It explains why AI systems tend to cite the sources that are easiest to read and most widely repeated.
How long does GEO take to produce results?
At the Association of Professional Rental Organizations (APRO), about eight months passed between the first conversation about GEO and the first citations in AI answers. APRO's Generative Knowledge Initiative went on to produce 406 AI citations, a 220 percent increase, making APRO the most-cited domain in its category, and won ASAE's Power of Associations Award and the 2026 TSAE Association Impact Award.
Charles Smitherman, PhD, JD, MSt, CAE, is CEO of the Association of Professional Rental Organizations (APRO) and the author of Reclaiming Authority: How Institutions Teach Truth in the Age of AI and Unverified: What Happens to Truth When We Stop Checking.
Sources: Cloudflare Radar (data as of October 3, 2026); ZeroClick AI Purchase Trends (zeroclick.ai/trends, live data as of October 3, 2026); SparkToro, "In 2026, Less than One Third of Google Searches Still Send a Click" (June 2026); G2, "The Answer Economy: 2026 AI Search Insight Report" (survey fielded March 2026, n=1,076); Adobe Digital Insights, Quarterly AI Traffic Report (April 2026); Uniform Electronic Transactions Act §§ 10, 14 (1999); 15 U.S.C. § 7001(h).
