A site owner put a one-cent price tag on his own pages and watched Claude Code pay it mid-task — while Google's new publisher-payment pilot hands you a monthly figure and won't show the math
Search Engine Journal · Suganthan Mohanadasan · September 17, 2026Patient acquisition
- Google's side, and it is news in its own right. The AI contribution pilot — confirmed by Google to Digiday on Sep 14, covered by Barry Schwartz at Search Engine Roundtable the same morning — pays publishers who accept the terms a monthly earnings figure in Search Console with no breakdown of how it was calculated. It pays only when a page contributed significantly while an answer was being generated in AI Overviews, AI Mode or the Gemini app; being linked after the answer is written does not qualify. Google has approached at least dozens of publishers, with smaller publishers showing more interest than large ones. One executive familiar with it called it a black box.
- The alternative he actually built. His page returns HTTP 402 Payment Required with a price attached; the agent signs a USDC authorization, a facilitator settles it on-chain, and the server returns the content plus a receipt. It runs on a Cloudflare Worker in roughly 300 lines of Hono code using the open x402 protocol, has been live since Aug 8, and five payments settled on the morning of Sep 15, each with a public transaction hash on a running earnings page.
- The refusal is the part worth stealing, not the payment. His buyer script enforces $0.05 per call and a $0.25 daily allowance, checked before signing — when the cap sits below the price it exits without ever creating a signature: `[refused] Price $0.0100 exceeds the per call cap of $0.0010. This agent does not negotiate.` He then wired it into Claude Code's WebFetch hook, so Claude hit a 402 during a task, paid one cent inside the limit, and carried on. "Claude can buy a page when it needs one, but it can't decide on its own that my five-cent limit no longer applies."
- The limits, stated plainly by the author himself. Every payment so far came from his own agents, in testnet USDC with no monetary value — no search crawler is paying anyone. GPTBot, ClaudeBot, PerplexityBot and Googlebot don't support this yet, so charging today simply blocks them and earns nothing. Cloudflare's Pay Per Crawl is closed beta and its Monetization Gateway is still waitlisted. And Decrypt reported Sep 13 that TRM Labs examined 198.9 million x402 settlements and concluded most of the volume was not coming from AI agents.
Why it matters for an independent practice: One — this is the first credible answer to "AI reads my client's content and sends nobody." And the split he lands on is the usable rule for a practice site: keep general explainers free, because being cited still has value, and reserve paid access for original datasets, research and tools. A regen practice's plain-English PRP explainer is discovery; its own outcome data is the asset, and those two should not be priced the same. Two — the independence angle is the whole point. Google's pilot pays what Google decides and shows no arithmetic; a priced request lets the site owner verify the request and the payment himself. Licensing deals scale to News Corp, not to a four-doctor clinic — a price in the request is the only version of this that scales down to one website. Three — and the stewardship read is the inverse of the usual one. The transferable idea here is not charging money, it is the budget check that happens before the signature. Every agent a practice deploys should be built to refuse first; "raise the cap to override" must be a human's decision, never the agent's. :warning: Disclose — author-level conflict, per this source's shape. Mohanadasan is reporting on his own demo site (`paid.suganthan.com`) and the post was originally published on his own blog, so the demo is self-promotional by construction. The Google-pilot facts are independently sourced to Digiday and Search Engine Roundtable, not to him. Search Engine Journal is a Candidate source — a vote here attributes to it. Search Engine Journal — "I Made My Website Charge AI Agents A Penny Per Page, Then I Watched Claude Pay It"
"Deployment means the technology is available. Adoption means clinicians voluntarily incorporate it" — and buried in part two is the likeliest HIPAA breach in a small practice this year: staff pasting patient records into public AI to write prior-auth letters
Healthcare IT News · Bill Siwicki, Managing Editor · September 17, 2026Practice operations
- The exposure, named. Dr. Stephen J. Morgan, practicing physician and CIO at Pediatric Associates of Greater Salem, Massachusetts, and a HIMSS Physician Community committee member: "when support staff upload patient information to a public AI system, say, to create a prior authorization letter, the output is useful but they are not realizing that this could be a HIPAA violation." :arrowright: No vendor sold anyone this. It is staff solving their own worst task with the tool that is already open in a tab.
- Why clinicians resist, from Dr. R. Ryan Sadeghian (chief AI officer, MITConn Advisors; physician committee co-chair, HIMSS Physician Community): "Physician resistance is not primarily a technology problem... Physicians are understandably skeptical because they have experienced years of technology that promised efficiency while frequently adding clicks, alerts and administrative responsibilities." Hence the distinction in the headline — available is not used.
- What he says to do before implementing anything. Pick a problem physicians actually want solved, establish baseline measurements, and define what success and failure will look like up front. At UToledo Health that means "smaller, measurable wins" through a Trusted Advisors model, concentrated on documentation, CDI, coding, denial reduction, prior authorization and medical education. And the governance line: "stop treating AI primarily as an IT project — AI implementation is clinical and operational transformation supported by technology," with governance that does not end at go-live because "as models and products change, organizations need a lifecycle approach."
- Morgan's design lesson is concrete and cheap to learn from. Developers rebuilt a school health form to match the state's paper form graphically, exactly — and it pulled so much unnecessary data out of the record that clinicians declared it unusable and it had to be redesigned. His conclusion: involve key users from initial specifications through UI and workflow, so the thing becomes "a feature that users actually want" rather than something meeting an enterprise or regulatory goal. Then, explicitly: "I believe this is true, not only in the large-scale institutions, but also the small independent practices."
- And the timing caution cuts both ways in the same breath: "we cannot be too slow to implement AI, in that younger patients are already expecting us to be using these tools," but "to rush to implementation without proper safety guardrails runs a significant risk of harm."
Why it matters for an independent practice: One — the prior-auth paste is a free fix and it is probably already happening. Prior auth is the most hated task in the building and a public chatbot is the most available relief; that combination is how a small practice generates a breach without buying anything. The remedy is two sentences and one tool: write down where PHI may and may not go, then give staff a sanctioned option that actually works — because a prohibition with no replacement just moves the behavior somewhere you can't see. Two — "deployment vs adoption" is the honest test for every AI thing sold into a practice, including ours. If no named clinician asked for it, it is a line item with a login, not a capability. Three — Morgan's small-practice sentence is the quotable asset. Independent practices are routinely told to copy health-system governance and reasonably ignore it; here a health-system CIO says the same discipline — involve your users, define success first — applies at four doctors as much as at four hundred. :warning: Shape note, said straight: this is Part 2 of the feature whose Part 1 ran yesterday — the same two named experts, still an interview rather than a study, which is the shape I flagged myself on 24 hours ago. I ran it because the HIPAA prior-auth warning is a specific action available this week and it was not in Part 1. If the same feature two days running is not wanted, :-1: and I will treat the second half of a two-parter as read in future. Healthcare IT News — "Effective clinical AI adoption starts before go-live"
