A Human Posing as AI: What Your Chat Widget Must Disclose

A San Francisco billboard advertised “the leading chat interface powered by AI.” The AI turned out to be one guy typing. WIRED told the story as an art project, which it is, but buried inside it is the clearest cost breakdown of human-versus-automated chat support anyone has published this year, plus a live demonstration of the disclosure rule that took effect across the EU eleven days ago.

“Bryant didn’t train an LLM to respond based on how he writes. He literally sits down and answers the prompts himself. If you request an image of a skunk wearing sunglasses, he’ll scribble one up. If you ask how to fry an egg, he’ll respond with a suggestion.”

— Caroline Haskins, WIRED

Our take

Tucker Bryant accidentally ran the experiment every business owner should run before buying a chat product, and the numbers in WIRED’s own reporting tell you exactly why “just hire humans to answer chat” has never worked.

Do the arithmetic the article leaves on the table. Bryant reported roughly 30,000 queries by August 6 and a peak of about 5,000 prompts per hour. Assume a fast three minutes per response, which is generous for anything involving a drawing. Five thousand prompts in an hour is 250 person-hours of work generated in sixty minutes. You would need 250 people sitting down simultaneously to keep pace with one hour of demand.

Price it. The U.S. Bureau of Labor Statistics puts the median wage for customer service representatives at $20.59 per hour as of May 2024. That peak hour costs about $5,150 in straight wages before benefits, software, or supervision. The full 30,000 queries represent roughly 1,500 person-hours, or about $30,900. Bryant’s revenue from ChatTJB Pro, per WIRED, is four subscribers at $5 per month. Twenty dollars.

He solved it the only way a viral art project can: 10 volunteers, selected from over 3,000 applicants. That is not a staffing model. It is a lottery win, and it is the part of the story that does not transfer to your business at all.

So here is the honest conclusion, and it cuts against the sentiment of the piece: the demand ChatTJB proved is real, the delivery method is not reproducible, and the correct response for a small business is not to replace your bot with people. It is to stop pretending the two are interchangeable. Bryant’s users, by his own account, valued “just getting a human to bounce those questions off of, as opposed to getting super expert advice.” That is a routing problem, not a hiring problem. Most inbound chat is informational and should be automated. A thin slice is emotional or high-stakes and is worth a person. The failure mode we see constantly is businesses that automate the second category and staff the first.

Now the part with legal teeth, which WIRED had no reason to cover. Bryant’s billboard called ChatTJB “AI” and the site quietly redefined AI as “average individual.” Funny as art. As commerce, that framing is a compliance problem in at least two jurisdictions.

California’s Bolstering Online Transparency Act, SB 1001, has been in force since July 1, 2019. It makes it unlawful to use a bot to communicate with a person in California with intent to mislead about the bot’s artificial identity in order to incentivize a purchase or sale of goods or services, and it requires disclosure that is clear, conspicuous, and reasonably designed to inform. Perkins Coie’s analysis notes the California Attorney General can seek penalties of up to $2,500 per violation. The law runs one direction — bots posing as humans — but it establishes the principle regulators now apply in both.

The bigger change is fresher. Article 50 of the EU AI Act took effect August 2, 2026 — less than two weeks ago. Systems that interact directly with people must be designed so users are informed they are dealing with AI, and the notice must arrive at or before the first interaction, be clear and distinguishable, and meet accessibility requirements. Fine print does not count. If your site takes orders from Europe, this applies to your chat widget today, not at some future compliance date.

Between those two rules sits the FTC, which announced Operation AI Comply on September 25, 2024 with five enforcement actions. DoNotPay settled for $193,000 over claims its “AI lawyer” could substitute for human legal expertise. Then-Chair Lina Khan’s line was blunt: “There is no AI exemption from existing laws.” Overstating what your AI does is a deception case. So is overstating that a human is involved when one is not — the “concierge” or “dedicated advisor” widget that is actually a decision tree.

The uncomfortable second-order effect Bryant’s project surfaces: as AI-labeled chat becomes universal, “you are talking to a real person” becomes the scarce, valuable claim. Thirty thousand people wrote to a stranger for slow, admittedly mediocre answers because a human wrote them. That is a positioning opportunity for service businesses — and the instant it becomes a selling point, it becomes a material claim you must be able to prove. Log it accordingly.

What this means for your business

• Add an AI disclosure to your chat widget this week if you sell into the EU. Article 50 has been live since August 2, 2026. One sentence in the opening message — “You’re chatting with an automated assistant; ask for a person any time” — satisfies the disclosure and, in our experience, reduces the angry escalations that come from users who feel tricked.
• Audit your marketing copy for human-implying language. Words like advisor, concierge, specialist, and team next to an automated widget are claims. Either staff them or rename them. This is the same substantiation standard the FTC applied in Operation AI Comply.
• Run the 250-hour math on your own volume before you buy a chat tool. Take your peak-hour message count, multiply by your realistic handle time in minutes, divide by 60. That is how many staff-hours per hour you would need. Compare against $20.59/hour median wage. The number decides automate-versus-staff faster than any vendor demo.
• Route by emotional stakes, not by volume. Automate order status, hours, pricing, and scheduling. Hand billing disputes, complaints, cancellations, and anything involving a deadline to a person within one exchange. Most businesses have this backward.
• Publish a one-line AI policy page and link it from the chat window. State what is automated, what is human-reviewed, and how to reach a person. It costs an hour, covers the disclosure requirement in multiple jurisdictions, and gives your sales team something competitors mostly cannot show.

Caroline Haskins’ interview with Tucker Bryant, including his thoughts on “cognitive surrender” and why he kept the project running, is worth your time: read the original story at WIRED.

---