Since March 2026, any writer publishing a book in the United States can pay ten dollars and receive a small circular seal for the back cover, one that certifies, in effect, that no machine wrote what is inside. The Authors Guild’s “Human Authored” certification, opened that month to all U.S. authors and publishers, works less like a spellchecker and more like a food label: a licensing agreement, an identity check for non-members, and a public, searchable registry where anyone can look up a title and confirm a person wrote it.
Guild members get the seal free. Everyone else pays per title and submits to verification.
A market for authorship
The certification began quietly, in a beta launch for Authors Guild members in January 2025, at a moment when Amazon’s bestseller lists were filling with AI-generated books assembled from scraped tropes and sold under invented author names, according to reporting at the time. The Authors Guild, which represents thousands of writers and counts prominent authors on its advisory board, had a straightforward interest in the project: protecting its members’ livelihoods from a market that could no longer easily tell their work apart from a bot’s. The mechanism is not a text scanner but a licensing contract: a book qualifies as “Human Authored” if the prose was written by a person, with narrow exceptions for spelling and grammar tools or research assistance, and the author signs an agreement that puts their name behind the claim. It is a promise with a paper trail, not a forensic result.
The same instinct shows up in art and photography
Writers are not the only creative workers building this kind of infrastructure. Cara, a portfolio and social platform founded by photographer Zhang Jingna, grew rapidly in 2024 as illustrators fled other platforms’ AI training policies. Cara screens uploads with detection technology and tags accepted images “NoAI,” while a partnership with the style-protection tool Glaze is meant to make an artist’s work harder for AI models to copy. The pitch is nearly identical to the Authors Guild’s: a visible mark that separates a human’s output from the flood around it.
Camera manufacturers have been building a parallel system for years. The Coalition for Content Provenance and Authenticity, founded in 2021 by Adobe, Arm, the BBC, Intel, Microsoft, and Truepic, embeds cryptographically signed metadata into a file recording who made it, when, and what tools touched it. Leica and Nikon added support for the standard, known as Content Credentials, in 2022; Sony and Fujifilm followed in 2024.
Proving realness is harder than printing a sticker
PetaPixel reported in June 2025 that the rollout of Content Credentials has stalled into a standoff: camera makers are waiting for publishers to display the provenance data before fully committing to embedding it, and publishers are waiting for cameras to embed it before building the tools to show it. Some phone makers’ approach to AI-editing metadata has compounded the confusion, tagging an image only when AI editing tools were used and leaving ordinary, untouched photos unmarked, the opposite of proving a real photo is real. A decade into the project, the outlet noted, most images published online still carry no provenance data at all. Proving a person wrote a paragraph turns out to be a matter of paperwork; proving a camera captured an unaltered scene requires an entire industry to move in step, and it has not.
The pitch banks on disclosure, not detection
The commercial logic behind all of this depends on a quirk of how people actually respond to AI content, and the data on that quirk complicates the case for “human-made” as a straightforward selling point. A 2024 study by the software company Bynder, which showed 2,000 consumers in the US and UK two unlabeled 300-word articles on the same topic, found that only half could correctly identify which one had been written by ChatGPT, and 56 percent said they preferred the AI-written piece when they did not know its source.
Told nothing, audiences show no consistent taste for human prose over machine prose. But attitudes flip once the label is attached: the same study found 52 percent of participants said they disengage from content they suspect is AI-generated, and roughly a fifth of respondents said AI-flavored web or social copy makes a brand look impersonal, untrustworthy, or lazy. Editors reviewing client copy see the same asymmetry played out line by line: readers rarely flag a sentence as wrong for being machine-written until they are told to look for it, at which point the same sentence reads as evidence of carelessness.
The trust these labels are chasing is already sliding
That gap between blind preference and labeled distrust is precisely the space certification programs are trying to occupy, and the market opportunity is shifting in real time. The Capgemini Research Institute found in mid-2023 that 73 percent of consumers said they trusted content made by generative AI; more recent Capgemini research has since tracked declining trust in AI tools generally, a slide the institute has flagged as a risk for brands. A small cluster of newer companies is stepping into that gap, among them VerifiedHuman, founded in 2023, which now operates field-specific certification standards for writers, musicians, and visual artists across dozens of countries. Others, like the entertainment-focused certifier Human Made, are built on the same premise: that a declared, checkable origin is now worth marketing on its own.
None of these systems can actually catch a determined liar.
The Authors Guild’s certification rests on a signed agreement and an identity check, not a scan of the manuscript. Cara’s detection tools can be fooled by anyone willing to disguise a file. Content Credentials metadata can be stripped by re-saving an image through the wrong piece of software.
What each of these programs sells is not proof that a human made something, but a party’s willingness to be held to that claim in public, on the record, with their name attached and a fee or a contract behind it. A seal on a book’s back cover says nothing about whether the writing inside is any good, only about who is willing to put their name to having typed it, a distinction editors have always had to make one manuscript at a time, seal or no seal, long before anyone thought to print the claim on a cover.
