The NewsRupt

A strange new price tier is appearing on subscription pages: the AI-free plan. Photography platforms, music services, stock libraries, e-book stores and a handful of software products now offer, for a premium, an assurance that no generative model touched what you are buying — that the images were photographed, the text written, the code typed by a person. The existence of this tier is one of the clearest signals yet of how the market is processing the flood of machine-made content. What the tier actually guarantees, however, depends entirely on fine print that almost nobody is reading.

The demand side is easy to explain and genuinely rational. Buyers of visual and written work have watched the supply curve shift violently: images that once required a photo shoot or an illustrator can now be generated in seconds, and markets have responded by discounting accordingly. For a commercial buyer, distinguishing human work from generated work has become a trust problem, and the AI-free plan converts trust into a line item. For creators, it is a survival mechanism — a way to segment the market rather than compete with generators on their own terms. Premiums observed so far are often modest, in the range of a fifth to a half above standard plans, which suggests operators see it as a positioning tool rather than a revenue engine.

The supply side is where the fine print lives. Almost no AI-free plan defines its terms precisely, and the definitional gaps are enormous. Does AI-free mean no generative models anywhere in the supply chain, or only none used at the point of creation? Is AI-assisted editing — upscaling, noise reduction, background removal, grammar repair — permitted, when those features have been standard in creative software for years? What about spellcheckers, translation memory, or a photographer whose camera applies computational processing so aggressive it is functionally synthesis? Several plans quietly permit AI in post-production while marketing reads as a purity guarantee. The distinction between "AI-made" and "AI-touched" is doing enormous unacknowledged work.

The verification problem is harder still. Few operators have any mechanism beyond the contributor's own declaration, and none can afford audits at catalogue scale. Detection tools — classifiers that guess whether text or images are machine-made — remain unreliable, particularly after ordinary editing, and false accusations carry their own reputational cost. What AI-free plans are actually selling today is an attestation, and attestations are only as strong as the consequences for lying. The plans that will matter are the ones that build verifiable provenance — camera-capture signatures, edit histories, cryptographically signed chains from shutter to sale — into the guarantee. The technology for this exists; almost no plan currently ships it.

There is a contradiction at the centre of the category that its marketers rarely acknowledge. Many of the same companies selling AI-free tiers are simultaneously selling AI-generation features on their standard plans, and in some cases training their models on the very contributors the premium plan protects. A photographer paying extra to keep AI out of a marketplace while the marketplace monetises her archive as training data is paying for a guarantee that runs in one direction only. Consumers reading these plans are entitled to ask whether the fee purchases a property right or merely a label.

None of this makes AI-free tiers a gimmick. Market segmentation around provenance is a legitimate response to a genuine supply shock, and the premium is a market signal that human work retains value precisely when its scarcity becomes visible. But a guarantee is only as good as its definition and its enforcement, and the current generation of plans has mostly neither. The ones worth paying for will be the ones that can prove their promise.

A limitation worth stating: plan terms analysed here are as published at the time of writing and are changing quickly; premium levels are drawn from advertised prices and vary by market. The state of AI-detection accuracy is assessed from published research, which lags deployed systems. No operator contacted made claims specific to this article.

The NewsRupt files this from the London desk. Corrections and right of reply are handled under our published standards policy.