Small print has only ever worked because reading was expensive. A ninety-page lease, a forty-page software licence, the insurance renewal that arrives as a PDF nobody opens: each of them was drafted on the safe assumption that the person signing would skim the headline terms, trust the rest, and get on with their day. That assumption held for the best part of a century, and it held on both sides of the table, because a small business rarely had a lawyer with the budget to read every word either. The small print was a bet on your inattention, and until very recently that bet paid every time.

The research on how safe the bet was is almost comic. In a 2016 experiment, Jonathan Obar and Anne Oeldorf-Hirsch invited 543 people to sign up for a fictional social network whose terms included handing over their firstborn child as payment, and 98 per cent agreed without noticing. A few years earlier, Aleecia McDonald and Lorrie Cranor had estimated that an American who read every privacy policy she encountered in a year would spend around 201 hours doing it. Nobody was going to do that, and everybody drafting those documents knew it. I signed more supplier agreements and office leases in seventeen years of running an agency than I ever read properly — and I suspect most of us who’ve run a business would admit the same.

What’s changed is that the cost of reading has collapsed to nothing. Any of us can now drop that ninety-page lease into Claude or ChatGPT and ask it the question the drafter hoped nobody ever would: what in here would embarrass the other side if I understood it? Thirty seconds later we’ve got the auto-renewal that triggers unless we give notice inside a two-week window, the clause that shifts responsibility for the building’s plant onto the tenant, and the termination right that only runs one way. The document is the same one it was last year; what’s gone is the economics that made its length a defence, and the bet on inattention no longer pays.

Both sides of the table

The obvious win is for the buyer, and it’s worth taking. Before we sign the next lease, supplier contract, insurance renewal or set of software terms, we should have an AI read every word and report back on four things: what obligations it creates for us, what renews automatically and on what notice, where liability moves from them to us, and how we get out. Then comes the sharper question, which is what the drafter would least like us to have spotted, because that question tends to surface whatever would otherwise have cost us money in year three. It’s the same instinct as asking a vendor what its revenue number actually means, and it costs about as much. None of this makes the AI a lawyer, and I’ll come back to that, but it does change who walks into the negotiation knowing what the document says.

The less obvious consequence lands on us as business owners, and it’s the one I’d think harder about. If our own terms rely on nobody reading them, that reliance has just become a liability, because the people who can now read them in thirty seconds include our customers, the journalist writing about our sector, the competitor looking for something to put in a comparison table, and before long the agent shopping on a customer’s behalf, which will read our terms before it reads our homepage.

WeTransfer found this out in July 2025, when a terms update granting the company a licence to use uploaded files for training machine-learning models was noticed within days by the designers and videographers who send unreleased client work through the service; the clause was rewritten inside a fortnight, with the company insisting it had never trained on anyone’s files. Disney had a rougher time of it the year before, arguing that a widower couldn’t sue over his wife’s death at a restaurant in Disney Springs because he had once agreed to Disney+’s arbitration terms, and dropping the argument once enough people had read it. In both cases the terms had been sitting there for anyone to see. What changed was that someone looked.

Plain, honest terms used to be a cost: the lawyer’s time to write them, and the leverage given away by not tucking a clause into section 14(c). Now that our customers can audit us for free, a set of terms we’d be comfortable having read aloud becomes something we can put on the website and sell with, and honest drafting turns from a compliance chore into a competitive position. Ask yourself what a customer would find if they ran your contract through an AI tonight, and whether you’d be happy to explain each answer to them. If you wouldn’t, it’s cheaper to fix the document than to wait for the screenshot.

Where the honesty has to go

Two caveats keep this from overclaiming, and both of them matter. The first is that the other side has AI too, and they’re using it to draft, so the asymmetry moves rather than disappears: the question becomes who bothers to use the tools rather than who has the stamina to read, and I’m not sure how that settles. A large landlord’s legal team with a contract platform at its disposal can produce terms that are technically transparent and practically opaque, dense with defined terms and cross-references that a general assistant will summarise fluently and slightly wrongly. My guess is that the advantage tilts towards whoever is more careful rather than whoever is bigger, which would be a real change from the last hundred years, but I wouldn’t bet the business on the guess.

The second caveat is reliability, and here the evidence cuts both ways. An AI contract reviewer beat twenty experienced corporate lawyers at spotting issues in NDAs back in 2018, scoring 94 per cent against their 85, in 26 seconds against their 92 minutes. Six years later, Stanford researchers tested the leading legal research tools and found them producing false or misleading answers between 17 and 33 per cent of the time, despite marketing that promised otherwise. Both findings are true at once — the machine reads faster and more consistently than a tired human, and it also makes confident mistakes that a tired human wouldn’t. The claim in this article is that there’s no place left to hide, which is a different claim from there being no need for a lawyer. Use the AI to find the clauses worth asking about, and then ask someone qualified about the ones that could hurt.

What happens when everyone can read

Beyond either side of any single deal sits the effect I find most interesting. Courts and regulators have tolerated small print for a century partly because nobody reading it was universal, and the law made its peace with that. Consumer protection in the UK has required contract terms to be transparent, meaning plain, intelligible and legible, since the Consumer Rights Act 2015, and the older principle that a signature binds you to what you signed has always carried an unspoken assumption that reading the whole thing was impractical. When reading costs nothing, both positions come under pressure from opposite directions. A company will find it harder to argue that a clause was reasonable because it followed industry convention, once a customer can show in seconds that the convention was designed not to be understood, and the customer will find it harder to plead surprise at a term that a free tool would have surfaced for them.

Britain’s new subscription rules, in the Digital Markets, Competition and Consumers Act 2024, which will require renewal reminders and an exit as easy as the entry once they finally commence, look to me like the first legislation drafted for a world where the old bet had already stopped paying. Drafting culture will follow, because a clause that survives only while it goes unread stops being worth the lawyer’s fee once everyone can read. I expect terms to get shorter and stranger over the next few years, as businesses discover which of their clauses were doing real work and which were simply hoping.

So before you sign the next contract, and before you send your own out, have an AI read every word of it and tell you what it found. One of those two exercises will be uncomfortable, and that’s fine, because it’s far better to have that conversation with a machine this week than with a customer, a journalist or a judge next year.