The Readability Problem: Why Nobody Reads What They Sign

The Reading Gap Nobody Measures

Ask ten adults to name the last contract they read start to finish. Most will laugh. One might mention a lease. The rest describe a habit we've all quietly adopted: scroll, squint at the bolded parts, hit Agree.

We sign constantly, software terms, leases, vendor agreements, employment offers, NDAs, insurance policies, click-through consent screens. The volume went up. The reading didn't. What we have instead is a reading gap: the distance between what we sign and what we actually understand.

The reading gap isn't about laziness or intelligence. It's structural. The documents handed to us were written to be defensible in a courtroom, not decipherable at a kitchen table. And the gap widens every time a clause gets copied forward without a second look.

The U.S. Securities and Exchange Commission has published a plain-English handbook since 1998. Federal agencies have been required to write clearly since the Plain Writing Act of 2010. And yet the average consumer still can't tell you what they agreed to when they clicked through a 9,000-word terms page. When was the last time you read one before hitting Agree?

There's a number that shows how badly this scales. Researchers Lorrie Cranor and Aleecia McDonald calculated that reading every privacy policy the average American encounters online would take roughly 244 hours a year, about six full work weeks, spent doing nothing but reading disclosure text. Nobody does that. The estimate isn't a description of behavior; it's a measurement of the gap.

And that gap has a cost. Procurement teams run terms pages through summarizers before they sign. Sales reps field questions about auto-renewal language. Somewhere in the middle, a 30-word subordinate clause decides whether a customer gets billed for another year, or whether a small business absorbs a liability cap it never saw coming.

plain language is a live issue in 2026 because AI has made it cheap to analyze documents at scale, and what those tools keep finding is that the hard part was never locating the clause. It's understanding it.

A summary you don't verify is just a faster way to be wrong. That's the tension worth unpacking.

Long Contracts Aren't the Problem. Dense Ones Are.

Density, not length, drives comprehension failure. A 60-page lease written in short sentences is easier to work through than a two-page NDA that buries four conditions inside a single clause. Readers quit on syntax, not page counts.

You can measure this. The readability score most editors rely on is Flesch Reading Ease, which weighs average sentence length against average syllables per word. Its sibling, Flesch-Kincaid grade level, converts that into a U.S. school grade. Both formulas are public, take about ten seconds to run on any document, and the Flesch-Kincaid math hasn't changed in decades.

Regulators have used those scores for real. Florida requires life and health insurance policies to clear a minimum Flesch reading ease threshold before they can be sold in the state. New York's plain English law applies to consumer contracts of $50,000 or less, demanding words with common, everyday meanings, divided and captioned into readable sections. Readability isn't an editor's pet project. It's been policy for years.

Here's the shape of the problem in one sentence:

Notwithstanding anything to the contrary set forth herein, in the event that Customer fails to provide written notice of termination no less than thirty (30) days prior to the expiration of the then-current term, this Agreement shall automatically renew for successive terms of equal duration.

Fifty-one words. One period. Three stacked conditions. By the time you reach successive terms, you've lost the notice deadline. Complexity, not word count, is what breaks reading.

Four patterns show up again and again in dense documents:

• Nested conditions, if X, unless Y, in which case Z
• Nominalizations, provide notification of instead of notify
• Exceptions stacked on exceptions, often signaled by the words provided, however
• Disappearing actors, Customer's continued use shall constitute acceptance, which hides who is doing what

That fourth one is the sneakiest. When a sentence has no visible subject, no reader feels responsible for the outcome. The obligation lands on nobody in particular, which is convenient if you drafted it and expensive if you signed it.

Run the same content through a plain rewrite and the grade level drops several notches without losing a single obligation. Same duties, same deadlines, same remedies, just placed where the eye can find them. Ask any editor who's converted template prose into subject-verb-object sentences: the obligations survive the surgery. Only the fog leaves.

Why Legalese Survives, And Why It's Not Laziness

The obvious explanation, lawyers can't write, is wrong. Dense drafting persists for three reasons, and two of them are good ones.

First, terms of art carry case law. Phrases like “time is of the essence,” “reasonable efforts,” and “material breach” aren't decoration; they've been litigated for decades, and courts know what they mean. Rewrite reasonable efforts as “a solid try” and you've thrown away the precedent that told both sides where the line sat. That isn't clarity. That's guessing.

Second, drafting against ambiguity is expensive. A lawyer writing a clause isn't aiming at you. They're aiming at the worst-case reader, which is a judge deciding what the words meant four years later, with both sides arguing. Over-specify and the contract gets ugly. Under-specify and it gets litigated. Most drafters choose ugly.

Third, and this is the bad one, copying is safer than thinking. Most boilerplate exists because someone, at some point, was told never to touch the template. Boilerplate isn't carefully designed. It's inherited.

There's a fourth pressure that's finally fading. The billable hour rewarded length, and hourly work still dominates parts of the market. As more review and drafting moves to fixed-fee and subscription models, that incentive weakens.

Picture a typical mutual NDA. Two companies want to talk about a partnership. Neither lawyer gets paid extra for tightening a paragraph. So the document arrives with a governing-law clause, a residuals clause, and a definition of Confidential Information that runs three sentences without a period. Both sides skim it. Both sides sign. Six months later, one of them shares a document they swore they'd protect, and the only question that matters is whether that definition actually covered it.

Legalese is a dialect with a real vocabulary and real costs. The goal isn't abolishing it. It's knowing which sentences actually need the dialect and which are just wearing it as armor.

A useful test: would a court read this sentence differently if it were plain? If the answer is no, the dialect is doing nothing but slowing the reader down.

The Rewrite Test: One Clause, Two Versions

Here's a composite clause, modeled on language that routinely shows up in vendor agreements and subscription terms.

Original: Provider may, in its sole discretion, modify the fees payable hereunder upon thirty (30) days' prior written notice to Customer, and Customer's continued use of the Services following such notice shall constitute Customer's acceptance of such modified fees.

Plain: We can raise prices. If we do, we'll tell you at least 30 days ahead. Keep using the service after that and you accept the new price.

Same legal effect, half the words. Three sentences instead of one. Active voice, second person, no defined-term scaffolding.

Now the same treatment on the termination side:

Original: In the event that either party desires to terminate this Agreement for convenience, such party shall provide the other party with no less than sixty (60) days' prior written notice of such termination, whereupon this Agreement shall terminate and neither party shall have any further obligation to the other except for those obligations which by their nature survive termination.

Plain: Either of us can end this agreement for any reason. Give 60 days' written notice. When it ends, only the obligations meant to outlast it stay in force.

What changed isn't friendliness, it's findability. In the first original, the operative fact (you're accepting a price increase by doing nothing) sits behind a 30-word subordinate clause. In the plain version, the surprise arrives first. Same story in the second: “for convenience” lands at word four of the plain version and word fourteen of the original.

Plain language isn't dumbing down. It's placing the same obligation where a reader can actually trip over it.

One caveat: this is analysis, not legal advice. Some terms genuinely can't be simplified without losing meaning, indemnities, IP assignment, and liability carve-outs come to mind. The right question isn't “can this be shorter,” but “does this sentence need to be this hard?” Guidelines from plainlanguage.gov are a solid starting test, and the SEC's Plain English Handbook still holds up after 28 years.

Four edits do most of the work:

1. Put the consequence before the condition.
2. Break any sentence carrying more than two commas.
3. Replace noun stacks with verbs, provide notification of becomes notify.
4. Name the actor. If nobody is doing the thing, the reader can't tell who's on the hook.

Where AI Helps, And What It Can't Do For You

AI is genuinely good at the structural half of this problem, and it's already in the workflow. Thomson Reuters reported that 41% of law firms and 47% of corporate legal departments were using generative AI in 2026, up from 28% and 23% the year before. The most common uses are document review, legal research, summarization, and drafting briefs or memos.

That's real. semantic search means you can ask “what happens if I pay late” and get the clause that governs it, rather than every instance of the word payment scattered across 40 pages. Summarization collapses a long agreement into a page of obligations. That's a map.

But a map isn't the territory. AI reduces cognitive load on the first pass; it doesn't reduce it to zero. The tool can tell you a contract auto-renews. It can't tell you whether auto-renewal at that price is acceptable to you. That judgment has no model.

And the failure mode isn't hypothetical. In 2023, two New York lawyers filed a brief citing six court decisions that didn't exist, ChatGPT invented them, and nobody checked. A federal judge sanctioned them $5,000. Now swap “fake case citations” for “summarized renewal terms” and you can see the shape of the risk. Fluency isn't accuracy. A confident sentence is not a verified one.

There's a second wrinkle. One 2026 report found 61% of legal professionals say AI saves them time each week, but fewer than half of firms provide training on responsible use. Nearly three-quarters now use general-purpose AI tools for work. So summaries are flowing while verification habits lag. If you hand someone a summarizer without teaching them what summaries drop, you haven't fixed the reading gap. You've relocated it.

Two habits close most of the distance:

• Read the five clauses that control money, term, termination, liability, and dispute resolution, verbatim, not summarized.
• Treat the summary as an index, not a verdict. Every claim in it should point to a line you can check in under a minute.

Do those two things and AI stops being a substitute for reading. It becomes a way to decide what deserves a closer read.

Plain Language Is Coming, Slowly

The regulatory direction has been one-way for thirty years. The SEC adopted plain-English disclosure rules in 1998. The Plain Writing Act of 2010 made clear writing a legal requirement for federal agencies. EU consumer law requires contract terms drafted in plain, intelligible language under the Unfair Contract Terms Directive. Canada's accessibility legislation pushed federal bodies toward the same standard. None of it eliminated legalese, but the floor moved.

Commercial pressure may do more than regulation ever did. When two vendors sell similar products and one's terms page can be read in four minutes, the dense version starts to look like it's hiding something. Readability has become a trust signal, whether or not anyone intended it that way.

Tech already proved the model. Apple's App Privacy Labels, introduced in 2020, forced developers to compress data collection into a standardized card. The labels aren't perfect, critics argue they flatten differences that matter. But they shifted the default: disclosure got shorter because users had to be able to compare at a glance.

Watch what's happening in procurement. Buyers who've spent a year watching AI pull obligations out of unstructured text now expect to find the important sentence without a flashlight. A vendor that hands over a 40-page agreement with no summary and no plain-language version is now the slower option. And slowness reads as risk.

Regulation moved the floor. Commerce will move the ceiling. That expectation doesn't reverse itself.

The firms and vendors that win the next decade probably won't have the shortest terms. They'll have the ones you can actually finish, and the ones where the sentence that matters is the sentence you can find.