All stories
Analysis /

Did the Plain Writing Act work?

The law telling federal agencies to write clearly turned sixteen last month. Thirty years of the Federal Register show the government's prose did get plainer — a decade before the Act, on the strength of a memo with a deadline, and not at all since.

By Joseph Kato LanguagePublic dataLaw

The Plain Writing Act of 2010 is a short law. It asks federal agencies for writing that is “clear, concise, well-organized,” and it gave them a year to start. It also contains, in the sentence that does the work, the word the government’s own guidelines say never to use: each agency shall use plain writing.

I build Vale, a linter for prose, so a law about prose is hard to leave alone. And the government happens to publish the perfect corpus for checking one: the Federal Register, every rule and notice it has issued, machine-readable back to the mid-1990s.

Reading the Act first saves some embarrassment. It covers letters, forms, notices, and instructions — documents about benefits, services, taxes, and how to comply. It says, in so many words, that a covered document “does not include a regulation.” And it rules out judicial review, so nobody can be made to follow it. The rules that fill the Register were never its target.

They were someone else’s. In June 1998 a presidential memo told agencies to use plain language in every proposed and final rule from January 1, 1999, and it said what that meant: common, everyday words; “you” and other pronouns; the active voice; short sentences. Three months after the Act, Executive Order 13563 repeated the demand: regulations must be “written in plain language, and easy to understand.”

So the Register has three dates to react to, and the memo’s list is a scorecard. I took the summary paragraph of 124,118 rules, proposed rules, and notices, and the full text of 3,732 rules — about 27 million words — from 1995 to 2026, and counted.

“You” arrived on schedule

Rules that address the reader as you

Share of sampled rules and proposed rules using you or your, anywhere and in the regulatory text itself.

  • Anywhere in the document
  • In the regulatory text
Show the data
YearAnywhere in the documentIn the regulatory text
199511.42.4
19965.22.8
19978.67.7
199811.93.8
199919.55.4
200036.48.3
200130.87.8
200231.68.7
200346.110.7
20045018.4
20055929.9
200657.124.7
200764.726.5
20085832.5
200970.618.4
201063.927.7
20116320.3
201265.226
201375.421.6
201471.321.3
20156120.6
201671.627.2
201773.528.4
201864.426.3
201967.520
202067.829.8
202166.934.6
202268.631.6
202379.545.6
202470.126
202565.826.9
202663.632.5

Source: Federal Register; counted in scripts/fetch-plain-writing.mjs

Before the memo, about 9% of rules said “you” anywhere. It was 12% in 1998 and 36% two years later, and by the five years before the Act it had reached 63%. Since then: 68%. The whole climb fits between the memo and the law.

Most of that is the front matter — you may submit comments, if you have questions — and it would be easy to wave off as boilerplate. But boilerplate is how an institution changes its voice, and the harder test moved too. In the regulatory text, the part that becomes law, “you” went from 4% of rules to 26%. A quarter of federal rules now bind you rather than the applicant.

The passive retreated

Passive voice in summaries

Passive constructions per 100 sentences in the SUMMARY paragraph, by document type.

  • Rules
  • Proposed rules
  • Notices
Show the data
YearRulesProposed rulesNotices
199536.443.444
199633.142.342.8
199735.34142.7
19983641.841.1
19993537.440.9
200032.23439
200132.331.738.5
200231.533.639.4
200329.731.240.2
200427.230.540.4
200525.226.337.6
200624.52637.2
200724.326.338.2
200823.925.436.4
200923.427.138.4
201023.224.136
201126.127.235.8
201226.42737.5
201325.326.635.9
20142524.537.1
20152424.836.2
20162423.132.6
201722.323.831.6
20182522.533.1
201925.422.632.2
202024.223.931.4
202125.823.631.1
20222523.730.7
202325.321.432.1
202423.622.130.9
20252320.832.9
202622.518.430.7

Source: Federal Register; counted in scripts/fetch-plain-writing.mjs

Every Register document opens with a summary, and in the 1990s that paragraph was a wall of is being issued and are hereby amended. Rule summaries ran 35 passives per 100 sentences before the memo and 24 in the run-up to the Act; proposed rules fell further, from 42 to 26. This rule is being issued to amend became we are amending.

Then it stopped. Rules have averaged 25 since 2011. Notices, the one document type here that the Act names, fell least: 43 to 33.

“Shall” lost, then held its ground

The plain-language guidelines are blunt about one word: “Use ‘must’ not ‘shall’ to impose requirements,” because nobody uses “shall” in common speech and courts have read it to mean almost anything.

From shall to must

The median rule's share of must among its musts and shalls, and the share of rules whose regulatory text has no shall at all.

  • Median share of "must"
  • Regulatory text with no "shall"
Show the data
YearMedian share of "must"Regulatory text with no "shall"
199566.742.9
199666.755.6
19977055.1
199867.740
199980.648.6
200084.656.9
20017648.1
20028055.1
20038057.3
200491.367.1
200590.974
200687.566.7
200787.769.9
200883.361.4
200986.866.7
201093.265.1
201185.764.6
201283.357.5
201383.358
201483.968.8
201587.560.3
201691.772.8
201792.267.6
20187562.5
201991.765.3
202087.565.5
20218773.1
202287.774.7
202393.573.4
20248961.6
202583.366.7
202688.975.3

Source: Federal Register; counted in scripts/fetch-plain-writing.mjs

In the median rule, “must” was 68% of the two in 1998 and 81% in 1999 — the year the deadline fell — and 88% by the late 2000s. Regulatory text with no “shall” in it went from 48% of rules to 66%.

And there it sits: 87% and 67% since 2011. The other 33% of new regulatory text still reaches for the word, sixteen years into a law that reaches for it too.

The words got plainer, mostly

What happened to the legalese

Change in how often each phrase appears per million words of rule text, 1995 to 1998 against 2011 on.

Show the data
PhraseChange (%)
herein-71
prior to-51
in accordance with-45
notwithstanding-43
thereof-39
set forth in-38
promulgate-36
pursuant to-36
in order to-12
utilize-1
provided that26

Source: Federal Register; phrases from the Federal Plain Language Guidelines

The guidelines keep a list of phrases with a simpler swap: prior to is before, pursuant to is under. I tracked 16 of them. Together they fell 38%, from 2.9 per thousand words to 1.8 before the Act and 1.8 after it.

herein took the worst of it, down 71%. Prior to is down 51%. Pursuant to is down 36% and still the most common of the lot, which is fair: it is the phrase a rule uses to say where its authority comes from. Utilize did not move at all. Nobody ever needed it, and nobody has given it up.

The sentences never moved

Words per sentence

Mean sentence length in the SUMMARY paragraph, by document type.

  • Rules
  • Proposed rules
  • Notices
Show the data
YearRulesProposed rulesNotices
199527.528.228.8
199627.72827.8
199727.427.828.3
199827.32828.2
199926.927.327.7
200027.62728.4
200127.42728.4
200226.927.828.3
200327.728.128
200427.527.630.3
200527.127.528
200627.328.328.9
200727.428.728.5
200828.42927.5
20092828.728.5
201028.529.128.3
201128.430.128
201228.429.628.8
201328.129.129.2
201427.629.229
201528.530.529.5
20162929.729.1
201727.428.829.7
201826.928.829
201928.529.329.3
20202929.729.4
202128.728.929.6
20222828.629.8
202328.429.229.7
202429.129.530.1
202528.128.130.6
202627.528.630.9

Source: Federal Register; counted in scripts/fetch-plain-writing.mjs

This is the trait every style guide leads with, and it is the one that never budged. A sentence in a rule summary averaged 27.5 words before the memo and 28.2 since the Act, and in thirty-two years it has never left the band between 26.9 and 29.1. By the Flesch-Kincaid formula that is a grade of 17.5 — past the end of college, where it has always been.

Notices went the wrong way. They held at 28.4 words through 2010 and have lengthened since, to 30.9 this year, with 27% of their sentences now running past 40 words, up from 20%.

My guess at why this one resisted: you can swap a word or flip a verb without touching what a sentence commits you to. Splitting it means deciding which clause is the rule and which is the exception, and that is a legal judgment, not an edit.

It depends who is writing

Who says you

Share of each agency's sampled rules that use you or your, 2011 to 2026.

Show the data
AgencyRules (%)
Coast Guard99
Fish and Wildlife Service91
Federal Aviation Administration88
Environmental Protection Agency84
Food and Drug Administration58
Defense Department46
National Oceanic and Atmospheric Administration44
Federal Communications Commission41
Agricultural Marketing Service14
Internal Revenue Service9

Source: Federal Register; counted in scripts/fetch-plain-writing.mjs

The average hides a split. At the Coast Guard, 99% of rules address the reader; at the Internal Revenue Service, 9%. Same memo, same order, same guidelines. My read is that the Coast Guard and the FAA publish the same kind of rule hundreds of times a year — a safety zone, an airworthiness directive — from a template, and a template only has to be rewritten once. A tax regulation is drafted one at a time.

So did it work? In the Federal Register, no — and it was never going to, because the Act left regulations out and gave nobody a way to enforce the rest. Every line on this page that moved did so between 1999 and the late 2000s, and every one has been flat since the law passed.

What worked was smaller. A memo named four habits anyone could check, set a date, and the agencies that wrote from templates changed the templates. Three of the four habits took. The fourth needed someone to rethink a sentence, and thirty years on, it is still 28.2 words long.