Shape at the document level
75 min
Two hosts talk the lesson through. The voices are synthetic; the script was written from this lesson and checked against it, and asserts nothing the lesson does not.
- Write headings that let a reader decide whether to stop, rather than naming a category
- Choose between prose, a list and a table by asking what relationship the content has
- Restructure a document without rewriting it, and test the result using its headings alone
Lesson 1 established that a reader takes in about a fifth of an average page, and gets through half the information only on pages of about 111 words or fewer. That finding has been driving everything since, but it has been driving it at the level of sentences and paragraphs, which quietly assumes the reader is reading.
Often they aren't. They're looking. They've opened a document to find one thing, they're moving down it at speed, and prose isn't what they're using to find their way.
What they are using is the part of writing that most writing courses do not teach, on the grounds that it belongs to somebody else. The US federal plain-language guide does not treat it that way: one of its four parts is Design for understanding, a quarter of the whole subject, and its opening claim is blunt. "Dense, cluttered writing deters people from taking the time to read."
Headings are the document a skimmer actually reads
The test for a heading is short, and it disqualifies most of the headings you've written.
A heading has to give a reader enough to decide whether to stop.
Now look at the ones everybody uses. Background. Context. Introduction. Next steps. Considerations. About this document. Every one of them is a category label. Every one would fit almost any document in the world, which means none of them tells a particular reader anything about this one.
The federal guidance asks for two things: "use heading levels consistently and appropriately to create hierarchy", and "write heading terms that are clear and descriptive". The second is the one people fail. Descriptive means it describes this content, not the kind of content.
GOV.UK's guidance on structure is more specific, and it's written by people who have to get this right across a whole government. It asks for headings that are "descriptive", "frontloaded", "active" and "removable", and tells writers to "avoid generic headings like 'Introduction'". Three of those four are this section so far. The fourth is the one worth stopping on.
"Removable – the content should still make sense with the headings removed." That sounds like the opposite of everything above, and it isn't. Both things have to be true at once. Strip the headings and the prose still has to work, because plenty of readers, including anyone listening rather than looking, will meet it that way. Keep the headings and they still have to tell the story on their own, because that's what a skimmer gets. A document that fails the first test has put content in its headings; a document that fails the second has put labels there. You're aiming at the narrow target where both hold, and the consultation note below hits it: pull the four bold lines out and the paragraphs still say everything.
The same page asks you to "put the most important information first", and recommends the inverted pyramid by name. That's lesson 2 arriving at document scale, and it decides the order your sections go in before any of this decides what they look like: the section a reader most needs goes first, and "Background" almost never is it.
| Category label | Same section, described |
|---|---|
| Background | Why the old system was replaced in April |
| Next steps | What we need from you by 14 March |
| Considerations | Three things that could delay this |
| Analysis | Costs came in eleven percent over budget |
| About this scheme | Who can apply, and who cannot |
Read down the right-hand column on its own. You now know roughly what that document says, which you can't do with the left-hand column at any length.
A report has these headings: Introduction, Methodology, Findings, Discussion, Conclusion, Appendices. Is that wrong?
Show the answer
Not necessarily, and this is worth being careful about. In a scientific paper those headings are a convention that readers know how to use, and a reader looking for the method knows exactly where to go. The labels are doing real work precisely because the structure is standardised across thousands of papers.
The failure is when those headings are imported into a document with no such convention behind it, which is where most of them live: internal reports, proposals, briefing notes, project updates. There, "Findings" tells a reader nothing, because they have no idea what kind of finding to expect.
So the rule isn't that category headings are banned. It's that a category heading is only doing work when the category is a convention your reader already knows the shape of. Otherwise it's a word occupying the most valuable line on the page.
Prose, list, or table
These three are not stylistic choices. Each one makes a different claim about the content, and choosing the wrong one tells the reader something untrue.
Prose says: these things are related, and the relations matter. Because, therefore, although, unless. Prose is the only form that can carry an argument, because an argument is made of joins.
A list says: these items are parallel and independent. Each one stands on its own and could be reordered without loss. The federal guide's advice is to "use lists for easy scanning", and that is exactly what parallel independent items are good for.
A table says: this has two dimensions. Every item is described by the same set of properties, and the reader will want to compare across them. The guidance is to "where appropriate, use tables to make complex relationships clear", and the operative word is relationships, in the plural.
The choice follows from asking one question: what relationship does this content actually have? Not what would look tidier.
A bulleted list claims the items are parallel and unordered. A numbered list claims they are in a sequence that matters.
Numbering a set of unordered options implies a priority that is not there, and a reader will look for it. Bulleting a set of steps that must happen in order throws away the part that mattered most. Neither is a formatting preference; each is a claim about the content.
A note as it arrived, and the same note given shape
This note arrived last week. Every word of it is fine.
The consultation on the parking scheme closed on 30 April. We received 412 responses, which is roughly four times what we normally see, and about a third of them came in during the final week after the local paper covered it. The main objection, raised in 260 responses, was the cost of a residents' permit. The second most common, in 114 responses, was the proposed boundary, particularly on the east side. A smaller number, 38, objected to the enforcement hours. We are proposing to reduce the permit charge from £90 to £60, to move the eastern boundary to exclude Mill Lane, and to leave the enforcement hours as consulted. The revised scheme will go to committee on 12 June. If members want a different option on the permit charge, we would need to know before 28 May in order to model it.
Before you look at the shaped version, decide for yourself. The note has four parts: the response figures, the objections, the proposals, and the committee timing. Which of them becomes a table, which a list, and which stays as prose?
Table, list, or prose, for each of the four parts?
Show the answer
The objections become a table: three things of one kind, each with a count, and the reader's question is which is biggest. The proposals become a bulleted list: three parallel, independent changes that could be reordered without loss. The response figures stay as prose, because "four times what we normally see" and "a third of them came in during the final week" are related to each other and to the newspaper coverage. And the committee timing stays as prose, because it contains a condition and a consequence: if members want a different option, they must say so by a date. A list would have cut that join.
Now the same note given shape. The four headings are the only new words. The two prose sections are the note's own sentences, unchanged. The table cells and the bullets are its own phrases, lifted out of the sentences that held them.
The consultation closed on 30 April, four times the usual response
The consultation on the parking scheme closed on 30 April. We received 412 responses, which is roughly four times what we normally see, and about a third of them came in during the final week after the local paper covered it.
What people objected to
Objection Responses the cost of a residents' permit 260 the proposed boundary, particularly on the east side 114 the enforcement hours 38 What we propose to change
- reduce the permit charge from £90 to £60
- move the eastern boundary to exclude Mill Lane
- leave the enforcement hours as consulted
What we need from members, and by when
The revised scheme will go to committee on 12 June. If members want a different option on the permit charge, we would need to know before 28 May in order to model it.
Something did get deleted, and it's worth seeing what. The note said "The main objection, raised in 260 responses, was the cost of a residents' permit", and the table row says "the cost of a residents' permit, 260". The main objection, raised in, was has gone. So has the second most common and a smaller number. Those words were doing a job in the prose: they told the reader which count was biggest and that the three were being ranked. In the table, the structure does that job, because the rows are in order and the numbers are in a column the eye can run down.
That's the trade in one example. A table deletes the words the prose needed to carry the comparison, and takes over the carrying itself. When the structure can't carry it, those words have to stay, which is why the last section is still prose.
Check the counts while you're here, because this course keeps telling you to. 260 and 114 and 38 come to 412, the number of responses, so the table accounts for every one and nobody is counted twice. Had those three summed past 412, the table would be making a claim the note never made, and any reader who added them up would have caught it.
Three things happened, and each was a decision about content rather than about looks.
The three objections became a table because they are one kind of thing measured on one axis, and the reader's question is which is biggest. The three proposals became a bulleted list because they are parallel and independent, and reordering them would lose nothing. The last section stayed as prose because it contains a condition and a consequence, which is a relation, and a list would have severed it.
And the headings tell the story on their own. Read only the four of them and you know what the note says.
One honest note about the demonstration. The bold lines above are standing in for real headings, because a block quotation can't contain them. In a real document they'd be proper heading elements, which matters for more than looks: a screen reader announces a heading and lets a listener jump between them, and bold text is just bold text. If you take one thing from this section into your own documents, make it that the heading is a heading.
The wrinkle: bullet points can destroy an argument
Bullets are the default in a great many workplaces, applied to everything. This is what they cost when they're applied to the wrong thing.
Original:
We should delay the launch until July. The regional teams have flagged capacity problems over the summer, and although the Q2 figures are strong, they are strong in exactly the regions that would be short-staffed. A May launch would therefore put our thinnest support cover against our highest demand.
Bulleted:
We should delay the launch until July.
- Regional teams have flagged capacity problems over the summer
- Q2 figures are strong
- The strong regions are the ones that would be short-staffed
- A May launch puts thin support cover against high demand
Every fact survives. The argument does not.
Three words from the original are missing from the bullets. Which are they, and what was each one doing?
Show the answer
Although, exactly and therefore. Although told you the strong Q2 figures cut against the case and were about to be answered. Exactly told you the third fact was the answer. Therefore told you the fourth was the conclusion. None of the three is a fact, which is why a bullet-by-bullet check finds nothing missing. They were the argument.
In the original, although tells you the second fact cuts against the case and is being answered. Exactly tells you the third fact is the answer. Therefore tells you the fourth is the conclusion. In the bulleted version those three words are gone, and four claims sit side by side as if they were four separate observations of equal weight. A reader now has to work out which one is the objection and which is the reply.
A list is a claim that the items are parallel and independent. When they are not, the list is a false statement about the content, and the falsehood is invisible because it lives in the formatting rather than in any sentence.
What form does each of these need, and what relationship in the content decides it?
Show the answer
The five documents someone must attach to an application. A bulleted list. Parallel, independent, order does not matter, and the reader is going to use it as a checklist.
How three pension options compare on charges, access and risk. A table. Two dimensions, three options against three properties, and the reader's whole purpose is to compare across.
Why a project was late. Prose. The reasons interact, some caused others, and one of them is probably the real one while the rest are context. All of that lives in the joins, and a list would flatten it into a set of equally weighted excuses.
The steps to reset a password. A numbered list. Parallel in form, strictly ordered in content, and the numbering is the part the reader uses when they lose their place halfway.
A single caveat that applies to everything above it. Prose, and probably one sentence. A one-item bulleted list is a bullet that has nothing to be parallel to, which is a common tic and always worth deleting.
What people get wrong
"Bullets are always clearer." They are clearer for parallel independent items and worse for everything else. The bulleted argument above is the standard demonstration.
"Headings are for long documents." A 300-word email with two informative headings is easier to act on than the same email in one block. The threshold isn't length; it's whether the reader might want to go straight to one part, which is true of a great many short documents that carry a deadline.
"Formatting is the designer's job." Deciding that three things are a table and three others are a list is deciding what relationship the content has. Nobody applying formatting afterwards can make that decision for you, because it is a claim about meaning.
"A table is for numbers." A table is for two dimensions. Plenty of good tables contain no numbers at all, including the heading table earlier in this lesson.
"White space is wasted space." The guide's own claim is that dense, cluttered writing deters people from reading at all, so space that gets someone to start is not waste.
"I'll add the headings at the end." A heading written last is a heading written by someone describing a section they've finished rather than telling a reader what's in it, and that's the frame that produces "Background". Write them as you go, when you still know what you're about to say, and rewrite them at the end when you know what you said.
Practice
Use the document you've carried since lesson 1, or a longer one if that's become short.
Take 35 minutes over this.
Change no sentences in this exercise. You may move text, split it, and add headings. If you catch yourself rewriting, you're avoiding the exercise, because the point is to show how much can be gained without touching the prose. The worked example above obeys the same rule: its four headings are new, and nothing else is.
Go through and mark every stretch where the content is a set of parallel independent items, and every stretch where it has two dimensions. Convert the first kind to lists and the second to tables. Leave everything with a because or an although in it as prose.
Give every section a heading that would let a reader decide whether to stop. Then check each one against the test: could this heading sit unchanged on somebody else's document about something else? If yes, it is a category label, and rewrite it.
Now the test, and its mirror. Write out your headings, in order, on their own, with nothing else. Do they tell the story? A reader who saw only those, which is close to what a skimming reader gets, would come away with what? Then do it the other way: cover the headings and read the document without them. It should still make sense, because that's GOV.UK's "removable", and a document that collapses without its headings has put content in them.
If the heading list doesn't tell the story, the problem is usually one of two things. Either a heading is a label, or the document's sections aren't the shape of its argument. The second is more common and more useful to find.
Last, count. How many words did you move, and how many did you change? Your headings are new words and don't count against you; everything else does. If that second number isn't close to zero, do step 1 again on a fresh copy.
Where this goes next
Everything in the course so far has been a claim about readers, and you have taken those claims on trust from research this course has read for you. The next lesson turns round and looks at where writing rules come from in the first place, using the most famous writing book in English and the most famous attack on it.
It is not a detour. The reason this course teaches what it teaches, and ignores a great deal of advice you have probably been given, is a claim about which kinds of rule are worth following. Lesson 8 makes that claim explicit and shows you how to test a rule yourself.
Go deeper
- GOV.UK's "Create a clear structure for your content", which is this lesson written as an operating standard, with the four heading criteria and the front-loading rule in their own context.
- Digital.gov's "Design for understanding" in full. It also covers visuals and links, which this lesson doesn't touch, and it's short.
- Nielsen's "How Little Do Users Read?", the source of the figures lesson 1 built on and this lesson's premise rests on, worth reading once for its caveats.
- If you write for the web or for anything that will be read on a screen, the two government guides above are the whole of this subject as two organisations that measure their results actually practise it.
Sources
- Digital.gov, "Design for understanding", part of the US federal plain-language guide series, opened and read on 2026-09-11 for this review. The quoted guidance on headings, lists and tables, and the claim that dense, cluttered writing deters people from reading.
- GOV.UK, "Create a clear structure for your content". The four heading criteria quoted above, "descriptive", "frontloaded", "active" and "removable", the instruction to avoid generic headings, "put the most important information first", and the inverted pyramid. All four criteria are quoted, including "removable", which pulls against this lesson's own test and is taught as a tension rather than dropped.
- The reading figures behind this lesson's premise come from lesson 1, which derives them from Jakob Nielsen's "How Little Do Users Read?" together with the caveats stated there.
- The consultation note is written for this lesson. Its response counts were checked to sum to its stated total: 260 plus 114 plus 38 is 412. The restructured version was compared with the original word by word rather than asserted: the four headings are new, the two prose sections are verbatim, and the table and list cells are verbatim phrases with the ranking language ("the main objection, raised in", "the second most common", "a smaller number") deleted, which the lesson names and explains rather than hiding.
Check your understanding
This lesson has a 6-question quiz. Pass it and the questions come back on a schedule in Review, so what you learned stays learned. Your progress is saved in your browser; no account needed.