I counted the browser tabs open on my machine the other day and stopped at forty-seven. Each one is a thought I wasn’t finished with. That is working memory in the wild: not a philosophy about focus, a tab hoarder’s confession.

Now put your website on the other side of that equation and ask what it costs a visitor to hold your page in mind.

This is the deep-dive behind the working memory section of How Humans Read Websites.

The number, and the argument about the number

The classic figure is seven, plus or minus two, from George Miller’s 1956 paper that every psychology student meets (full text preserved at York University).

The modern picture is smaller. Nelson Cowan’s 2001 review re-examined the literature and argued the true capacity of the focus of attention is closer to four chunks, and a generation of working-memory research has clustered around that estimate since.

Which one is right? Genuinely contested, and it depends partly on what you count as a chunk and how you measure. I am not going to hand you a settled number, because there isn’t one.

What matters for web design is the floor: design for four and the sevens are covered. Design for seven and a chunk of your audience is already drowning.

What spends the budget

The four slots are shared by everything currently on screen. Your headline is holding one. Your navigation is holding several, depending on how many items you stacked in it.

The cookie banner is holding one hostage until dismissed. The promo popup, the sticky header, the form field the visitor is halfway through, the thing they were still carrying over from the previous page. It all draws from the same tiny workspace, and it all draws at once.

This is why “just add one more banner” is never just one banner. The budget doesn’t grow. The new element takes a slot from something else, usually from the paragraph the visitor was reading, which is why they arrive at the end of it with no idea how they got there.

Why walls of text are expensive

Reading a 400-word paragraph is not one act of memory. It is the visitor holding hundreds of words in a four-slot workspace and assembling the meaning by hand, live, without dropping a piece.

Skilled readers do this constantly and never notice the machinery, the same way you don’t notice your legs on a staircase until the staircase gets long.

Some visitors will do that work. Most will correctly conclude that your page is not going to pay them back for the effort, and the tab closes with your paragraph half-assembled. Nothing was wrong with your ideas. The packaging charged a deposit the visitor refused to pay.

The reframe

Chunking is not dumbing it down. A tight list of five bullets is not five items in memory; it is one chunk labeled “list of reasons,” with the details available on demand. You are not losing depth. You are moving the assembly cost from the visitor to the page, where it belongs.

Chunking that actually works

Chunking fails when it’s decorative. White space around an undisciplined paragraph is not a chunk, it’s a margin. The version that works has rules:

  • One idea per paragraph, three or four lines. If a paragraph needs “and also,” it is two chunks pretending to be one.
  • Lists for genuine lists. Three to five items, each item grammatically complete, because scanners read the items as sentences anyway.
  • Grouped navigation. Eight top-level items are eight claims on the budget; four groups of two are four chunks, and the visitor only opens the group they need.
  • Progressive disclosure for depth. Summarize first, link to the full treatment, and the visitor spends a slot only when they choose to.

That last rule is how this blog handles it too, for what it’s worth. Every overview page links to its deep-dives and carries nothing but the summary weight.

But experts can handle more, surely

They can, in their own domain, and it doesn’t transfer. Expertise works by chunking: a chess master sees one board configuration where a novice sees thirty-two pieces.

Your visitors are experts at their business, not at your website, and your pages are not their domain. The senior engineer drowning in your nine-item services menu is an expert at distributed systems, not at your information architecture.

There is also the simpler point, made by Nielsen Norman’s plain-language research: even experts complete tasks faster when the reading load drops. The four slots belong to everyone. The smart ones just notice sooner when a page wastes them.

Scanning behavior is the visible half of this story (the F-pattern piece covers what it looks like), and working memory is the machinery underneath. If your page structure respects the budget, the rest of the checklist mostly follows.

And if you are now looking at your homepage counting its slot-claimants like I counted tabs, that instinct is correct. Bring me the count and we’ll cut it down to what a visitor can actually hold.

Sources & further reading

  1. Cowan, N. (2001). The Magical Number 4 in Short-Term Memory. Behavioral and Brain Sciences, 24(1).
  2. Miller, G. A. (1956). The Magical Number Seven, Plus or Minus Two. Psychological Review, 63(2). (full text preserved at York University)
  3. How Little Do Users Read? — Nielsen Norman Group, 2008
  4. Plain Language Is for Everyone, Even Experts — Nielsen Norman Group