Layer cake pattern

Put subheadings down a page of text and the gaze plot separates into bands: a run of fixations along each heading, almost nothing in the paragraphs between them, then the next heading. Nielsen Norman Group named the shape the layer cake, and unlike most of the scanning patterns it is a compliment to the page. It means the reader found a table of contents built into the text and used it, sampling the headings to decide where to stop, which is the behaviour good subheadings are for.

It also sets the standard subheadings are held to. If the layers are the only thing being read, then a heading has to survive being read alone, out of order, with none of the paragraph under it. “Fees and deposits” survives. “A note on the above” does not, and neither does a heading that is really a joke whose punchline is in the first sentence. The same test kills the decorative subheading that repeats the section’s first line, because a layer that adds nothing is a layer the eye stops trusting, and once two are ignored the reader goes back to skimming the left edge.

Take the subheadings out and the pattern collapses. The eye has nothing to sample, so it falls back to the F pattern: a sweep or two across the top, then a run down the left picking up first words. That is the whole argument for a heading hierarchy in body content, and it is why the fix for a wall of text is rarely shorter sentences. It is structure the scan can land on, which also means enough space above each heading that it belongs to what follows rather than floating between two paragraphs.

Eye tracking has named a handful of these shapes, and between them they cover most of what people actually do with text. The F pattern, the Z pattern and the Gutenberg diagram describe the page as a whole; layer cake, spotted, bypassing, lawn mower, marking and commitment describe what happens inside a block of text. Only the last of those is reading.