Original: http://www.useit.com/alertbox/20000514.html

Jakob Nielsen's Alertbox, May 14, 2000:

Eyetracking Study of Web Readers

In May 2000, the Poynter Institute released an eyetracking study of how people read news on the Web, mainly focusing on newspaper sites. Their results confirm the findings from my previous studies in 1994 and 1997 of how users read on the Web. This despite the fact that these studies used different methodologies, tested different users and different sites, had different goals, and were conducted at very different stages of the growth of the Web.

As discussed in a sidebar, there are a few methodological weaknesses in the Poynter study that make a few of their minor conclusions suspect, but the main findings are very robust and credible. When different people keep finding the same results year by year, it is time to take the findings seriously and to base Web design on the data and not on wishful thinking.

Web content is intellectually bankrupt and almost never designed to comply with the way users behave online. Almost all websites contain content that would have worked just as well in print. Even online-only webzines are filled with linear articles with traditional blocks-of-text layouts. No hyperlinks, no scannability. New forms of content that are optimized for online are exceedingly rare, and I keep returning to the same four examples when I am asked to name good writing for the Web: Tomalak's Realm, AnchorDesk, the Feed Daily mini-column, and Yahoo Full Coverage.

Main Findings

Text Attracts Attention Before Graphics

It was almost twice as common for users to fixate on the text as on the images upon their initial visit to a page. In general, users were first drawn to headlines, article summaries, and captions. They often did not look at the images at all until the second or third visit to a page.

Keep Headlines Simple and Direct

Confirming our findings from 1997, the users in the current study also preferred straightforward headlines to funny or cute ones. A new finding was that users often praised the Web headlines for being better than the headlines in print newspapers. It seems that several of the news sites have taken the earlier findings to heart and have started rewriting their headlines for online.

Shallow Reading Combined With Selected Depth

It was more than three times as common for users to limit their reading to a brief as opposed to reading a full article. Even when reading a "full" article, users only read about 75% of the text.

In other words, the most common behavior is to hunt for information and be ruthless in ignoring details. But once the prey has been caught, users will sometimes dive in more deeply. Thus, Web content needs to support both aspects of information access: foraging and consumption. Text needs to be scannable, but it also needs to provide the answers users seek.

Interlaced Browsing

Users in the Poynter study frequently alternated between multiple sites:
  • they would read something in one window
  • then switch to another window and visit another site
  • and then return to the first window and read some more on the first site; possibly to turn to the second window again later in the session

I observed this behavior as early as 1994: users would interlace browsing sessions in several windows. Doing so is particularly easy on big monitors that show several full-page windows simultaneously, but can also be done on small screens. The Windows task bar facilitates session interlaced browsing as long as users stay below eight sessions or so.

I admit that I was surprised when we started seeing interlaced browsing in 1994. Previous studies had not identified this behavior, so I originally expected people to browse a specific site and stay with its navigation features until they decided that they were done with it. In retrospect it is clear why interlacing was not seen in the old days: we were simply not studying sufficiently rich hyperspaces.

The lesson for site designers is that users are not focused on any single site. There is not even such a thing as "a visit" to a site: even while the user is "visiting" your site, he or she is also checking out the competition. Truly, the Web as a whole forms the user experience.

Site design must accommodate people who leave and return frequently:

  • help users reorient themselves
    • plain and simple headlines immediately tell users what each page is about
    • simple page titles that start with a salient keyword help users pick out pages from the minimized tiles in the Windows task bar
  • do not assume users can remember their entire browsing session:
    • provide breadcrumbs and other location tools
    • do not change the standard link colors - doing so makes it harder to recognize what pages the user has already seen
  • use standard terminology to minimize the need for users to switch context and remember what you call things
  • during user testing, interrupt the users for a few minutes if they don't leave your site on their own (in order to test their ability to return to the site)

Implications for Non-Newspaper Sites

It is not really a weakness of the Poynter study that it focused on newspaper sites. It is fair enough to study a narrow genre. But since the vast majority of websites are not newspaper sites, I want to mention some limitations in transferring the findings to other types of sites such as corporate sites, ecommerce sites, or intranets:
  • Trust is less of an issue for newspapers which usually have high integrity. Other sites need to fight for credibility and must reduce marketese, slogans, and other elements that generate distrust.
  • Users are likely to spend much less time on other sites. Ten minutes would be a long visit to most sites.
  • Users will read fewer words on other sites than they do on newspaper sites. Editorial integrity and journalistic objectivity makes people more willing to read a larger percentage of the material. The task of reading news implies a willingness to process more words more than the average Web task which is directed at finding specific information and solutions.
The new eyetracking study is mostly applicable to all types of websites. Most of the Poynter findings confirm earlier findings from studies of many other types of sites, so they relate to basic characteristics of reading on the Web and are not limited to newspapers.

Adding one more study to the list of evidence for different reading behavior will hopefully convince more Internet executives of the need to write differently for the Web and hire specialized Web editors who understand online content.

Read All About It


Previous: April 30, 2000: Progress in Internet client design
Next: May 28, 2000: Alertbox Five Years Retrospective

See Also: Complete list of other Alertbox columns