of the American Society for Information Science and Technology          Vol. 28, No. 5         June / July 2002


Go to
Bulletin Index

bookstore2Go to the ASIST Bookstore



Do Your Links Stink? Techniques for Good Web Information Scent

by Jason Withrow

Jason Withrow can be reached at Diamond Bullet Design, 315 West Huron, #140, Ann Arbor, MI 48103; telephone: 734-665-9353; e-mail: jay@diamondbullet.com. This article is based on the author's recent presentation at the ASIST 2002 IA Summit in Baltimore.

If someone tells you that your website stinks, my advice is to ask them first if it is a good scent or a bad scent and what they thought of the scent quality. You may get an odd look in return, but eventually they will realize you aren't joking. But what scent does a website have? How could a website have a smell? Well, this scent is not a smell that is detected with the nose; it is an information scent, based on our mental associations between concepts.

In an information space like the Web, the goal is to get from one place to another, from here to there. The Web is vast, however, and spatial notions of up and down, right and left do not apply. About the only movements possible on the Web are from general to specific (when browsing through a website hierarchy), from one website to another following a link, or direct access using a search engine. In all these cases, but especially when browsing, visual and verbal cues are required to let us know that we are on the right path to the desired information. These cues are the pieces of information scent that we are following, and we choose the cues with the best likelihood (the strongest scent) of getting us to the desired destination.

An example will help clarify how information scent functions in our use of the Web. Perhaps you are visiting an e-commerce website that carries a wide variety of products. Your interest is in purchasing a movie in DVD format. Looking at the navigation bar for the website, you notice a number of different labels for the links. One link is labeled "Films," which seems more closely related to your desired item than the links labeled "Music" or "Software." In this situation the scent is stronger for the "Films" link, because conceptually that term is closer to the DVD movie than the other two labels. If you were interested in buying a movie soundtrack, though, the decision is not as easy to make. This problem arises because both "Films" and "Music" seem plausible labels. The scent is no longer distinctive. Which path should you choose?

Exploring Scent

Much of the work on information scent has come from researchers in the field of human-computer interaction (HCI), with Xerox's Palo Alto Research Center generating a significant portion of the research in the area. In exploring information scent, these researchers draw on theories of how humans locate information, including information foraging theory. This theory views humans as informavores, continually seeking information from our environment. In a sense we are foraging for information, a process with parallels to how animals forage for food. For both human and animal there are cues in the environment that help us judge whether to continue foraging in the same location or to forage elsewhere. On the Web poor information scent may be one cue that leads a person to leave one website and visit another, because it is unclear where to start looking on the current website and there is also uncertainty whether the effort expended in looking will be justified.

The mechanisms supporting information scent likely draw on the semantic networks that are unique to each individual. We build these vast networks of interconnected concepts from our experiences. The connections between nodes differ not only in terms of distance but also in strength, both of which represent our understanding of the concepts and their interrelationships. Each label or other visual cue on a website activates nodes in our networks that then activate connected nodes in a spreading activation pattern. As the activation spreads it weakens, so that concepts further away from the point of origination receive less stimulation.

Based on these patterns of activation we make our best judgment concerning the link to click when browsing. The process of activating the nodes in the networks and choosing the best option appears to be mainly preconscious, as decisions about where to click next can be made without conscious deliberation if the scent is distinct. Unclear scent, however, or multiple labels with equivalent scent for the desired content will often require conscious deliberation.

The Quality of Scent

How do you know if your website has good information scent? If a user is browsing the website, scent is good if that individual navigates to the desired content simply by choosing the best link at each level of the website. If a search facility is used, the results page should either list the desired page or a page in the same section of the website with sufficient information for the user to make an informed judgment about which link to click. When the users search and don't find the desired content but find something related, the hope is that local navigation on the related pages will provide enough scent for them to browse to the desired page.

There are also a number of behaviors and emotions that indicate scent is poor. If a user is indecisive, moving the mouse between two or three links in the navigation bar and not making a decision, it is very likely that the scent is unclear. The labels probably seem equally good, or perhaps none of the links in the navigation bar seem especially promising, but those two or three are the best options available. This highlights the need for labels to provide clear, distinct scent, suggesting not only what content is found under them but also what content is not found under them.

Additional indicators of poor information scent are feelings of frustration and confusion during browsing. Frustration may manifest in random clicking and frequent use of the browser's "Back" button, until finally the user leaves the website. Confusion arises from labels that convey an unknown scent, generating minimal activation in the semantic networks because related concepts cannot be found. Labels using unfamiliar terminology often give rise to this confusion and the user may remark, "What does that even mean?" Clear labeling is a crucial part of good information scent.

Developing Good Scent

Before discussing approaches and recommendations for developing good information scent, it should be noted that scent is not going to be perfect for every user of your website. Scent cannot be ideal for all users because of the unique composition of their semantic networks. The goal is to choose labels and other cues that convey meaning clearly enough that most users (especially users from target audiences) experience little difficulty browsing or searching.

One practice supportive of good scent is the construction of broader hierarchies when organizing website content. A navigation bar containing 10 top-level links can assign more specific, descriptive labels to those links, compared to the same content being represented with five top-level labels. Those five labels are almost certainly going to use vague terminology, because they need to represent all the content underneath them; the 10 links can use more focused terminology. Generally speaking, labeling ambiguity corresponds to scent ambiguity, so narrowly structured sites may encounter difficulties in this area. A final benefit of greater breadth is that it tends to make the overall structure shallower, creating shorter paths through the website.

Another beneficial practice is to perform an exploratory card sort once the content for the website is determined. In this activity you have users group content pieces into whatever groupings make sense to them and then assign a label to each grouping. These labels are probably not going to be the final ones chosen, but they can be quite revealing about what characteristics the content pieces share. If possible, further subdivision occurs within each grouping, with additional labels assigned to the smaller sub-piles. When performed with multiple users (each user is tested separately), common groupings (and possibly common labels) can emerge, allowing you to derive groupings of content and labels that will accommodate most semantic networks fairly well.

Assessing Scent Quality

Two techniques are helpful in determining the quality of scent at an existing website. The first technique is confirmatory card sorting. The difference between confirmatory and exploratory card sorts are that the exploratory are asking the user to generate groupings and labels, while the confirmatory asks the user to sort the content pieces based on the existing hierarchy and labels. The first step in confirmatory card sorting is to provide the user with the top-level labels and all the content pieces and have him or her sort the content under those labels. Once that is completed, second-level labels are revealed and content is sorted again. Content that does not 'fit' anywhere on the website is indicative of scent issues with the current labels, as well as content that is sorted in a manner different from how the website is presently structured.

The second technique that is quite helpful in assessing scent quality is user testing. Focused tasks that ask the user to locate specific information are best, as they force the user into considering labels at a conscious level. A variety of quantitative and qualitative metrics can be used for judging scent quality. The quantitative measures include path directness (number of clicks required to reach the content), path frequency (which paths are chosen and how often), time to complete, success rate and satisfaction with the labeling. Qualitative data comes from verbal and written user comments, as well as from the signs of indecision, frustration and confusion noted previously.

Supporting Scent

A number of practices are supportive of scent, including the provision of scope information, the use of "see also" links, and shaping browsing around facets. Scope information is exactly what the name suggests; you are providing information about the nature and extent of the content in part of the website. Indicating the scope of a top-level label is an excellent way to provide context for that label. This is useful if two labels seem to have equally good scent; the scope indication clarifies the meaning of each label.

Scope is frequently provided through either a text description of the content in that part of the website or through listing the subnavigation links for that section. If you are interested in researching upcoming concerts at a website but the two most likely links in the navigation bar are labeled "Entertainment" and "Recreation," which one should be clicked? A text description on the home page could clarify that "Entertainment" contains listings of plays and concerts while "Recreation" lists sporting events. If the subnavigation links are labeled "Plays," "Concerts" and "Sporting Events," showing those under the top-level navigation label would also provide necessary context for the two labels.

"See also" links are cross-links between related pages on the website. Typically they are shown in a box on the right-hand side of the page or are displayed in an inline box. These links go by a variety of names: "related pages," "related links" and of course "see also." These links are useful because they help people get back on the right track. During card sorting there are usually some cards that fit in more than one pile, depending on the rationale used for grouping. To minimize confusion it is generally best if content is only located in one place on a website, so in implementing the website provide "see also" links to that location from all the other parts of the site where it could have been located. If a user goes to those other locations first, the "see also" links provide a quick way to return to the correct path.

The final practice that is supportive of scent is the use of facet-based browsing. A facet is an aspect or dimension of an object or piece of information. Each facet is also a potential scent trail to the desired content. If you were looking for a specific book, being able to browse by author, year or publisher (among a variety of possible facets) would improve your chances of locating the book, provided that you knew the book's information for at least one of those facets.

Clearly, there are a number of factors to consider in developing, assessing and supporting scent. Keeping those factors in mind, however, and designing for good scent is likely to result in a number of payoffs, including improved usability and increased user satisfaction. Perhaps a smelly website isn't so bad after all.

How to Order

ASIST Home Page

American Society for Information Science and Technology
8555 16th Street, Suite 850, Silver Spring, Maryland 20910, USA
Tel. 301-495-0900, Fax: 301-495-0810 | E-mail:

Copyright © 2002, American Society for Information Science and Technology