Information Scent

Also called Scent of Information

Quick answer

Information scent is a person's estimate of how likely a link, label or search result is to lead to what they are looking for, made before they open it. The estimate comes from cues on the screen and from what the person already knows. Information scent is a core idea of information foraging theory by Peter Pirolli and Stuart Card.

~20 sec read

012 min

What Information Scent actually says

Information scent belongs to information foraging theory. The theory describes how people choose between many possible sources of information. It says that a person with a question picks the next source by comparing two things: how much useful information the source will probably give, and how much time and effort it will cost to get it.

The theory writes this as a ratio. The rate of gain is the information value divided by the cost of obtaining that information. A person tries to keep this ratio as high as they can. This is why people do not click every link or scroll every page. Another click usually adds little value and a lot of cost.

A person cannot know the true value of a page before they open it. They only know what the link and its surroundings show. The theory calls these visible signals proximal cues, meaning cues that are near the person. The page at the other end is the distal content, meaning content that is far away until the person clicks. Information scent is the person's imperfect estimate of the distal content's value, built from the proximal cues.

Three properties of this definition matter for design.

  • Scent is an estimate. The estimate can be wrong. A link can promise more than its page delivers, and a page can hold exactly what a person needs behind a link they skipped.
  • Scent is relative to a goal. The same link has high scent for one person and almost none for another. A link titled Food has high scent for someone looking for cheese and low scent for someone looking for a facial cleanser.
  • Scent includes cost. The original definitions name both value and cost. A link that looks relevant but leads to a long, dense page can have lower scent than a plainer link that leads to a short answer.

A link has high scent when a person with a specific goal can tell from the cues alone that the page is probably worth opening.

The person does not calculate any of this. They scan the screen and form a quick feeling that a link looks right or wrong. Herbert Simon described the economics behind this behaviour. He wrote that a wealth of information creates a poverty of attention, so attention has to be allocated among many sources. Information scent is the signal that helps people allocate it.

022 min

Where Information Scent comes from

Peter Pirolli and Stuart Card developed information foraging at Xerox PARC, the Palo Alto Research Center, in the 1990s. Their full statement of the theory is the article "Information Foraging" in Psychological Review in 1999. They borrowed ideas from optimal foraging theory in biology. That theory predicts that animals choose food sources so as to gain the most energy for the least effort.

The vocabulary comes from that source. Animal foragers search for food in patches, and they judge a patch from cues such as smell. Information foragers search for information in patches, such as websites, and they judge a patch from cues such as link text. Nielsen Norman Group's summary pairs the terms directly: an animal's assessment of how likely a patch is to provide food becomes how promising a source of information appears to the user.

Pirolli and Card did not name the underlying observation first. Their 2001 paper says that intermediate information had earlier been called "residue" by Furnas, and that they called it scent "in keeping with foraging terminology".

Two papers at the CHI 2001 conference applied the idea to web use. In one, Card, Pirolli and five co-authors analysed eight recorded web sessions from a laboratory study. They logged every browser action and videotaped each person speaking their thoughts aloud. They reported a strong effect of information scent on the path each person followed. In the other, Ed Chi, Pirolli, Kim Chen and James Pitkow published two algorithms that predict user paths from scent. Chi's team computed scent from the text of links and pages, so a designer could estimate which links a person with a stated goal would follow.

032 min

What makes up Information Scent

Nielsen Norman Group sorts the sources of scent into two groups: what the person sees, and what the person already knows. Designers control the first group directly. They influence the second only over a long period, through the reputation a brand builds.

What the person sees

  • The link label. This is usually the strongest cue. A label should be a short and accurate description of the destination, in words the audience understands. Jargon and branded names lower scent for anyone who does not know them.
  • Content next to the link. A summary line or a thumbnail adds cues. People may not read all of it, but they scan it. A summary that repeats the label wastes the chance to add detail.
  • Context around the link. The other content on the page changes what a label means. The word Christmas signals table settings on a kitchenware site and signals gift baskets on a food-gift site. Position matters too, as the examples below show.

What the person knows

  • Trust in the brand. A person who has had good results from a company will click a vaguer label from that company.
  • Familiarity with the topic. A person who knows the field's vocabulary reads labels that a newcomer cannot decode.
  • Other people's reports. A review that says product B is better than product A raises the scent of product B for the next click. Nielsen Norman Group calls this social foraging.

The two groups combine. Strong prior knowledge can compensate for a weak label, and a precise label can compensate for no prior knowledge. The worst case is a vague label shown to a person who also lacks the background.

042 min

How Information Scent shows up in real interfaces

The clearest evidence comes from named pages that Nielsen Norman Group reviewed. Each shows a different cue going right or wrong.

Labels that a visitor cannot decode

A Cisco.com headline reviewed by Raluca Budiu of Nielsen Norman Group used the branded terms NGFW, AnyConnect and Duo. NGFW stands for Next Generation Firewall, which is Cisco's name for its firewall technology. A visitor who does not already know the products gets no cue about where the link leads. The summary under the headline added nothing to the titles. The label and the summary were both unclear.

Images that do not represent their category

Costco's category Coffee & Sweeteners used a picture that does not look like what most people picture for coffee. The image added no scent for a person looking for that category. Nielsen Norman Group ties this to Eleanor Rosch's work on categories. People take longer to see a chicken as a bird than a robin, because a robin is a more typical bird. A category image should show a typical member of the category.

The same label in two contexts

For a person looking for Christmas table settings, the link Christmas has different scent on two sites. On Williams Sonoma, pictures of dishes and kettles raise it. On Harry & David, the surrounding categories such as bakery and gift baskets lower it. On mobile, less context is visible, so Harry & David changed the label to Christmas gifts. A specific label works even when the context is hidden.

Position as a cue

Position on the page is a cue too. Nielsen Norman Group reports that information presented in the right rail may be read as an ad, because people have learned that ads sit on the right side. A link in that position loses scent even when its text is clear. This is related to banner blindness, the habit of skipping page areas that usually hold ads.

Landing pages that give no early cue

A landing page with a large image and little text gives too little context in the first screen. The PosturePillow page had a bed image, a logo and no clear statement of what the site sells. Nielsen Norman Group describes the likely result: visitors who arrive from a search engine decide the page is not worth exploring and leave. The search result gave a strong cue. The landing page gave no matching cue.

051 min

A second case: the common data set

The second case differs from the first in one variable: the person's prior knowledge. The cues on the page are fixed. What changes is who reads them.

Universities publish a common data set. It holds statistics about each year's incoming students and the admission criteria. Nielsen Norman Group found that it is often hosted in the finance section of the university site, as on the sites of UC Berkeley and Caltech.

A person who has applied to universities before may know this. For that person, a link labelled Office of Vice Chancellor of Finance has high scent. A person who is hearing about the common data set for the first time has two problems. They do not know the name of the document, and they do not know where it lives. Each problem lowers scent, so the information is very hard to find.

The contrast teaches something the first case does not. Fixing a label can raise scent for one audience and leave it low for another. A designer has to decide which audience the label is for, then add a cue for the other, such as a short line that says Admissions statistics are here.

062 min

Applying Information Scent

These moves can be made in a single working week. Each one adds a cue that a person uses to estimate value before they click.

  1. Rewrite vague labels as specific ones.

    Replace More, Learn more and Resources with the name of the destination. A label cannot have high scent when no goal can be matched to it.

  2. Replace jargon and internal product names with the words the audience uses.

    Where the product name must stay, add a plain-language description beside it.

  3. Add a summary line to every link that leads to an article or a category.

    The line should add detail to the label, not repeat it.

  4. Check images against their category.

    Use a typical member of the category, not the most attractive photo.

  5. Make the first screen of each landing page state what the page is for.

    Scent that a search result or link created should still be visible on arrival.

  6. Test labels with a goal.

    Give a participant a task, show only the links, and ask where they would click. This checks scent in the way a person uses it.

Where this stops working

The first limit is that scent works for a goal, not for everyone. Nielsen Norman Group's advice is to optimize for the top task of the page. Users with other goals will not be drawn to a page that is not relevant to them, which is the correct outcome.

The second limit is that raising scent on a link that does not deliver does harm. A strong label on a weak page produces a click, then a return, and then less trust in the next label.

The third limit is that too much scent on every link makes pages crowded. Long summaries under every link lengthen the page and raise the cost of scanning it. Progressive disclosure is one way to balance the two: show a short cue first and the detail on request.

071 min

How to spot weak Information Scent

Weak scent leaves traces that can be checked in analytics, session recordings and support tickets.

  • High exit rate on a landing page that was reached from a link with high scent. The link promised something the page did not show quickly.
  • People clicking a link and returning at once. The back button is the signal that the page did not match the estimate.
  • Heavy use of the site's search box on pages that already have navigation. People who cannot judge the links use search instead.
  • Low click rate on a link in a prominent place. Check whether the label is vague, uses unfamiliar terms, or sits in a position people read as advertising.
  • Support requests asking where a known page or document is. The document exists, but no label led the person to it.
  • Task tests where participants point to the wrong menu item first. Ask each one why. The reason is usually a cue that matched a different goal.

A single metric rarely identifies the cause. Compare the label, the surrounding content, and the audience's likely prior knowledge before changing anything.

081 min

Information Scent vs. nearby concepts

Several concepts deal with how people find and act on things in an interface. They answer different questions.

ConceptQuestion it answersDifference from information scent
DesignInformation architectureIt is the structure. Scent is how well that structure predicts the destination to a person with a goal.Question it answers: How is content organized and named?
DesignSignifiersA signifier shows that an action is possible. Scent estimates whether the result will be worth having.Question it answers: Where and how can the person act?
DesignDiscoverabilityIt is the outcome. Scent is one cause of it.Question it answers: Can people find a feature or page?
DesignHick's LawIt measures time per choice. Scent explains which choice the person picks.Question it answers: How does decision time grow with the number of choices?
DesignBanner blindnessIt is a learned cue that lowers scent for content in those areas.Question it answers: Why do people skip areas that look like ads?

The most common confusion is with signifiers. A button that looks clickable has a clear signifier but may have no scent, if its label says nothing about where it leads. A plain text link with a precise label has high scent even if it has no visual emphasis.

091 min

Limits of Information Scent

Information scent is a model of behaviour, not a measurement a tool can return. Chi and colleagues estimated it from the words on links and pages, and their method needs a stated information need to work. Without a goal there is nothing to compare the cues against.

The model assumes people try to maximize their rate of gain. People do not calculate this exactly. Nielsen Norman Group cites Herbert Simon's bounded rationality: people cannot estimate value and cost precisely, so they use satisficing, which means taking the first option that looks good enough. The model describes the overall tendency, not every decision.

The scent of a link also depends on trust that can be spent. Nielsen Norman Group warns against writing an intriguing title for content that does not match it. In its words: "Yes, you will get your click, but at the same time, you will use up your visitors' trust." The next honest label from the same site then has lower scent than it deserves.

Finally, scent is an estimate from outside the destination. A page with strong content and weak cues will be skipped, and a page with weak content and strong cues will be opened. The remedy for the first is better cues. The remedy for the second is better content, not weaker cues.

102 min

Frequently asked questions

What is information scent in simple terms?

Information scent is a person's guess about whether a link will lead to what they want. The guess comes from the link label, the text and images near it, the rest of the page, and what the person already knows.

What is the difference between information scent and information foraging?

Information foraging is the whole theory of how people choose and read sources. Information scent is one part of it: the estimate of a source's value made before opening it.

Who created the concept?

Peter Pirolli and Stuart Card developed it at Xerox PARC in the 1990s. Their 1999 article "Information Foraging" in Psychological Review is the full statement of the theory.

Can a label have high information scent for everyone?

No. Scent depends on the person's goal. A label such as News is clear for people who want news and irrelevant for people who want something else.

Is a clear label enough to create high scent?

No. The label is the strongest cue, but the image, the summary, the context and the person's prior knowledge also count. A clear label next to a misleading image can still fail.

How do I test information scent?

Give participants a task and show only the links, menu items or search results. Ask where they would click and why. The wrong first choices show which cues failed.

Does information scent apply to mobile apps?

Yes. Less context is visible on a small screen, so the label must carry more of the meaning. Harry & David made this change by lengthening Christmas to Christmas gifts on mobile.

?4 questions

Questions people ask

What is information scent?

Information scent is a person's estimate, made before opening a link, of how likely it is to lead to what they want. It comes from the link label, nearby text and images, page context and prior knowledge.

Who created information scent?

Peter Pirolli and Stuart Card developed it at Xerox PARC in the 1990s, as part of information foraging theory. Their 1999 article in Psychological Review is the full statement.

What is the difference between information scent and a signifier?

A signifier shows that an action is possible and where to do it. Information scent estimates whether the result of the action will be worth having.

How do you improve information scent?

Use specific labels, add a summary that adds detail to the label, choose images typical of the category, and state the page's purpose in the first screen.

§5 sources

Sources

Show all 5 sources
  1. Peter Pirolli and Stuart Card, "Information Foraging", Psychological Review 106(4), 1999. doi.org

  2. Ed H. Chi, Peter Pirolli, Kim Chen and James Pitkow, "Using Information Scent to Model User Information Needs and Actions on the Web", CHI 2001. cs.fit.edu

  3. Stuart K. Card et al., "Information Scent as a Driver of Web Behavior Graphs", CHI 2001. research.cs.vt.edu

  4. Raluca Budiu, "Information Foraging: A Theory of How People Navigate on the Web", Nielsen Norman Group, 2019. nngroup.com

  5. Raluca Budiu, "Information Scent: How Users Decide Where to Go Next", Nielsen Norman Group, 2020. nngroup.com

Keep reading

More from Design

All of Design
All of Design

Missing a concept?

Tell us what to write next. We’ll email you when it’s live.

Several? Separate them with commas.

Only to tell you when it’s live.

How we use your email: privacy.