Twitter  "Algorithms have consequences." Zeynep Tufekci on Ferguson and net neutrality:  
Nieman Journalism Lab
Pushing to the future of journalism — A project of the Nieman Foundation at Harvard

Word clouds considered harmful

The New York Times senior software architect would like the newest “mullets of the Internet” to go back from whence they came.

In his 2003 novel Pattern Recognition, William Gibson created a character named Cayce Pollard with an unusual psychosomatic affliction: She was allergic to brands. Even the logos on clothing were enough to make her skin crawl, but her worst reactions were triggered by the Michelin Tire mascot, Bibendum.

Although it’s mildly satirical, I can relate to this condition, since I have a similar visceral reaction to word clouds, especially those produced as data visualization for stories.

If you are fortunate enough to have no idea what a word cloud is, here is some background. A word cloud represents word usage in a document by resizing individual words in said document proportionally to how frequently they are used, and then jumbling them into some vaguely artistic arrangement. This technique first originated online in the 1990s as tag clouds (famously described as “the mullets of the Internet“), which were used to display the popularity of keywords in bookmarks.

More recently, a site named Wordle has made it radically simpler to generate such word clouds, ensuring their accelerated use as filler visualization, much to my personal pain.

So what’s so wrong with word clouds, anyway? To understand that, it helps to understand the principles we strive for in data journalism. At The New York Times, we strongly believe that visualization is reporting, with many of the same elements that would make a traditional story effective: a narrative that pares away extraneous information to find a story in the data; context to help the reader understand the basics of the subject; interviewing the data to find its flaws and be sure of our conclusions. Prettiness is a bonus; if it obliterates the ability to read the story of the visualization, it’s not worth adding some wild new visualization style or strange interface.

Of course, word clouds throw all these principles out the window. Here’s an example to illustrate. About six months ago, I had the privilege of giving a talk about how we visualized civilian deaths in the WikiLeaks War Logs at a meeting of the New York City Hacks/Hackers. I wanted my talk to be more than “look what I did!” but also to touch on some key principles of good data journalism. What better way to illustrate these principles than with a foil, a Goofus to my Gallant?

And I found one: the word cloud. Please compare these two visualizations — derived from the same data set — and the differences should be apparent:

I’m sorry to harp on Fast Company in particular here, since I’ve seen this pattern across many news organizations: reporters sidestepping their limited knowledge of the subject material by peering for patterns in a word cloud — like reading tea leaves at the bottom of a cup. What you’re left with is a shoddy visualization that fails all the principles I hold dear.

Every time I see a word cloud presented as insight, I die a little inside.

For starters, word clouds support only the crudest sorts of textual analysis, much like figuring out a protein by getting a count only of its amino acids. This can be wildly misleading; I created a word cloud of Tea Party feelings about Obama, and the two largest words were implausibly “like” and “policy,” mainly because the importuned word “don’t” was automatically excluded. (Fair enough: Such stopwords would otherwise dominate the word clouds.) A phrase or thematic analysis would reach more accurate conclusions. When looking at the word cloud of the War Logs, does the equal sizing of the words “car” and “blast” indicate a large number of reports about car bombs or just many reports about cars or explosions? How do I compare the relative frequency of lesser-used words? Also, doesn’t focusing on the occurrence of specific words instead of concepts or themes miss the fact that different reports about truck bombs might be use the words “truck,” “vehicle,” or even “bongo” (since the Kia Bongo is very popular in Iraq)?

Of course, the biggest problem with word clouds is that they are often applied to situations where textual analysis is not appropriate. One could argue that word clouds make sense when the point is to specifically analyze word usage (though I’d still suggest alternatives), but it’s ludicrous to make sense of a complex topic like the Iraq War by looking only at the words used to describe the events. Don’t confuse signifiers with what they signify.

And what about the readers? Word clouds leave them to figure out the context of the data by themselves. How is the reader to know from this word cloud that LN is a “Local National” or COP is “Combat Outpost” (and not a police officer)? Most interesting data requires some form of translation or explanation to bring the reader quickly up to speed, word clouds provide nothing in that regard.

Visualization is reporting, with many of the same elements that would make a traditional story effective.

Furthermore, where is the narrative? For our visualization, we chose to focus on one narrative out of the many within the Iraq War Logs, and we displayed the data to make that clear. Word clouds, on the other hand, require the reader to squint at them like stereograms until a narrative pops into place. In this case, you can figure out that the Iraq occupation involved a lot of IEDs and explosions. Which is likely news to nobody.

As an example of how this might lead the reader astray, we initially thought we saw surprising and dramatic rise in sectarian violence after the Surge, because of the word “sect” was appearing in many more reports. We soon figured out that what we were seeing had less to do with violence levels and more to do with bureaucracy: the adoption of new Army requirements requiring the reporting of the sect of detainees. Of course, the horrific violence we visualized in Baghdad was sectarian, but this was not something indicated in the text of the reports at the time. If we had visualized the violence in Baghdad as a series of word clouds for each year, we might have thought that the violence was not sectarian at all.

In conclusion: Every time I see a word cloud presented as insight, I die a little inside. Hopefully, by now, you can understand why. But if you are still sadistically inclined enough to make a word cloud of this piece, don’t worry. I’ve got you covered.

Jacob Harris is a senior software architect at The New York Times.

What to read next
BuzzFeed HQ
Caroline O'Donovan    Aug. 11, 2014
With a new round of investment from Andreessen Horowitz, the viral giant aims to get even bigger. One step down that path: making it more clear to readers what kinds of content BuzzFeed really offers.
  • Chuck Taylor

    The “Deadly Day in Baghdad” link is broken.

  • Megan Garber

    Thank you! Fixed.

  • Michael Becker

    It would be great if all news organizations had the programming talent that the New York Times has, but we don’t. Sometimes, shortcuts and pre-fab services need to be used.

    Now, I’m not defending word clouds, but if we believe data-driven interactives are going to play a big role in the future of news, then we need to be willing to accept and pardon a few “mullets” now and again, at least until the majority of us learn how to craft narrative out of code.

  • Seth Grimes

    I made pretty much the same points in a 2009 article:


  • Michelle Walkden

    Love the post and completely agree. Only problem is I now know about Wordle! How can I possibly resist using it?

  • Tom Scheinfeldt

    I share your frustration with word clouds, but the problem is neither inherent in the technique nor unique to it. The problem is in their facile use. The same can be said of geographic visualizations, so many of which are offered without context or qualification. Pins on a map don’t say much more than word clouds and can be just as misleading if they aren’t explained and substantiated by other evidence. I wouldn’t say your example from the Times falls into this camp, but lots do. 

    And don’t get me started on timelines…

  • Anonymous

    It’s best to differentiate “tag clouds” from “word clouds,” the latter you might also call “prose clouds.” Tags are keyword descriptors that are well-chosen specifically because they capture some essential meaning of the referenced content in a single word, much like a category name. Words, in the general sense, are used for prose and their direct relationship to the meaning of the content is far, far more removed. The utility of a “tag cloud” vs a “word cloud”  is thus very different because tags are a better representation of the content by design. In the case of aggregating tags for bookmarks on social sites like Delicious or Pinboard, tag clouds are really quite useful. Also please note that the Zeldman post you reference regarding “mullets of the internet” actually argues that tag clouds are smart and useful, but just that they are overused and annoyingly trendy. I will agree with you though that the clouds shown in this post are pretty useless and miss the point.

  • Dan Nguyen

    While timelines and geolocation maps are often poorly used, when done well, they provide a storytelling element that no other visualization can. However, there is nothing informationally useful that a word cloud conveys, even if it is well-designed (for a word cloud). 

    In fact, its very physical design obscures even the scant modicum of data that it purports to tell, as it is very difficult to visually determine which words were more frequently used, as font-color and word-length may appear more prominent than size (which I’m assuming is the visual indicator of frequency in most word clouds).

    So we have a visualization that requires little reporting, shows very little (useful or not) data, and is visually difficult to make quantifiable sense of. And oftentimes, it’s just plain ugly. So why use them at all?

  • Tom Scheinfeldt

    A couple points, in no particular order: 1) I disagree that maps and timelines tell stories — they can illustrate stories or help inspire and inform stories, but they can’t tell them. 2) Wordles are poor visualizations of (also poor) word frequency data, but that doesn’t mean word frequency is always a useless analytic tool. 3) I agree that geo-temporal visualizations, when well done, are richer than word clouds or other word frequency visualizations. However, when done poorly (i.e. most of the time), they’re just as bad.

  • Anonymous

    The link to the commencement speech example is broken also. Here is the correct link:

  • Anonymous

    Pretty much what I was going to say.

  • Brenda Bernstein

    Word Clouds can be quite useful to job seekers who want to know what keywords to emphasize in their resumes.  As a resume writing professional, I love word clouds because they help identify what is important to a company, whether in a job description or on a website.  I also put my company’s core values into a word cloud on my website…  there is only one instance of each term, so the usefulness of the cloud is simply a graphically appealing representation of my values.  So I’m a fan!

  • l e

    blogs critizing word clouds now _that’s_ old hat

  • franksting

    Have you just discovered it, or is “More recently, a site named Wordle” supposed to be relative to your earlier use of 1997?
    I’ve been using Wordle for about three years – an aeon on the Internet
    Let’s not take Wordle TOO SERIOUSLY. I use it for art, for adornment. I use words, the words aren’t often beautiful. It helps make them so.Context is everything, if anyone reads my Wordle’s and thinks it represents anything but an idea of what I write, then they probably aren’t going to read what I write for very long.And that’s fine.

  • Anonymous

    I don’t have a problem with people doing word clouds for fun or because they find them pretty. I just have a beef with them being presented as a visualization that offers insight. But word cloud away all you want. Just know its limitations.

  • Anonymous

    Good point that was a little lost. Tag clouds were more useful because the vocabulary was usually a lot more sparse and the arrangement was designed so you would take action on your links as opposed to swimming in some artful arrangement hoping to emerge with wisdom.

  • Anonymous

    Tom, I sympathize with your points, but I think maps and timelines are at least familiar interfaces and echo the physical and temporal qualities of the story you are trying to reveal. But word clouds are divorced from reality in a sense that maps and events are not.

    I also agree that textual analysis makes sense when the text IS the story (ie, analyzing speeches) and not just primary documents of the story. But even then, I feel like there are many more sophisticated tools.

  • Anonymous

    Indeed, we’ve all been there. Even the Times. So I don’t want to come off as a judgmental sourpuss, but at the same time, I think we have to remember we’re trying to tell a story in our data, which we all can forget in the face of Filler Shiny. That’s all. Thanks.


    wow…. really cool. what’s the difference with mid mapping? is it same concept?


    wow…. really cool. what’s the difference with mid mapping? is it same concept?

  • Nancy Day

    Interesting journalistic point of view

  • Reid

    I usually don’t find word clouds to be very helpful, but I did see a good use of it recently: at the beginning of a presentation, there was a slide that was a word cloud of the presentation itself. It was only shown for a few seconds, but it was a nice, very quick view of what the presentation was going to be about.

    I think you’re right that word clouds can’t give narrative or even make sense of something, but I do think they still have use as a very high level overview.

  • Jonathan O’Donnell

    It’s a very interesting article, however I think your frustration is the use of anything without context. That’s the key. Using a Wordle without context is Lazy (at best)  and a possible misrepresentation of the information allowing anyone to conjure up their own ideas and possibly missing the point.

    To be fair, Wordle is, well, for words and your Baghdad graphic is a completely different mechanism for delivering information.  That is not to say Wordle cannot be used effectively (no, I have no affiliation with Wordle :)). But I have seen and personally used Wordle (with comparisons and context) along with a narrative to drive a point home.

  • Janelle Urchenko

    Automatically generated word clouds can be the bane of any designer, as well, since the battle still rages to dispell the layperson’s misconception that good design is just about making your message pretty, or attention-grabbing. 
    Agree with Jonathan that it’s all about context in
    communications/advertising/education/what have you. No matter your
    angle, you’ve got to understand your message first.

    That said, I think representations like word clouds, when used thoughtfully, are extremely effective at grabbing the attention of those who would never read an article in the first place. Dangerous if the message isn’t well-crafted, but effective nonetheless.

  • David Gadarian

    Fun piece – congratulations as it made it to my Top Headlines in LinkedIn.  Saw the headline and new I had to read this one.

    In terms of pure journalism, your point as you constructed it is very valid, but I’m not sure you give word clouds a fair shake (Shame on You?).  I can’t say I use them a ton on my site, but I do have a post on how to make them that is probably the most popular post I have written to date.  

    I think their popularity stems from the fact that they can be visually compelling and certainly can add considerably to a narrative.  I also believe that for the creator of a new word cloud, when using a service like Wordle there is an amazing sense of discovery that frankly was not available to the average user not too long ago – most people don’t have the staff and resources of say a NYT or a Fast Company and being able to step that much closer to something “creative” can be extremely gratifying.  

    Lastly +1 for William Gibson.  Have a great weekend. 

  • Karl Roche

    So when I analyse a speech someone makes, I’ll now make a map of Baghdad instead – thanks for the pointer.

  • Dan Nguyen

    > I disagree that maps and timelines tell stories

    I highly, highly disagree with this statement. Great maps and timelines require great journalism, research, and analysis. When given that attention, their visual output can be even more informative than any narrative text.
    The perfect example is covered by Edward Tufte in one of his books (maybe the second one?), in which he examine’s the work of Dr. John Snow in investigating the cause of a cholera outbreak. After interviewing residents, he created a map showing the location and number of outbreaks:

    His work convinced authorities to disable the water pumps near the outbreak areas and is largely considered to be the start of epidemiology.

    The map, of course, illustrates the story. But how does that not count as *telling* the story? There’s no narrative that can fulfill the informational value and empirical argument that this map makes. Maps often accompany text stories, here, the text is just an accompaniment to the map. The reason why the latter case happens far less often than the former is because we are less accustomed to building visualizations. Hence, the need to fallback on a useless tool like Wordle.

    Another great example spotlighted by Tufte is Charles Minyard and his flow map of Napoleon’s failed Russian campaign. It serves as both a map and a timeline and tells a far clearer story than any amount of text would.

  • l e

    the author doesn’t link to the article the fastcompany wordcloud is attached to

    do they really claim this to be cutting edge data journalist or a quick simple peek…

    you obviously spend far longer on your work the two are just not comparable

    again oldhat, this article is nearly as much linkbait as any old worldcloud post.

    should have started with the meaty stuff

  • Alexander Hotz

    couldn’t agree more

  • Dan Nguyen

    Yes, I echo what harrisj says about your second point. No one is saying that word frequency analysis is not a useful tool. I’m saying, and I think Jake is too, that word clouds fail at communicating even the most basic datapoints of that frequency analysis. Even a simple bar chart would show the data more accurately, and also allow for the plotting of other datapoints relevant to language use (such as rate and timeframe of the usage, common surrounding words and/or sentiment analysis, etc).

    A word cloud is a popular option because it is easy to make. If being easy-to-produce is a primary concern (and that is not in itself a bad thing), then word clouds should be seriously considered. But let’s not fool ourselves into thinking that we’re doing it because it’s an actual informational service.

  • Tom Scheinfeldt

    I am familiar with both of your examples, and I agree they’re revealing … in the right hands. Snow’s map spoke to him because he already knew something about the geography, demography, and water supply of London, the symptoms of Cholera, and the progress of the epidemic. If he hadn’t known those things, the map would just have been so many dots on a grid. Same for Minard. Give Mindard’s map to an average undergraduate who knows nothing about the geography of Europe, the way pre-industrial armies traveled and were supplied, or even who Napoleon was, and then ask him to explain the significance of the Napoleonic Wars. He’ll be able to tell you that lots of people died in the cold on their way to Russia. He won’t be able to tell you why that matters. These graphics are excellent at conveying information, but they do little to convey the significance of that information. To me, “the story” is in the significance, not in the facts.

  • Tom Scheinfeldt

    Jacob, This is exactly what I was (ham-handedly) getting at: “textual analysis makes sense when the text is the story.” And, yes, word clouds are lousy tools for doing that textual analysis. Thanks for clarifying this for me, and thanks again for the post.

  • Tom Scheinfeldt

    I think we agree here. I’m certainly not arguing for Wordles. Just not to throw the baby (textual analysis) out with the bath water (Wordle).

  • Toon Van de Putte

    You’re quite right. Anyone who tries to work with word/tag clouds in an intelligent manner has noticed that they lose all meaning once coherent statements (sentences) are pulled apart. has going for it that you can also upload a tab-separated set of statement/frequency pairs. It then generates an image where each statement (word or sentence) is sized according to the number that went with it. This is a really quick and fairly OK way to create a basic visualization of this kind of data.
    In my experience, a properly done word cloud (ie a tag cloud, actually) is never really a data vizualisation, but more of an illustration. Like a photograph, it grabs the reader’s attention and pulls them into the text. The point, then, is to create a raw starting point with a service like and turn it into an illustration piece to accompany the story.
    In my opinion, there is a huge difference between full-blown data visualization that tells a story all on its own (a story that couldn’t be told with text alone, usually), and supporting illustrations that happen to use data points. Keeping the distinction and the emphasis (aesthetics vs. story) right is key to success.

  • Anonymous

    No, Fastcompany doesn’t really do war reporting obviously, but the question is why should we produce these kind of low-analysis pieces with filler visualizations in the first place?

    I do object to the linkbait, since I forgot to do a slideshow of top 10 word cloud mistakes. Plus, this is kinda the type of guest essays that Nieman does.

  • Anonymous

    Thanks for the comment. The Shame On You joke was targeted at people who wanted to do a word cloud of this piece to tease me, so I’m sorry that came off the wrong way.

    I have seen references to the vast resources of the NYT, so I’m going to shed a little light on the process here. I had access to the Wikileaks data and used SQL and Ruby to dump a crude version of the deaths in Baghdad over the years just by writing out KML to Google Earth. Graphics helped to make it prettier using ArcGIS, and the reporter helped weed out some duplicates and provide context. And that was it. More than a word cloud, but not weeks of work.

    I guess I feel like there never will be a silver bullet tool that will do it all for you, but a little programming knowledge goes a long way to making this work much simpler.

  • Macsym

    Also worth-reading for design

  • Trevor Owens

    I agree on most points bit would suggest that the value of quick and dirty visualizations (like wordle) should come in the way they can serve as steps in playing with and exploring data. In general, the idea that a visualization is the end product misses out on many ouppertunites to mess around with and explore data as part of a sense making process.

  • Npattison

    And from the crusty old editing side … let’s send editors who think dog-earred and imprecise phrases are suitable for headlines “back from whence they came.”

  • David Gadarian

    No worries Jacob – I thought your final word cloud link was funny.  I think you are OK there.

    Also thanks for sharing a little “inside baseball” on the NYT – interesting stuff.  

    Have a great weekend.  

  • Kenneth Gregory Prud’homme

    I totally agree that this cloud phenomenon erodes at the very Principle of accurate and insightful reporting..i.e. journalism…and will hopefully go the way of the Dinosaures…Kenneth Gregory Prud’homme

  • Tagxedo

    I’m the creator of Tagxedo, which makes amazing word clouds with shapes. A reader at my blog asked me to respond to your criticism, and that’s why I’m here, and let me start with the short reply I gave him (see for the context of the conversation):

    I’ve read that piece. I sympathize with the author’s disgust of seeing word cloud as a kind of “pseudo insight” in lieu of actual thoughtful journalistic analysis. Just as hammer is a poor replacement for screwdriver, or vice versa. On the other hand, I hope this blog demonstrates that word cloud could be captivating, artistic, original, and in some cases, even appropriate. And to see that you don’t need to look further than my tribute to the late Dennis Ritchie. Don’t you think that the word cloud serve as a much better tribute than any other textual analysis of his great book? Granted, Tagxedo, unlike other word cloud tools, emphasizes a lot more on its artistic form.

    Word cloud as a generic replacement for thoughtful journalistic analysis is bad, but so are most other visualization techniques. I don’t think we have a disagreement here. However, your title and conclusion, a complete dismissal of word cloud – are something I object to. Word clouds are much more than textual analysis.

    It could be art.

    It could be a quick way to capture emotion (e.g. “3 words about Obama”)

    It could be a tribute.

    – Hardy Leung (creator of Tagxedo)

  • The Wildfire Effect

    Thanks for your perspective about word clouds.

    As with all forms of communication the context must be understood and simple tools like word clouds also have their place.  As a busy professional processing over 100 daily unique pieces of information in the forms of email, readers, social networks, magazines and person to person conversation’s I love visual aggregators   

    It places the control of time back in my hands with a snapshot that I can then use to determine if I want to devote my thoughts to learning more.  I appreciate infographics, photographs, curated lists and suggestions from friends and colleges  for the same reasons.  

    All good things to you, and your readers
    Stephanie Michelle Scott

  • Anonymous

    Well, if you want to do a tag cloud as art or engagement, bully for you. I don’t have a problem with that. I just don’t think they work for analysis or substantive insight.

  • Mac McCarthy


    1. Word clouds are not generated for the benefit of the readers; no ordinary reader pays any attention to word clouds. They are uninterpretable, lead nowhere, have no evident meaning, and don’t link to anything.
    2. Word clouds are purely decorative; publishers think they add something visual to articles that otherwise don’t have much art, or are in too much of a hurry to come up with art.
    3. Word clouds can be produced programmatically, which is their main appeal to publishers – automatic, costless, decorative. The same thing happened when they discovered that service that automatically links any word in an article–any word at all–to a ‘definition’ and to an advertisement tied to that word. Don’t you hate that? I do. This is the same thing, just without the excuse of generating cash.

    4. The real proof that 1. and 2. are true is the fact that the words are arranged in that supposedly artful way, which makes the word cloud even harder to read. And of course the arrangement of the words is meaningless; only the size correlates to anything — nothing useful, though, as you point out.

    5. The only way word clouds could be useful is if each word links to a list of other articles that had the same keywords in similar density. Even that is stretching the concept of usefulness to its breaking point.

    6. “Mullets of the Interne” ha ha ha ha ha!!!

  • I808

    So, how many of us ran this article through Wordle?

  • Kacharagadlatechno

    Nice review

  • Erik Devlies

    Haha, the word cloud at the end :-)
    But you do have a point. Never looked at it that way.
    I still believe though for some education purpose it can serve well.
    Even if it only was to proove to the pupils what you said .

  • Jenni

    I think the problem is that word clouds (tag clouds) have been mis-translated from the web into journalism.

    Tag clouds were (are) predominantly used for site visitors to grasp the main topics of a website at a glance, and potentially click through to the most interesting topics. They are also good for search engine optimisation. Using them to try and assess the most popular words in a single article or presentation is beyond pointless, as is literally trying to figure out exactly which words of the same size text are most dominant.

  • Christen Bouffard

    I’m glad I’m not the only one who dies a little at the sight of a Wordle. It isn’t just the shallowness of explanation and lack of deeper meaning, but also the lack of control over what is created by Wordles that I take issue with. I would prefer to employ a tool such as IBM’s ManyEyes ( to create such visualizations. With these types of tools creators have more control over the final product AND it can be edited and revised.

    The proliferation of visualization tools that make it simple for even the least technically inclined of us is fantastic, but there is a new kind of literacy required to understand which type of visualization is most appropriate for a situation.