Wikipedia & Knol: Why Knol Already Failed

Why Knol is worse than Wikipedia, has failed, and will continue to fail.
Wikipedia, predictions, statistics, survival-analysis, Google
2009-01-212013-05-04 finished certainty: highly likely importance: 6


Google’s Wikipedi­a-killer, , has passed some an­niver­saries; in Jan­u­ary 2009, it reached its 6-month an­niver­sary. 6 months is a fairly good time-pe­riod to look back­—­long enough that the hype has died down, and for big things to have been done (8500 hours is a long time on­line).

How had it done by then? Faster than Cit­i­zendium, even, it was dead. As a , one might ques­tion my ob­jec­tiv­i­ty. So we’ll see what a rel­a­tively trusted third-par­ty, , says:

What hap­pened to Knol? An­nounced by Google in late 2007 and launched in July 2008, the site was meant to bring more cred­i­ble (read: not writ­ten by anony­mous Wikipedi­ans) “knowl­edge units” to the web, and it would al­low the au­thors to cash in on their work. But it’s 2009, and Knol ap­pears to be no­table largely for its non-no­ta­bil­i­ty…Now the bad news: no one’s read­ing the site, and it’s awash in poor con­tent.

As for the qual­ity of the con­tent, Google’s at­tempt at mon­e­tiz­ing (both for it­self and for its au­thors) the Knol en­tries has had a per­verse effect. While it has at­tracted plenty of de­tailed com­men­tary from learned pro­fes­sion­als, it’s drown­ing in plenty more that is ba­si­cally spam, pla­gia­rism, or a stub, thrown up in the ap­par­ent hope of mak­ing some quick cash. (Though be­cause of point num­ber one, that’s not hap­pen­ing, ei­ther.)

Take “Barack Obama”, for in­stance. A search for his name brings up 809 en­tries; since most Knol users ap­pear to write their own en­tries rather than add to oth­ers (for which no com­pen­sa­tion is forth­com­ing), the pro­lif­er­a­tion of en­tries is in­evitable. And it’s not at all clear that the best ones are ris­ing to the top.

It’s larger than Wikipedia was?

At six months, Wikipedia only had a few hun­dred con­trib­u­tors and ~6,000 ar­ti­cles; but Knol had ~100,000. Surely this demon­strates that Knol has met with some sort of suc­cess?

Well, Knol ‘knols’ do not cor­re­spond to Wikipedia ar­ti­cles. Knols re­peat them­selves. Worse, they’re often spam, copy­vios, repet­i­tive. The true count of knols that are orig­i­nal to Knol and unique is prob­a­bly very low—quite pos­si­bly down to ~6,000.

Growth is important

But get­ting bogged down in ar­ti­cle count omits a ma­jor met­ric, and per­haps the most im­por­tant met­ric: rate of growth. Wikipedia at 6 months could claim 6,000 ar­ti­cles, but what’s im­por­tant is that just 5 months (a lit­tle less than twice as) lat­er, the ar­ti­cle count had more than dou­bled to >13,000. This was the knee of the curve that would lead to Wikipedi­a’s >3 mil­lion ar­ti­cles (for up­-to-date sta­tis­tics, see Spe­cial:S­ta­tis­tics).

An equal comparison is not equal

An­other pro­found point is that if Knol is only equal to Wikipedia, then it is a worse model than Wikipedia.

We should ex­pect more of Knol than of Wikipedia at sim­i­lar stages! Knol has, by virtue of its po­si­tion in time, nu­mer­ous ad­van­tages over /Wikipedia. We should ex­pect a lot more.

It has:

  1. A clear li­cense regime. Thanks to 8 years of , the choice is not lim­ited to just the prob­lem­atic .

  2. 8 years of hard­ware ad­vances; ~5 it­er­a­tions of Moore’s law.

  3. 8 years of wiki de­vel­op­ment, demon­strat­ing dead ends, the good ideas, & what re­mains to be im­proved. Imag­ine if Knol had to start with the state of the art in 2001. It would be truly grue­some. (Any­one looked at the very old Wikipedias in Nos­tal­gia, or old wikis like ? They’re hideous and un­us­able! They make me quite grate­ful for 2009 Me­di­aWiki with all its mod­ern con­ve­niences.)

  4. The back­ing of a com­mer­cial jug­ger­naut, with its many ben­e­fits:

    1. world-class host­ing & tech­ni­cal sup­port
    2. which they did not have to de­velop
    3. long-term back­ing (Bomis turned out to sup­port Wikipedia for a good 5 years, but there was no guar­an­tee)
    4. world-class soft­ware de­vel­op­ment re­sources1
    5. Mas­sive pub­lic­i­ty. At launch Knol had in­fi­nitely more pub­lic­ity than did Wikipedia; much of this buzz was due sim­ply to who the backer was.
  5. A pub­lic ed­u­cated to read wik­is, and to use them.

    This is a greater ad­van­tage than it seems. How many peo­ple could Wikipedia hope to draw on at day 1—that cared even a lit­tle about , that knew what a wiki was, that would­n’t dis­miss it as hope­less, and had an edit­ing fa­mil­iar­ity with wik­is? Darned few. We had to con­stantly evan­ge­lize and ed­u­cate peo­ple about wik­is, and by dint of un­remit­ting effort cre­ate the Eng­lish Wikipedia and make it in­ter­est­ing and valu­able enough that peo­ple would con­tribute who did­n’t ful­fill any of those cri­te­ria. En was the ex­is­tence proof that large-s­cale wikis were pos­si­ble and valu­able. Knol, on the other hand, can draw im­me­di­ately on that pool of peo­ple Wikipedia cre­at­ed.

  6. A model tar­geted di­rectly at peo­ple un­happy with Wikipedia. Are you an ex­pert tired of ‘an­ti-ex­per­tism’ on Wikipedia? Why try to get along with those bump­kins when you could have your own ar­ti­cle com­pletely to your­self on Google Knol (and get paid for’t)? Wikipedia ap­pealed to those un­happy with Nu­pe­dia. Nu­pe­dia when Wikipedia launched was a lot smaller than Wikipedia was when Knol launched. I think this pool of pos­si­ble con­trib­u­tors was thus also much larger for Knol than it was for Wikipedia.

Why Knol was doomed to fail

With Wikipedia by this point, the ba­sic con­cept of col­lab­o­ra­tion had been proved. With Knol, we see only the di­vi­sive­ness of the pay­ments sys­tem, and a few iso­lated au­thors striv­ing on their own.

The im­por­tant thing here is that peo­ple work on differ­ent things. No ar­ti­cle writer on Wikipedia is dis­cour­aged by the fact that, say, Tim Star­ling is be­ing paid for his sys­tem ad­min­is­tra­tion and PHP hack­ing—they don’t want to do it and are happy to ex­change do­na­tions for the prac­ti­cal re­sults of Tim’s la­bor viz. high and faster servers. But if, say, every FA au­thor got 1000 dol­lars, then per­haps writ­ers of or­di­nary ar­ti­cles will get dis­cour­aged, or the bit­ter bat­tles over what will be an FA or not will burn out the best ed­i­tors.

In­tu­ition sug­gests, and some re­sults sup­port, the hy­poth­e­sis that pay­ing only some con­trib­u­tors can “” other un­paid con­tri­bu­tions (through shift­ing the rea­son for con­tri­bu­tion from in­trin­sic to ex­trin­sic mo­ti­va­tion). See for ex­am­ple:

There may be no happy medium be­tween pay­ing only for things which won’t get done oth­er­wise (te­dious sys­tem ad­min­is­tra­tion), and pay­ing for most every­thing; I think Knol is in that un­happy mid­dle.

Knol did fail

How did Knol fare since I di­ag­nosed its high lev­els of fail­ure and pre­dicted gen­eral doom in Jan­u­ary 2009? Google con­tin­ued to up­date the site, but then stopped any up­dates in De­cem­ber 2009, and the site stag­nated over the 94 weeks since. In­ter­mit­tent tech­ni­cal prob­lems have lead to the ac­cu­sa­tion that Knol has been aban­doned—pre­sum­ably for hot­ter ini­tia­tives like , which was pub­licly an­nounced, re­leased, and aban­doned dur­ing the in­ter­val that Knol has stag­nat­ed; or , which looks like it will have more stay­ing power than its pre­de­ces­sor Google Buzz.

User ac­tiv­ity is sim­i­larly pretty low. On 2011-10-13, I took a look at Knol’s “What’s New” sec­tion on its front page.

Views Ver­sions Edited
3,000 83 6 min­utes ago
900 23 45 min­utes ago
90 20 One hour ago
80,000 59 One hour ago
10,000 58 5 hours ago
23,000 74 5 hours ago
7,000 90 8 hours ago
1,000 29 9 hours ago
19,000 135 9 hours ago
300 7 10 hours ago

Over half a day for the en­tire site, 10 knols were changed in some fash­ion; ed­its prob­a­bly are not huge up­dates given the rel­a­tively large num­ber of re­vi­sions like 135 or 90. A quick­-and-dirty es­ti­mate of the user­base: 1 edit per hour, each by a unique user, who ed­its on av­er­age once a week im­plies a user­base of .

Traffic-wise, one of them has a re­spectable 80k views (an ac­count­ing knol), but this seems to be life­time views. Wikipedia, of course, has more like 10 ed­its a sec­ond, and even a dry ar­ti­cle like will pick up ~62k views a month. (I could­n’t find a di­rect equiv­a­lent for that knol—Wik­i­books has so many ac­count­ing-re­lated texts, and Stat­s.­grok.se does­n’t cover Wik­i­book­s.) An­other amus­ing com­par­i­son is see­ing how often Knol is sub­mit­ted to Red­dit ver­sus Wikipedia sub­mis­sion. In Au­gust 2009, TechCrunch re­ported on Knol’s de­clin­ing traffic as mea­sured by ; in Oc­to­ber 2011, Quant­cast’s es­ti­mates shows Knol cy­cling from a high of <300k monthly vis­i­tors to a low of ~170k, with ~260k at the end­point in Ju­ly, al­though it ap­pears to have never again reached its Jan­u­ary 2009 peak of 300k monthly vis­i­tors. (For com­par­i­son pur­pos­es, one site claims that the num­ber of global In­ter­net users in­creased from 1.596 bil­lion in March 2009 to 2.110 bil­lion in June 2011; Wikipedia con­tin­ues to re­main pop­u­lar, #6 world­wide ac­cord­ing to Alexa.)

Knol death-watch

So even in ab­solute num­bers, Knol has stag­nated in most every re­spect. Some of the links above are from months or years ago—no one cares enough about Knol even to dis­cuss it. (Some­what like .) Knol has all the char­ac­ter­is­tics of a doomed Google ser­vice, which leads to an in­ter­est­ing ques­tion: when will Google kill off Knol?

To pre­dict, we should con­sider past ex­am­ples: ser­vices can linger for a very long time, en­tirely ne­glect­ed, or die very quick­ly. Which group does Knol fall in­to?

  • Ya­hoo: is very like Knol in be­ing a col­lec­tion of pub­lic con­tent cre­ated by in­di­vid­u­als and hosted on a cor­po­rate site, and was ig­nored by Ya­hoo for al­most a decade be­fore its shock­ingly abrupt shut­down, did lit­tle bet­ter while is still with Ya­hoo (and still ne­glect­ed).

  • Google: A num­ber of ser­vices with steady users, like or Google Wave, have been killed off. We can look at them one by one:

    1. Google An­swers: pub­lic con­tent; clo­sure an­nounced No­vem­ber 2006, shut down De­cem­ber 2006. All the archived ques­tion-an­swers and pages are still hosted by Google.
    2. : pub­lic con­tent ap­pear­ing in Google search­es; vis­i­bil­ity elim­i­nated Oc­to­ber 2009, APIs dep­re­cated De­cem­ber 2010. Un­clear where the con­tent wen­t—seems to’ve been in­cor­po­rated into other Google ser­vices.
    3. : pri­vate ser­vice; clo­sure not pre­vi­ously an­nounced, No­vem­ber 2010. The con­tent is pri­vate and folded into Google’s au­dio analy­sis al­go­rithms.
    4. : in­tro­duced July 2008, closed De­cem­ber 2008. Ap­par­ently lit­tle or no user-gen­er­ated con­tent, and the con­tent/­code­base is un­avail­able.
    5. : clo­sure an­nounced Jan­u­ary 2009, closed Au­gust 2009. User-gen­er­at­ed, their con­tent ap­par­ently made avail­able in­defi­nitely on Code.­google.­com (un­clear how use­ful that would be to the users)
    6. : clo­sure an­nounced 2007, user-signups dis­abled Sep­tem­ber 2008, closed 2009. User-gen­er­at­ed, all con­tent shifted to host­ing on and pre­sum­ably still avail­able.
    7. : opened No­vem­ber 2008, closed March 2010. User-gen­er­ated but mostly pri­vate, the con­tent (such as it is) is avail­able to users in pri­vate.
    8. Google Video Mar­ket­place: an­nounced 2006, closed Au­gust 2008. All con­tent un­avail­able, but the con­tent ap­pears to have been com­pletely com­mer­cial, and none of the con­tent seems to be unique, so the dis­ap­pear­ance is­n’t a big con­cern.
    9. : some­thing of a joke or mis­take; ir­rel­e­vant.
    10. : clo­sure an­nounced Oc­to­ber 2011, closed Jan­u­ary 2012. Pre­sum­ably en­tirely un­avail­able, but as a search util­i­ty, it was pri­vate use on­ly.
    11. so­cial fea­tures & : clo­sure an­nounced Oc­to­ber 2011, closed No­vem­ber/De­cem­ber 2011. User-gen­er­ated con­tent, but the posts will re­main pub­licly vis­i­ble and users can down­load their ma­te­r­i­al.
    12. : clo­sure an­nounced Oc­to­ber 2011, closed Jan­u­ary 2012. User-gen­er­ated con­tent; ap­par­ently Google will not con­tinue to host the con­tent like Buzz or Page Cre­ator, but the code­base was open-sourced in 2009 and Google will let users down­load their own ma­te­r­i­al.
    13. Aca­d­e­mic API ac­cess: clo­sure an­nounced Oc­to­ber 2011, closed Jan­u­ary 2012. Pri­vate, for aca­d­e­mics on­ly; no user-gen­er­ated or pub­lic as­pects. (Pre­sum­ably most aca­d­e­mics us­ing it have pub­lished their re­sults al­ready.)

Look­ing through the most rel­e­vant ex­am­ples, Google tends to shut down ser­vices fairly fast and is sur­pris­ingly good about pre­serv­ing data, in line with its data lib­er­a­tion views. Other ex­am­ples are rel­e­vant: Google elim­i­nated file stor­age for its , it gave the lists some­thing like half a year of lead­-time, and its Usenet archives have been poorly main­tained since 2001 with­out be­ing elim­i­nat­ed.

That it might do so is not a great chal­lenge: disk space is cheap­—frac­tions of pen­nies per gi­ga­byte—and Google could shift Knol to all-, elim­i­nat­ing any soft­ware main­te­nance over­head. Un­like Google Note­book, knols are in­her­ently pub­lic works, meant for pub­lic dis­tri­b­u­tion and con­sump­tion, so it may be bet­ter to com­pare Knol against Google Groups rather than Google Note­book. Knol has not been shut down yet 2 years after its lack of no­tice­able suc­cess be­came clear, sug­gest­ing that it won’t be shut down any time soon. Weigh­ing Google’s usual an­nounce lead­-time, the pub­lic na­ture of the con­tent, its sur­vival un­til now, and so on sub­jec­tive­ly, these prob­a­bil­i­ties feel about right:

  • Google Knol will be shut­down in 2011: 10%
  • Google Knol will be shut­down in 2012: 25%
  • Google Knol will be shut­down in 2013: 25%
  • Google Knol will be shut­down in 2014: 20%
  • Google Knol will be shut­down in 2015: 15%
  • Google Knol will be shut­down in 2016: 5%2
  • Google Knol’s shut­down will leave the knols’ con­tent pub­licly ac­ces­si­ble at some URL: 90%

(In March-May 2013 I con­ducted to come up with some pre­dic­tive model of shut­downs; while ob­vi­ously Knol had been shut­down al­ready, I still used the model to cal­cu­late a 5-year sur­vival es­ti­mate for Knol from the day the shut­down hap­pened, which turned out to be a suit­ably-low 43%.)

Chronicle of a Death Foretold

On 2011-11-22, Google put the spec­u­la­tion to rest when it an­nounced:

We launched Knol in 2007 to help im­prove web con­tent by en­abling ex­perts to col­lab­o­rate on in­-depth ar­ti­cles. In or­der to con­tinue this work, we’ve been work­ing with Solvi­tor and Crowd Fa­vorite to cre­ate An­no­tum, an open-source schol­arly au­thor­ing and pub­lish­ing plat­form based on Word­Press. Knol will work as usual un­til April 30, 2012, and you can down­load your knols to a file and/or mi­grate them to Word­Press.­com. From May 1 through Oc­to­ber 1, 2012, knols will no longer be view­able, but can be down­loaded and ex­port­ed. After that time, Knol con­tent will no longer be ac­ces­si­ble.

This im­me­di­ately re­solved the above 7 pre­dic­tions. I was cor­rect to most ex­pect Knol’s shut­down to come in 2012 or 2013 (and not ear­lier or lat­er), how­ev­er, my fi­nal pre­dic­tion sig­nally failed—the an­nounce­ment is clear that the knol con­tent will not be hosted in­defi­nitely but will be deleted3 (‘ac­ces­si­ble’ be­ing an eu­phemis­m). Why did I blow it? My rea­son­ing was based on one of my stan­dard tools while : the Out­side View. Google had kept pub­lic con­tent hosted in the past when it shut down projects (see the pre­vi­ous list), so it would prob­a­bly con­tinue to do so. The Out­side View tech­nique fails, how­ev­er, when the same fun­da­men­tals that give it pre­dic­tive power do fi­nally change and a new regime is en­tered. In ret­ro­spect, there may have been such a regime change within Google when be­came CEO in April 2011 and be­gan push­ing for a Google-wide fo­cus on so­cial net­work­ing, with a ruth­less ap­proach to projects which were not suc­ceed­ing or con­nected to so­cial net­work. This was all laid out in Wired’s “In­side Google+—How the Search Gi­ant Plans to Go So­cial”, which I read at the time but ex­am­ined more for how Google+ might differ from Face­book than how it might affect al­lo­ca­tion of Google re­sources.


  1. Not to den­i­grate the efforts of Mag­nus Manske and Tim and all the other Me­di­aWiki de­vel­op­ers over the years, but one sim­ply ex­pects more of Google’s sup­pos­edly world-class ful­l-time de­vel­op­ers us­ing the fa­mous Google in­fra­struc­ture.↩︎

  2. I can’t re­ally see Knol sur­viv­ing com­pletely in­tact un­til 2017; I’m sure enough that it’ll be changed some­how be­fore get­ting near its 10th an­niver­sary.↩︎

  3. Thank­ful­ly, Google be­ing Google, they are be­ing fairly gen­er­ous in mak­ing it eas­ier to save one’s stuff. Ex­port­ing still re­quires ac­tion on the au­thor’s part, which will fre­quently not be taken (eg. if the au­thor has died), and so hope­fully the Archive Team will be able to do a clean sweep.↩︎