How an algorithm came up with Amazon's KEEP CALM AND RAPE A LOT t-shirt

You may have heard that Amazon is selling a "KEEP CALM AND RAPE A LOT" t-shirt. How did such a thing come to pass? Well, as Pete Ashton explains, this is a weird outcome of an automated algorithm that just tries random variations on "KEEP CALM AND," offering them for sale in Amazon's third-party marketplace and printing them on demand if any of them manage to find a buyer.

The t-shirts are created by an algorithm. The word “algorithm” is a little scary to some people because they don’t know what it means. It’s basically a process automated by a computer programme, sometimes simple, sometimes complex as hell. Amazon’s recommendations are powered by an algorithm. They look at what you’ve been browsing and buying, find patterns in that behaviour and show you things the algorithm things you might like to buy. Amazons algorithms are very complex and powerful, which is why they work. The algorithm that creates these t-shirts is not complex or powerful. This is how I expect it works.

1) Start a sentence with the words KEEP CALM AND.

2) Pick a word from this long list of verbs. Any word will do. Don’t worry, I’m sure they’re all fine.

3) Finish the sentence with one of the following: OFF, THEM, IF, THEM or US.

4) Lay these words out in the classic Keep Calm style.

5) Create a mockup jpeg of a t-shirt.

6) Submit the design to Amazon using our boilerplate t-shirt description.

7) Go back to 1 and start again.

There are currently 529,493 Solid Gold Bomb clothing items on Amazon. Assuming they survive this and don’t get shitcanned by Amazon I wouldn’t be at all surprised if they top a million in a few months.

It costs nothing to create the design, nothing to submit it to Amazon and nothing for Amazon to host the product. If no-one buys it then the total cost of the experiment is effectively zero. But if the algorithm stumbles upon something special, something that is both unique and funny and actually sells, then everyone makes money.

Dictionary + algorithm + PoD t-shirt printer + lucrative meme = rape t-shirts on Amazon



  1. The world’s first t-shirt spam? Do we need thousands of random variations on this theme? Is there really someone out there looking for a Keep Calm and Wipe On t-shirt?

    1. Not the world’s first; I’ve seen other companies doing similar things with t-shirts and other products, mostly with agrammatical, useless results.

      Obviously we do not need thousands of variations on the theme, and nobody is looking for “Wipe On.” That’s sort of the point of the article: throwing everything to the wall and seeing what sticks is easier than ever, and no longer even requires human involvement, so this’ll keep happening.

      In fact, it might start happening a lot, very soon. I’m sure this has generated lots of publicity for Solid Gold Bomb, and made people who wouldn’t otherwise have heard about their products aware of their products. They’ll probably sell more t-shirts as a result. This will cause more companies to make sure to include edgy/offensive words in their algorithmically-generated t-shirt designs. Eventually one of those designs will become momentarily popular (in my imagination, this involves Ke$ha somehow. YIMV.), and we’ll all have another way to identify horrible people from a distance.

    2.  Don’t know about t-shirts but Amazon is clogged with “books” scraped together from fragments of Wikipedia and titled so as to confuse people with the book they’re actually looking for. It’s nasty.

      1. Yup. I don’t need the equivalent of CafePress spam in all marketplaces, as the Kindle wikiscammers did in the ebook market.

      2. Amazon is clogged with “books” scraped together from fragments of Wikipedia and titled so as to confuse people 

        And also books printed from public-domain sources, without any of the original illustrations, sometimes rendered incomprehensible by OCR.

  2. It only got a 1 star review. Sadly, the page is taken down so I can’t read the reviews. I was hoping to hear about people who got their asses kicked while wearing this shirt.

  3. The word “algorithm” is more than a little scary to me because of the mental image it conjures of our Nobel-winning former VP attempting to boogie down.

  4. Okay, I see an annoyance: first I find out Google apps are superior on iOS, and it looks like the Amazon app for iOS might be better than the Amazon app on the Kindle Fire.

  5. It costs nothing to create the design, nothing to submit it to Amazon and nothing for Amazon to host the product.

    I think that’s an overly simplistic way to look at it.

    There may be practically no cost in running the algorithm and posting the result to Amazon. But cluttering up the store shelf, even a virtual one, makes it that much harder for customers to find something they actually want to buy. It hurts sales of any product that actually has a chance, when you bury them in a big pile of shit.

    There’s no way I’d sift through half a million t-shirt examples, not even close. After the first twenty or thirty useless ones, I’d give up and go shop somewhere else.

    If Amazon’s isolated this vendor to their own store, great. At least the vendor is hurting only themselves. But if these t-shirts are advertised next to t-shirts from other vendors, then everyone loses.

    1.  That’s a real cost, but not a cost to the t-shirt company, and I’m sure they don’t care about anyone else. They’re pissing in the swimming pool because it’s convenient for them to do so.

      1. Not being able to sell their own t-shirts isn’t a cost to the t-shirt company?


        (Though, I see now that the whole thing was just a big mistake. But still…it’d be a mistake even if there really was someone trying to sell a half-million different variations on the t-shirt, and even for the vendor selling the shirts themselves).

        1. It seems to be a mistake only in the sense that it was a really bad idea, not in the sense that they created and listed their million nonsense variations unintentionally.

          1. Yes, to me that seems like a valid use of the word “mistake”. Though it does seem that they never intended to actually sell those shirts.

            I’ve made mistakes with intent. Hasn’t everyone?

            Poor judgment, to be sure. But to my point about it hurting sales, it appears it was never really the intent to expect anyone to sift through, never mind buy one of the half-million or so variations.

            I did not mean to suggest that the person responsible didn’t intend to list all those t-shirts. But neither does it appear that they thought they were engaging in real commerce either. They were making a statement, in a poorly conceived way.

            Thus, “mistake”.

    2.  What if they show a random assortment of t-shirts to each site visitor? 50% can be the most popular ones, the other 50% are truly selected at random, a kind of mechanical turk where the reward for the user is a cool shirt.

      1. If. Sure, as long as the total number of shirts any one consumer can see is small enough, your suggestion has the potential to suss out the shirts with potential without hurting sales as much (they will still lose some, but at least not all consumers would be affected).

        I was not under the impression that this particular situation was implementation that way.

    1. Was this apology created by an algorithm, too? “I apologise for the offensive response this has created across the world.” Apologies for the response. Huh.

      1. yeah, but one paragraph later:

        This was a computer error of my creation and I accept my responsibility in the matter.

        He repeats a variation of accepting responsibility several other times.

        But, you know. first impressions, tl;dr, right?

    1.  But according to Todd if you enjoy the rape you’ll become pregnant. I guess that gives “keep calm and carry on” a whole new meaning.

  6. Bummer. I really preferred the world of novelty t-shirts created by stoners & old men who both happened to understand how iron-ons, silk-screens & printing in general worked. 

    FWIW, I just tried to do a Google search for vintage novelty t-shirts & it’s all seemingly filled with SEO listings from modern t-shirt vendors selling neo-faux-retro t-shirts. So can we assume that the Internet & automation has really glutted the t-shirt business with dreck?

    1. No algorithm could ever come up with “Keep on Truckin’” I mean, what does that even mean?

  7. Amazon recommendation algorithm in a nutshell:

    We noticed that you recently bought Jim Butcher’s Cold Days in hardback.  Perhaps you’d also be interested in buying Jim Butcher’s Cold Days in hardback.

    1. Seriously.  Spewing 14 different variations and version of the same book back at you after having bought one version is annoying.  You would think that would be one of the simpler things to cull from the algorithm.

      When Amazon gets it right, they get it very right.  I bought some board games and Amazon spit back in an e-mail a dozen other board games that I might also like, and they were right.  I threw money at them.  When they get it wrong though, they get it stupid wrong, like in the cases of getting offered three versions of the same book I just bought.  Don’t get me wrong, I like Dresden, mildly misogynistic writing aside (is there a better word than misogyny… it isn’t even misogyny, it is just painfully awkward when it comes to women), but, um, one copy is enough.  Thanks.

      1. After that dialogue with Titania in the gay cruising area, I’ve given up on trying to figure out his sexual politics.  I’m pretty sure that he hasn’t figured them out yet, but I don’t think that he means any harm.

    2. Good gods Amazon is annoying that way – and it’s not just trying to resell books, but every product.  I can’t seem to buy anything without getting emails from Amazon suggesting I’d like to buy another variation of that exact same thing.  When did that algorithm get so broken, or was it always so and I just ignored it previously?

    1. The short words that ended the sentences were compiled by hand, probably more randomly than misogynistically, which explains why there is a “Keep calm and love her” shirt but no “Keep calm and love him”.

  8. Where did they get this list of verbs that they’re algorithmically recommending to people (in addition to keeping calm)? Why didn’t anyone look through it, or, worse, why didn’t the person looking through it just decide we oughtn’t have shirts advocating rape?

    1. In this post-Google world, any human interaction would ruin the pristine nature of the algorithm.

      And that sounds like real work, anyway, which is not the point of this business model.

  9. A couple years ago there was an Amazon Mother’s Day gift book suggestion – the Guide to Anal Sex. 

    1.  And your point is….?

      I’m sure I know some mothers who would find it a very thoughtful gift.

      Me, I’d rather have European chocolate (insert your own joke here)….proof that there’s no accounting for tastes.

  10. Surprised no one has pointed out that ‘A LOT’ should be one word. Shouldn’t it?

  11. Why do I see a bunch of people trying to find the MOST ridiculous t-shirt out of 529,493 in an ongoing game of sartorial one-upsmanship? I think a standard color should be picked, so you’ll know when you see an intentionally absurd t-shirt combination in the field.

  12. Great t-shirt.  If the text were white, on white fabric, I’d wear it.

    (Inside out, just to play it safe.)

    But seriously.  Aside from the brief Zach Galifianakis phase that I (and many of us paunchy, bearded, nerdy guys probably went through a while back) I’ve never really been a wearer of t-shirts with clever sayings, graphics, memes or advertising (retro or non) of any kind on ’em.  I’m more of a solid brown (John Locke) or gray or OD green Beefy-T guy, myself.  $5.64 each all day long, online.  Hell yeah!!!

  13. It might be interesting to estimate what proportion would be tautological (essentially, variations on a theme of KEEP CALM AND KEEP CALM) or oxymoronic (likewise, KEEP CALM AND PANIC LOADS).

  14. A fascinating story, but what I found most interesting was something in the company’s formal apology: the algorithm’s creator claims that the procedurally generated shirts were a form of parody of the ludicrous proliferation of ‘carry on…’ shirts, rather than the logical conclusion of it. A case of Poe’s law for craven capitalism?

  15. How did the verb “rape” get in their database in the first place? If the word wasn’t there, the algorithm wouldn’t be able to generate such sentence. 

    1.  Not until you wrest my bright red coffee mug from my cold dead hands.

      Which is kind of funny, because it was a gift from a friend when she learned I was going through chemo.  What a brilliant choice!  (No, I’m not dying.  But I will keep the mug until I do.)

      1. True story: My Mom (not known for her great gifts) bought my daughter a “Keep Calm and Carry On” shirt in London because she thought it was, “unique.” Yeah, never seen one of those before.

        Coffee cup was a nice giftie.

  16. Except for possible triggers, it does seem like a good way to let people know whether you’re someone worth talking to.

  17.  I think in this case the triggers are a feature, not a bug. I hope some of those triggered might resort to justified violence to deal with such an obvious threat.

  18.  Oh, ghods, how I miss disemvoweling.  I wish Antinous had been provided a tool to do it instead of having to do it by hand (I should have written him one… PHP is dead easy).

    Disemvoweling did a wonderfully elegant job of short-stopping the echo chamber effect that post deleting encourages.  It let Antinous have three levels of response.

  19. Does Disqus allow a moderator to forbid editing on certain comments? I guess it’s not the most obvious feature, but it’s certainly got practical applications.

  20. Disqus massively downgraded the quality of BB threads.  For starters, by removing a barrier to entry – you used to have to really WANT to say something, not just be a drive-by with a google ID.  (And the barrier was not insignificant at times; I never quite understood how wordpress could be made so hostile to logins).  Also, because disqus uses so many separate authentication sources, it inherently encourages sock-puppetry…  I’m using a BB id right now, but honestly most people probably have half a dozen valid disqus ids and can use three or four simultaneously by just using multiple browsers.  Finally, while you know (a recent incident comes to mind) that I have used the edit ability a lot, it does mean that threads are less coherent overall since everyone can back-edit themselves to look taller or smarter (or, in my case, more polite) than they really are.

Comments are closed.