Someone at Amazon’s Seattle headquarters probably has the strangest job in tech: explaining why their recommendation algorithm thinks explosive chemicals and ball bearings make a great combo purchase.
Back in 2017, British journalists at Channel 4 News were poking around Amazon when they stumbled onto something disturbing. The site’s “Frequently Bought Together” feature was helpfully suggesting bomb-making ingredients to anyone shopping for basic chemicals.
Not kidding. Buy some potassium nitrate (perfectly legal food preservative), and Amazon would cheerfully recommend sulfur and charcoal to complete your black powder starter kit. Need some thermite ingredients? Amazon’s got suggestions for that too, plus some ball bearings in case you want to add shrapel.
The Channel 4 investigation found that “ingredients for black powder and thermite are grouped together under a ‘Frequently bought together’ section on listings for specific chemicals.” The reporters managed to fill a shopping cart with 45 kilograms of explosive materials before Amazon caught on and started damage control.
When Robots Learn from Weird Humans
Here’s the thing about Amazon’s frequently bought together algorithm—it’s not actually smart. It just watches what millions of people buy and assumes that if enough customers put two things in the same cart, they must go together.
This works great for obvious stuff. Buy a camera, get suggested a memory card. Buy a phone, here’s a case. Makes sense.
But humans are weird. We buy random combinations of stuff for reasons no computer could understand. Maybe you’re shopping for a white elephant gift exchange, or you’re having a mental breakdown and need ice cream AND a self-help book AND batteries for some reason. Or maybe you’re just trying to hit the free shipping minimum.
The algorithm doesn’t know. It just sees patterns and assumes they mean something.
The Headphone Deodorant Conspiracy
Take this gem discovered by someone on the Linus Tech Tips forum: Audio-Technica ATH-A990Z headphones were being paired with Old Spice deodorant in Amazon’s suggestions.
Their theory? “When you get some new headphones and sit down for a long sweaty listening session, you need to smell good.”
Honestly, that’s not the worst logic. Gamers ordering new gear probably do think about hygiene for those marathon sessions. But imagine being the Amazon engineer who has to explain why their AI thinks audio equipment and antiperspirant are related products.
The Gaming Community’s Greatest Hits
Gaming forums are goldmines for weird Amazon combinations. The Giant Bomb community once noticed that Zojirushi thermoses were getting paired with random gaming accessories—entirely because their community had an inside joke about the brand during some livestream.
Other highlights include:
- PlayStation TV suggested alongside completely random household items
- People deliberately buying games with “random comical household objects just to play hell with this algorithm”
- One user’s observation about buying “Final Fantasy X, toilet roll and massage oil all in one order”
Why Amazon’s Algorithm Gets Confused
Amazon processes millions of orders daily. Even if 0.01% of customers make weird purchases, that’s still thousands of people creating “patterns” for the algorithm to find.
Some reasons for strange frequently bought together combinations:
Gag gifts: People love buying ridiculous stuff together for laughs. Order a pregnancy test with novelty underwear? Algorithm thinks they’re related.
Emergency shopping: Your toilet breaks at 2 AM, so you order a plunger, candles (in case the power goes out), and stress-relief chocolate in one panic purchase.
College students: Moving into dorms creates the weirdest shopping lists. Electronics, hygiene products, ramen, energy drinks, and a desk lamp all seem essential.
Sellers gaming the system: Smart sellers figured out they can create promotional bundles to “teach” the algorithm new associations, then unbundle them later once Amazon learns the pattern.
Real Examples You Can Still Find
Pet Products with Identity Crises
The Marshall Pet Banana Hammock (a ferret bed shaped like a banana) gets suggested alongside actual fruit storage and even men’s underwear, purely because of the word “banana.”
Survival Gear Clusters
Camping and survival gear creates some of the most logical-yet-bizarre combinations. A simple camping hammock might get paired with emergency food, tactical flashlights, and water purification tablets. Not because they’re related, but because certain types of customers tend to buy all these things.
The Pregnancy Test Ecosystem
Look at any pregnancy test listing and you’ll find interesting combinations. Tests often get paired with ovulation predictors, but also with completely unrelated stress relief items and comfort foods. The algorithm has learned that people buying pregnancy tests are often in emotionally heightened states.
Amazon’s Damage Control
After the bomb ingredients story broke worldwide, Amazon went into full PR mode. They issued the standard corporate response about “reviewing our website to ensure products are presented appropriately” and promised to “work closely with law enforcement.”
Amazon’s official statement said: “All products sold on Amazon must adhere to our selling guidelines and we only sell products that comply with U.K. laws. In light of recent events, we are reviewing our website to ensure that all these products are presented in an appropriate manner.”
Translation: They probably hired some intern to manually flag dangerous combinations and added some basic filters to catch the most obvious problems.
But the fundamental issue remains. Amazon’s scale is so massive that manually reviewing every possible product combination is impossible. They’re stuck trying to balance algorithm efficiency with not accidentally helping terrorists.
How the Algorithm Actually Works
Amazon’s recommendation system analyzes massive datasets of customer purchasing behavior to identify products commonly bought together. According to McKinsey research, recommendation engines like Amazon’s drive up to 35% of the company’s revenue.
The system considers:
- Purchase history patterns across millions of transactions
- Shopping cart combinations from completed orders
- Browsing behavior and product view sequences
- Customer demographics and buying preferences
- Seasonal trends and timing correlations
When enough customers buy seemingly unrelated items together, the algorithm assumes they’re complementary—regardless of logical connection.
What This Says About Us
The weird part isn’t that Amazon’s algorithm makes strange suggestions. It’s that humans are making these purchases in the first place.
We’re all just stumbling through life, buying random stuff for reasons that make sense to us in the moment but look odd in a spreadsheet. Someone really did buy headphones with deodorant enough times for Amazon to notice a pattern.
Maybe that person was preparing for a gaming marathon, or maybe they were shopping for a smelly friend. Maybe they just needed both things and Amazon was convenient.
The algorithm doesn’t judge. It just watches and learns from our chaos.
The Bigger Picture
Every recommendation system faces this problem. Netflix suggests movies based on what you’ve watched, but doesn’t know you fell asleep during that documentary or that your kid hijacked your account. Spotify thinks you love death metal because you forgot to turn off shuffle.
Amazon’s “Frequently Bought Together” is just the most visible example of machines trying to understand human behavior and failing spectacularly.
The next time you see a bizarre product suggestion, remember: somewhere out there, enough real people made that exact purchase for Amazon’s computer to think it was worth recommending.
And honestly? That might be the most human thing about online shopping. We’re all just making it up as we go along, buying weird combinations of stuff and confusing the robots in the process.
At least until they figure out how to read our minds. Then we’re really screwed.
Want to understand your customers’ real shopping patterns? Our team helps brands navigate marketplace analytics and customer behavior insights to drive growth.