I went looking for the source of one little statistic. That was easier said than done.

Apparently, 75% of your emotions are generated by smell.

Or that’s what the internet says, anyway. The statistic appears everywhere from marketing books and articles to academic papers about the power of scent. As a perfumer (someone who creates fragrances) and journalist who writes about these things, I came across it several times, and when I finally wanted to cite it for an article about the impact of scent on mood and memory, I went looking for the original research.

But I couldn’t find it.

This wasn’t a totally new experience for me. One time, while working on a workplace story as the editor of a hospitality trade magazine, I couldn’t find the source of an employee retention statistic, so I wrote the people who included it in a study. They wrote back to say they had no idea what I was talking about.

In this case, there was no such dead end. I followed link after link, figuring surely I would get to the bottom of it, but every source pointed to another source, which pointed to another, and somehow none of them led to the original research I was looking for.

I asked ChatGPT, Google Gemini and Claude for help. They gave me links to wellness blogs, aromatherapy companies, and authoritative-looking PDF reports. But as I ruled each one out as the original source, the answers became more circular, and the trail ate itself.

Me: Where did the statistic that 75% of emotions are generated by smell come from?

Google Gemini The Sense of Smell Institute.

Me: Great. Which study?

Gemini: There isn’t a specific study. The statistic was popularized through scent marketing research.

Me: So where did the number come from?

Gemini: Martin Lindstrom’s 2005 book Brand Sense.

Me: Does the book contain the original research?

Gemini: Lindstrom commissioned extensive research into sensory branding through the research firm Millward Brown.

Me: Does that research establish that 75 percent of emotions are generated by smell?

Gemini: Would you like to know how lavender and citrus can influence your mood?

A breakthrough! Or so I thought

Lindstrom did look like a real lead. Several writers and businesses traced the number to the Danish author’s 2005 book Brand Sense: Sensory Secrets Behind the Stuff We Buy. Simon & Schuster’s page for the book says it outright: “Research shows that a full 75% of our emotions are in fact generated by what we smell.” Aha!

I bought the book and read it. While there is talk of emotions and the senses, the number isn’t in there.

Next, I wrote to Lindstrom’s company asking them to point me toward the original research. They sent me a copy of Brand Sense and told me I would find the research in the book.

I tried another route. Brand Sense was based in large part on an international study by market research company Millward Brown. The 75% figure has been attributed directly to that research. So I went looking for the Millward Brown study. But I couldn’t find it anywhere.

Does 75% even make sense?

There were other questions: Does the stat even make scientific sense? How would you measure the percentage of emotions generated by smell? Do people with no sense of smell experience 75% fewer emotions? And what about emotions that seem to have nothing to do with the senses at all, like grief over someone’s death, a breakup, or anxiety over losing your job?

I asked Kristen Lindquist, a social and affective neuroscientist and psychology professor at Ohio State University who studies emotion. She had never heard of the statistic.

“I imagine if somebody had done that study I would be aware of it,” she told me.

There is a relationship between smell and emotion, and it may be possible to measure something like this, she said. Researchers could, for example, ask people throughout the day how they’re feeling and whether what they’re feeling is related to something they’re seeing, hearing, touching or smelling. But that still wouldn’t make emotion the product of one sense. “Emotions really arise from all of our sensations,” Lindquist said, including signals from within our bodies. Our brains process all of that information together to figure out what is happening and what it means for us.

So, what about that breakup?

“If you get dumped and you smell some nice cookies, are you going to feel 75% better?” Lindquist said. “Probably not.”

When repetition becomes evidence

There’s a name for the particular way an unsupported claim can acquire the appearance of fact through citation: citogenesis. Coined by Randall Munroe in an XKCD comic, the term describes the process by which an unsupported claim gets repeatedly cited until the citations themselves make it look like truth.

I called Jodi Schneider, an associate professor at the University of Wisconsin-Madison Information School whose research includes scientific evidence and citation integrity, to talk about it. I sent her the 75% statistic, and she went looking for the source herself.

“It was not a casual ‘Let me look for 20 minutes, and there it is,'” she told me. “I would find a thing, [it] would go to something else, and that would be a dead end.”

Her conclusion: “It’s hard to source.”

The problem, Schneider said, is that repeating a claim is much easier than investigating it. “Somebody says something, it sounds like a thing you’d like to know or repeat, and you don’t trace it back to where it came from. You just pass it on.” Then repetition becomes the “evidence,” and “if you hear the same thing from multiple places, it’s harder to question.”

This psychological phenomenon is known as the illusory truth effect: We’re more likely to believe information we’ve encountered before. The effect was first demonstrated in the 1970s and has since been replicated many times. More unsettlingly, research has found that repetition can make a false claim seem more believable even when it contradicts something we know to be true.

In other words, seeing the same statistic everywhere doesn’t just make it look well sourced; it makes it feel true.

It’s hard to fully check every claim. Writers build arguments from dozens of pieces of information, and tracing each one all the way back to its origin can take an absurd amount of time.

Newsrooms have spent years cutting staff and budgets, including the fact-checkers and copy editors whose job was to catch this kind of thing. At the same time, the volume of information journalists have to sort through has exploded. Finding a source for a claim can take seconds. Establishing where it actually came from can take hours (or an eternity).

Meanwhile, the consequences of not checking facts compound. Once a bad claim gets published, it becomes another source for the next journalist, another search result for the next reader and another apparent confirmation that the claim must be true.

Then came AI

AI is making that problem worse. It repeats the same unsupported claims while stripping away some of the clues that might otherwise make us suspicious of the source.

“The polish that generative AI can put on just about anything is already a challenge,” Schneider said. More importantly, “Generative AI is really, really good at being persuasive, even if the claims inside the persuasion have no basis in reality.”

Most people are not going to track a statistic through six articles, buy the book everyone cites, email the author, and start calling academics. They don’t have the time or energy for that, and they shouldn’t have to do it anyway. One of journalism’s roles is to provide information that’s more reliable than TikTok or whatever nonsense Google Gemini spits out.

When we get it wrong, we don’t just add another bad source to the pile. We give the claim credibility and put it in front of more people. And once they’ve seen it, the illusory truth effect suggests they’re more likely to believe it the next time. (Research has also found that repeated misinformation is more likely to be repeated than new misinformation, and can continue to influence people even after it has been corrected.)

Nobody is going to die because they believe smell generates 75 percent of their emotions. But replace that claim with one about a vaccine, a war, an election or a medical treatment and the same failure becomes considerably less benign.

That’s why the difference between finding a source and finding the source matters. I found hundreds of sources for the 75% statistic. I still haven’t found the source. That doesn’t mean it’s not out there.

I just can’t find it.

Can you find the source of the 75% stat? Email staff@niemanlab.org.