It was tricky to find specific bugs associated with specific foods, but the presence of one particular microbe was a strong indicator of whether or not a person had been drinking coffee. Basically, if you’re a coffee drinker, a microbe in your feces will give you away.

The new study, by Hannah Holscher at the University of Illinois at Urbana-Champaign and her colleagues, takes a slightly different approach. Here, the team looked at fecal samples from volunteers who ate set amounts of specific foods on a daily basis. And rather than look at the presence of microbes themselves, Holscher’s team looked for metabolites—the chemicals microbes produce when they break down food.

The team looked at the impact of six specific foods: almonds, avocados, broccoli, walnuts, barley, and oats. The researchers first looked to see if there were any links between metabolites in poo and whether a particular person had eaten any of these foods. They used any patterns they identified to guess whether other people had eaten the same foods.

Again, it was tricky—but the team was able to tell whether people had eaten almonds, broccoli or walnuts with 80 to 87% accuracy, depending on the food. The study was published online at the preprint server bioRxiv and has not yet been peer-reviewed. But it builds on similar work the team published last year.

Studies like these provide a tantalizing glimpse into the potential future of fecal analysis. It’s early days, and the accuracy of these tests is likely to improve over the coming years. But the ability to understand the impact of individual foods on our microbiomes, and our health, could revolutionize research and nutrition. “This is really the frontier of what’s next,” says Emily Leeming, a nutrition scientist at Zoe, the maker of a personalized nutrition app, who coauthored Berry’s study.

Holscher’s team hopes to improve nutrition research. Studies that aim to figure out how certain foods affect our health usually rely on volunteers to keep food diaries. They’re a pain to maintain, and they’re usually inaccurate or incomplete. Analyzing a person’s poo instead could one day provide a painless alternative.