Olfactory Research & Our Scoring

The science of smell is advancing rapidly. Here are the research threads that shape how we think about fragrance similarity — and what they might mean for AI-powered scent analysis.

Osmo.ai — Google DeepMind

Molecular Smell Prediction ↗

Osmo.ai, backed by Google DeepMind, has built a machine learning model that predicts how a molecule smells from its chemical structure alone. Their Principal Odor Map maps thousands of aroma compounds into a unified odor space — meaning chemical similarity can now be translated directly into perceptual similarity.

How it informs our scoring

This research validates AI-powered similarity scoring. Rather than relying solely on note names (which are subjective descriptors), future scoring can incorporate molecular-level distance metrics. We plan to integrate the Osmo API when a public endpoint is available.

MIT Media Lab — Smellnet

Linguistic Olfactory Mapping ↗

MIT's Smellnet project maps the linguistic space of smell — how different cultures and languages describe the same odours, and what that reveals about the perceptual structure of olfaction. The research shows that smell descriptors cluster into consistent families across languages, suggesting that "woody-amber" and "oriental-balsamic" are real perceptual categories, not just marketing terms.

How it informs our scoring

This supports our vibe and note page taxonomy. When we categorise fragrances as "woody-oriental" or "powdery-feminine," we're using descriptor clusters with genuine perceptual validity.

E-Nose Research — Overview

Electronic Nose Technology

Electronic noses use sensor arrays to detect volatile organic compounds (VOCs) and classify odours. Modern e-nose research has demonstrated reliable discrimination between fragrance families, concentration levels, and even individual perfumes. Combined with machine learning, e-nose systems can now match human expert assessors on blind classification tasks.

How it informs our scoring

E-nose benchmarks give us confidence that AI systems trained on molecular and linguistic odour data can produce reliable similarity scores — the science supports the approach, even if we're using note-pyramid data rather than mass spectrometry.

Our AI scoring is informed by published olfactory research. None of these organisations are affiliated with PerfumeCloneFinder — we reference their work for credibility and context.

How We Score → · Fragrance Families →

See the scoring in practice: Oud fragrances, Vanilla fragrances, Amber fragrances, or browse all designer originals and the best-value clones.

See this research applied — run an AI side-by-side comparison

Compare Originals vs Clones →