Free is a map. Paid is the reading.

The Scent Coach is an AI recommendation system built specifically for fragrance discovery. Two technical components make it work together.

A frontier large language model. 'Frontier' here means a state-of-the-art model — the same class of systems that power the most capable AI tools available today. It reads your answers, interprets intent, picks up on subtle cues, and reasons about which fragrances would fit. Unlike a rules-based quiz, it handles ambiguity and open-ended answers. Unlike a keyword-match system, it understands context.

A structured olfactory graph. A proprietary database of 26,000+ perfumes from 1,421 brands, each classified across a dozen dimensions: top / heart / base notes, olfactory accords, fragrance family, subfamily, sillage, longevity, seasonality, occasion, gender expression, price tier, and retailer availability. Think of it as a map of the fragrance world, where every bottle is placed in relation to every other through shared attributes.

The reasoning engine interprets what you want. The olfactory graph grounds the recommendations in real, structured fragrance data. Without the graph, the LLM would hallucinate notes that aren't in the bottle. Without the LLM, the graph would return mechanical lookups instead of nuanced picks. The combination is what makes the Scent Coach work.

Here is what happens in the three minutes or so between your first click and your shortlist appearing on screen.

Step 1 — Conversational intake. Ten questions organised as ten chapters. No fragrance jargon. Questions probe different axes: olfactory memory (a smell that reminds you of someone, a place), aspirational identity (how you want to be perceived), context (professional, social, intimate), sensory preference (how much you want to be noticed by sillage), and directional pull (which families attract or repel you even before you know their names).

Step 2 — Profile inference. The LLM reads your ten answers holistically and infers a structured olfactory profile — not a single label like 'woody masculine' but a weighted distribution across families, sub-families, accord preferences, sillage tolerance, seasonal fit, and occasion coverage. Importantly, the model also infers what you probably want to avoid, which is often as informative as what you want.

Step 3 — Graph query and scoring. Your inferred profile becomes a query against the olfactory graph. Every perfume in the database is scored on how well it matches across the relevant dimensions. Scoring is multi-dimensional — a perfume can be a near-perfect family match but fail on sillage or seasonality, and the overall score reflects that. Hundreds of candidates survive initial filtering, then get ranked.

Step 4 — Shortlist selection with diversity constraints. The free Scent Coach returns six matched bottles; the paid Premium Report (in Monthly and Annual) narrows them to three hand-picked picks with deeper context. In either case, the top picks are not just 'the highest-scoring'. We apply diversity constraints so that your shortlist represents real alternatives: usually one safe anchor pick, adjacent-family variants, and at least one slightly bolder option. Near-identical picks would be a worse answer than genuinely different directions.

Step 5 — Reasoning generation. The LLM writes a human-readable explanation for each pick. Why this one fits your profile, which of your answers it responds to, what you can expect from it in terms of dry-down, longevity and sillage. This step is what makes the Scent Coach different from a typical recommendation engine — you don't just get a list, you get the reasoning.

Step 6 — Live retailer comparison. For each pick, we query current prices across major European retailers — rolling out as our affiliate partners come online — and display them side by side. You go to wherever it's cheapest. We earn an affiliate commission regardless of which retailer you choose, so our recommendations are never skewed toward a higher-margin retailer.

The olfactory graph is the backbone of the Scent Coach. How it's built, maintained, and updated matters — so here is the full picture.

Source data. The initial dataset is assembled from public fragrance databases and brand-direct information (note pyramids, official descriptions, fragrance family classifications), augmented with community data where available. We are building direct relationships with niche and indie houses for direct listing updates.

Classification. Each perfume is classified across roughly a dozen structured attributes: top / heart / base notes, olfactory family and sub-family, sillage rating, longevity rating, seasonality fit, occasion suitability, gender expression (where declared), and price tier. Classifications come from a combination of brand-declared data, structured community consensus, and our own editorial review for niche and indie pieces where public data is thin.

Retailer availability. For each perfume we track which European retailers currently carry it. For perfumes where retailer partnerships are live, stock status and price are queried in near-real-time. The retailer network is rolling out across major European storefronts.

Update cadence. The database is refreshed continuously. New releases are added within days of launch. Existing classifications are periodically reviewed against new community consensus. Retailer pricing is near-live. We built the system so that 'stale data' is not a problem we live with.

What we don't do. We don't sell our database. We don't train third-party models on your quiz answers. We don't promote perfumes in the graph because a brand paid us — classification is editorial, not commercial.

We are a European company, built to European standards. Here is what we do and don't do with your data.

A free account is required to take the quiz — we use it to save your profile, build a collection, let you come back for fresh readings, and give better recommendations over time. You can delete your account and all associated data at any time.

We are direct about this because we think you deserve to know.

Three ways to use the Scent Coach. Same engine across all three, different depth and commitment.

The free Scent Coach is enough for many people to find direction — six carefully matched perfumes based on your ten quiz answers, with the option to take a fresh quiz every three days. Monthly is for the deeper monthly reading: a Premium Report with three hand-picked bottles, match scores, wear guidance, and honest guardrails on what to skip. Annual is for enthusiasts who want an evolving reading every month, locked in at the lower yearly rate.

The European Union's AI Act comes into full effect on 2 August 2026. Parfuses was designed with its transparency principles in mind from day one.

Concretely: we tell you when you're interacting with AI. We explain our reasoning. We document how our system works (this page is part of that documentation). We don't pretend to be human, and we don't pretend to be infallible.

If you ever want to know more — about the data we use, how our models are trained, what the system's known limitations are, or how to request your data — write to us at hello@parfuses.com. We will answer.