4. Discussion
4.1 Answering the three questions
Q1 — Coverage. The four-level chain is complete for 5 of the 52 molecules, and for all five of them each level rests on only one to three small studies. There is no molecule about which we know enough to say: this substance, at this concentration, via this receptor, produces this pattern of brain activity, leading to this emotional state, with this effect size. The molecule–emotion map that the applied industry assumes to exist does not exist.
Q2 — Reliability. Where data exist, quality is systematically low rather than randomly so. Sample sizes are in single digits to low double digits at the neural level. There are no confidence intervals at the emotional level for single molecules by inhalation. There is no molecule-level meta-analysis. And twenty popular claims do not match the sources they cite.
Q3 — Levers. This is the constructive part. The molecule is not the main control variable, but there are other variables that are far better quantified and have far larger effects.
4.2 Why the molecule is not the control variable
Three independent results converge on the same conclusion, and because they come from three different levels they cannot be explained by a common methodological error.
At the perceptual level, chemical structural features do not predict activity patterns in the human olfactory cortex, whereas perceptual features do. In other words, the brain does not represent molecules; it represents learned odour objects.
At the valence level, about 54% of the variance in pleasantness is individual difference and 37% is familiarity, whereas molecular identity accounts for a share that varies with the odour set, and culture accounts for only 6–7%. A molecule does not carry its emotional value with it; that value is assigned by each person’s history of exposure.
At the context level, the same chemical under two labels produces two different brain responses, and in expectancy-controlled trials the instructions determined the direction of physiological change regardless of whether the odour was real or a placebo.
Taken together: the decisive variable is not in the bottle. It lies in the relationship between the person smelling and the substance — a relationship built from experience, context and labels. This is not a pessimistic conclusion; it points to exactly where investment is needed.
One more thing needs to be said about the question of “emotional control”. The data above show that the idea of influencing people without their knowledge has no technical basis, for three independent reasons: effects below the threshold of awareness are small and unstable across individuals; they depend on learned associations and so do not transfer between people with different histories; and they disappear or reverse in sign when the experimental design controls for expectancy. What exists and can be measured is a cooperative relationship: users know what they are smelling, have their own associations with it, and the scent works with those associations.
4.3 What can actually be controlled
| Lever | Evidence | Data | How to use it |
|---|---|---|---|
| Learned association between odour and context | Strong | Familiarity explains 37% of the variance in pleasantness — more than molecular identity in some odour sets, and more than five times the share of culture. | Pair a fixed scent with a repeated positive experience. This is the strongest and cheapest lever, and it lies in experience design rather than in the formula. |
| Labels, context and expectancy | Strong | The same chemical under two labels produces two different brain responses in the mOFC and amygdala. In expectancy-controlled trials, the instructions determined the direction of physiological change regardless of whether the odour was real or a placebo. | Product name, story, packaging and context of use are not the outer shell of the effect — they are part of the mechanism. But this also means the effect does not hold when the context changes. |
| Intensity relative to the smeller’s threshold | Strong | The concentration–emotion relationship expressed in steps above individual threshold gives |ρ| = 0.79–0.95; expressed in physical concentration, only 0.51–0.90. Individual thresholds in a small sample spanned up to 38 dilution steps. | Dosage should be set not as a percentage of the formula but by the distance to the threshold of the target user group. This requires measuring thresholds in that very group. |
| Intensity ratio between components | Strong | Mixture valence is an intensity-weighted average, R² = 0.87–0.95. An arithmetic mean reaches only R² = 0.14. | For a note to dominate perception, it must dominate in intensity, not in mass. And a strong unpleasant note will drag the whole mixture down regardless of the pleasant notes around it. |
| Keeping the number of notes within the perceptual limit | Strong | Three-component limit; olfactory white convergence from 20–30 widely spanning components. | Complex formulas may be necessary for longevity, tenacity and structure over time, but they do not add perceptual layers. If the goal is a recognisable identity, design around three lead notes. |
| Temporal structure and adaptation | Moderate | Intensity levels off at 25–40% after ~4 minutes and then stops falling; recovery can take hours; the trigeminal component does not adapt. | Top notes do not need a high concentration to make an impression, because the impression forms in the first few seconds. But do not rely on the assumption that the odour will disappear; it does not. |
| Genetic differences within the user group | Moderate | β-ionone: ~49% of East Asians have low sensitivity. trans-anethole: a variant with 57% frequency in East Asia increases both intensity and pleasantness. But receptor genotype as a whole explains a median of only 4.77% of the variance. | For a few specific ingredients the difference is large and worth taking into account. For most ingredients it is negligible compared with experience-driven individual differences. |
| Choosing molecules by published pharmacological effects | No basis | Not a single single-molecule inhalation study reported a confidence interval. No molecule-level meta-analysis exists. The pooled effect of aromatherapy shrinks to −0.84 STAI points when small trials are excluded. | There is no basis for choosing ingredients on the strength of claims of specific physiological effects. The only exception with solid mechanistic evidence is cedrol, and its mechanism is pulmonary absorption, not olfaction. |
The order in the table above reverses the industry’s usual order of priorities. The three strongest levers — learned association, context and labels, relative intensity — are none of them decisions about ingredients. The weakest lever, choosing molecules by published pharmacological effects, is the one that receives the most publicity.
4.4 Implications for formula design
Five principles follow directly from the quantitative rules in section 3.6, not from inference:
- Design around three recognisable notes. The perceptual limit is three components, and training does not raise it. Other ingredients serve longevity, tenacity, temporal structure and roundness — they do not add perceptual layers.
- Balance by intensity, not by mass. Mixture valence is an intensity-weighted average. Because detection thresholds span more than four orders of magnitude across ingredients, percentages in a formula say almost nothing about proportions in perception.
- Set dosage by distance to threshold, and measure thresholds in the real user group. The variable that predicts emotion is steps above individual threshold, not physical concentration.
- Avoid the convergence zone. If a formula has more than about twenty equal-intensity components spanning the odour space widely, it is moving towards being indistinguishable from other formulas of the same kind. Keeping components within a narrow cluster of odour space is how identity is preserved.
- Take the trigeminal component into account deliberately. Cooling, warming and spicy sensations do not adapt as odour does, and they are rated very highly for pleasantness. This is a separate channel, often wrongly lumped in with the “odour effect”, and it holds up better over exposure time.
As for choosing ingredients by claimed effects: the only basis still standing after filtering is the cultural and personal associations of that ingredient within the target user group. That is a market research question that can be answered by measurement, not a chemistry question.
4.5 A research programme for the Vietnamese market
The Vietnamese data gap is one that can be filled at moderate cost, and filling it creates an asset no one else has. Four studies, ranked by value-to-cost ratio:
- Valence and familiarity norms for Vietnamese people. The odour set of the VSIT identification test has been standardised on 391 people and already includes the important local odours. Adding pleasantness and familiarity scales to that same set is the cheapest possible extension. Extend it with a further 30–40 commercial fragrance ingredients. About 400 people per age and region subgroup are needed to estimate means with an error of ±0.1 standard deviations.
- Measuring detection thresholds in the target user group for 10–15 key ingredients. These are the data that determine dosage, and they currently do not exist at all for the Vietnamese population. A standard two-fold dilution series method with n of about 60–100 is enough to estimate the distribution.
- Building an odour-evoked emotion scale specific to Vietnamese. Do not translate European scales: the Singapore version showed that there are emotional dimensions that do not exist in the European structure, and each culture’s structure fits its own data better. The standard procedure is to collect descriptive terms, run exploratory factor analysis, then confirm on an independent sample.
- A two-factor randomised trial: expectancy × scent, with a control odour matched for pleasantness. This is the only design that can separate the effect of the substance from the effect of expectancy, and its results apply directly to the design of product communication.
The first three studies require no hypothesis about pharmacological effects and no expensive equipment, and they produce data that can be used immediately for product decisions. The fourth answers the question the whole industry still avoids.
4.6 Limitations
Molecules were not selected according to a pre-registered search strategy, so the selection is biased towards well-studied substances and popular commercial ingredients; the absence of a molecule from this database does not prove that no data on it exist. Many key sources were accessible only as abstracts, including some foundational papers on binary mixtures and on thresholds; values taken from secondary sources are all flagged. CAS numbers have so far been cross-checked for only 24 of the 52 entries; of the 24 numbers checked, one (4%) was wrong — androstadienone had been mistakenly assigned the registry number of another steroid — so the unchecked numbers in this database should be treated as not yet reliable. Extraction was carried out by a language-model-assisted process in four independent streams with cross-checking, but without two independent human reviewers. And finally, the high proportion of negative data in this field means the picture may be even more pessimistic than this report presents, because negative results are less often published.
5. Conclusion
Going down to the molecular level does not give the fragrance industry control over emotion, but it gives something more useful: knowing exactly what does not work and what does.
What does not work is the “choose a molecule to achieve an emotional state” model. The data chain that model requires is complete for 5 of the 52 molecules and is not robust enough for any substance. The brain represents not molecules but learned odour objects; more than half of the emotional variance is individual difference; and labels together with context can change even the direction of the physiological response.
What does work is three levers with large, measurable effects: learned association between a fixed scent and a repeated experience; the context and labels in which the product is placed; and relative intensity with respect to the smeller’s threshold. Alongside these is a set of psychophysical rules robust enough to serve as engineering principles: the three-note limit, the intensity-weighted average, olfactory white convergence, and adaptation kinetics.
For a fragrance business in Vietnam, the most actionable conclusion lies not in chemistry but in data: no one has valence, threshold or odour-emotion structure data for Vietnamese people. The first three studies in section 4.5 fill precisely that gap, require no special equipment, and create an advantage that competitors cannot buy from the international literature.
Scientific reference material, not medical advice and not promotional material. All assessments of evidence quality are the compilers’ own. Figures are redrawn from published data.