A vanilla extract can have a plausible vanillin isotope result and still need a separate check for added flavor compounds. That is not a contradiction. A test of the vanillin molecule and a test for ethyl vanillin answer different questions.
A 2025 study in Molecules makes the distinction unusually concrete. Its researchers examined ten purchased vanilla samples, compared their volatile compounds and carbon and hydrogen isotope ratios, and separately added synthetic ethyl vanillin to a commercial vanilla extract. The work found useful differences between the two vanilla species studied, but it did not achieve complete geographical discrimination.[1] For a buyer reviewing a laboratory report, its value lies less in a promise of tracing every bean to a country than in showing what each analytical result actually covers.
Ten samples are informative, but they are not a global reference library
The sample set comprised four entries from Madagascar, two from Réunion and one from Mexico, all identified as V. planifolia, plus two from French Polynesia and one from Papua New Guinea identified as V. tahitensis.[1] These were ten purchased sample entries—not ten countries, ten documented farms or a multi-year validation population. The methods describe origins and quality labels but do not establish a farm-authenticated, harvest-stratified reference collection.[1]
Even the physical materials were not entirely uniform. Table 5 describes MAD3 as a glass-tube product containing syrup sugar, pods at 300 g/L and vanilla seeds, whereas other entries give pod lengths and quality descriptions.[1] That matters when interpreting differences within Madagascar: the table does not justify assuming that every entry began as an equivalent dry bean. Nor does it establish that syrup caused any particular analytical difference.
The laboratory prepared extracts using equal volumes of water and ethanol, 15 g of sample per 100 mL, and 72 hours of maceration with stirring, followed by further preparation for chromatography.[1] Results obtained through that workflow should not be silently transferred to a finished beverage, a carrier-containing powder or a sugar-rich paste without checking the laboratory’s validation for the relevant matrix.
A large vanillin peak is not the same as a vanillin assay by bean weight
The study reported vanillin at 35.1–81.9% of the measured volatile-compound profile.[1] Those are relative abundances within the VOC analysis, not a claim that a bean is 35–82% vanillin. The supplementary-table description explicitly identifies the denominator as total VOCs.[1]
This distinction changes how a buyer reads a certificate. A relative chromatographic profile may help compare patterns, but it does not by itself supply the mass concentration needed for an ingredient specification. Ask whether a reported percentage is a share of a measured profile or a mass-based assay before comparing suppliers.
The authors observed clearer species separation than geographical separation in their VOC principal-component analysis, or PCA.[1] A PCA plot is a way to display patterns in the measured samples; visible clusters are not a validated acceptance rule for future shipments.
The isotope results separated these species more clearly than these origins
For vanillin, the seven V. planifolia entries had carbon isotope values, δ13C, from −20.5 to −19.1‰; the three V. tahitensis entries ranged from −16.7 to −16.4‰ in Table 4.[1] These are the observed ranges in this sample set, not universal cutoffs for accepting or rejecting a species declaration. The ‰ unit describes the relative difference between the sample’s heavy-to-light isotope ratio and a reference ratio: VPDB for carbon and VSMOW for hydrogen. It is not a concentration or a percentage of bean-derived material.[1]
Hydrogen isotope values added another dimension, but not a unique country signature. The French Polynesian entries had vanillin δ2H values of −79 and −82‰, compared with −99‰ for the single Papua New Guinea entry; the four Madagascar entries ranged from −80 to −64‰.[1] The comparison is interesting, but one Papua New Guinea entry cannot establish that country’s full range.
More importantly, the combined VOC/isotope PCA placed Madagascar and Réunion together, with the Mexican sample also in the same group as Madagascar.[1] The authors suggest similar climate as a possible explanation, rather than demonstrating that these measurements can resolve national boundaries. Their conclusion is explicit: VOCs and vanillin isotope ratios could not completely discriminate the samples by geographical origin.[1]
There is also a design limitation that a neat plot can hide. Each country or region in this set contributes only one species, although each species appears in more than one origin. Species and origin are therefore not fully crossed in the design: separation between the two species cannot establish geographical discrimination. A stronger follow-up would expand sampling within each species across origins, farms and harvests and, where feasible, include both species from the same origins, with independently documented provenance.
Analytical validation is not origin-classification validation
The paper did real analytical-method work: it assessed within-run and between-run precision, investigated isotope fractionation during solid-phase extraction, and studied concentration effects.[1] Its within-run precision experiment used 15 runs of a standard solution—not 15 additional independent beans or harvests.[1]
That distinction should remain visible in a laboratory’s claims. Precision asks how consistently an analytical procedure measures a sample. Origin-classification validation asks how often a decision rule correctly identifies genuinely new samples from the relevant population.
The reported statistical work consists of exploratory PCA and species comparisons; the methods’ statistics section names R 4.4.0 but does not describe a held-out test set, cross-validation, an external sample cohort or classification accuracy.[1] There is consequently no defensible percentage here for “country-of-origin identification accuracy.”
It would also be wrong to accuse the authors of demonstrated train/test leakage: no such train/test scheme is reported. The relevant risk arises if someone later presents separation of these same exploratory samples as predictive validation. If several measurements or extracts from one lot are used in a future model, keeping that lot entirely on one side of a validation split would be a prudent design requirement, not something this paper demonstrates having done.
Added ethyl vanillin and added vanillin are different problems
The researchers did not detect ethyl vanillin in the analyzed vanilla extracts. They also investigated a proposed glycosylated precursor using high-resolution mass spectrometry and a targeted PRM experiment; none of the ten in-house extracts showed that precursor.[1] The latter is useful because the authors encountered isomeric peaks and used additional fragmentation evidence to avoid false positives.[1]
These are bounded negative findings. They do not establish absence in every vanilla plant or provide a universal detection threshold. For a sample represented as an unmodified bean extract, a confirmed ethyl-vanillin result would warrant investigation. For a product explicitly formulated and described with added flavoring, the interpretation must instead start with its actual declaration—not an automatic accusation of fraud.
The separate spiking experiment added ethyl vanillin, not non-bean vanillin, to a commercial vanilla extract.[1] Under adequate chromatographic separation, the authors concluded that ethyl vanillin did not interfere with the isotope ratios measured for vanillin.[1] This is precisely why a plausible vanillin isotope result cannot, by itself, exclude the presence of ethyl vanillin elsewhere in that extract.
Concentration mattered. The discussion reports stable ethyl-vanillin δ13C values over 5–144 mg/L within ±0.2‰, but unreliable δ2H determination below 21 mg/L and synchronized trends at 72 mg/L and above that the authors associate with a possible instrument memory effect.[1] The conclusion summarizes a hydrogen interval of 21–72 mg/L, while the discussion flags synchronized trends at 72 mg/L and above; 72 mg/L should therefore not be presented as an unqualified reliable upper limit.[1] These mg/L values refer to ethyl vanillin in the reported analytical concentration series, not its mass fraction in beans or an addition rate for a finished product. The methods describe dilution and SPE preparation; the reported series should not be converted into an original-product concentration without a verified preparation and recovery basis. These are method-specific analytical observations, not legal addition limits or universal adulteration-detection limits.
Added non-bean vanillin raises a different source-attribution question. The paper’s introduction discusses carbon-isotope overlap and mixtures that can mimic a natural-vanillin δ13C value, drawing on earlier literature; the ten-sample experiment is not a fresh validation of every such mixture.[1] Its ethyl-vanillin findings therefore should not be advertised as proof that any addition of vanillin can be detected or excluded.
What a useful laboratory conclusion should say
A report becomes more useful when the conclusion names the question it resolved. “Ethyl vanillin detected” identifies a compound. “Vanillin isotope result consistent with the laboratory’s reference population” states a bounded comparison. Neither statement alone establishes the country where the beans grew.
For a geographical-origin claim, ask what authenticated reference material supports the comparison, which species and harvests it covers, how the laboratory handles overlapping origins, and what performance it has measured on genuinely independent samples. For an added-flavor question, ask which compounds were targeted, whether identification was confirmed, and whether the sample preparation and concentration were appropriate for the actual material.
The study’s three proposed minor markers—homovanillic acid, 4-ethoxymethyl-phenol and vanillyl alcohol—are promising research directions, not a ready-made origin guarantee. They were selected for sufficient abundance and chromatographic separation, and the authors explicitly call for more samples to confirm and refine the preliminary findings.[1]
The practical lesson is to keep compound identification, molecular-source evidence and geographical attribution separate. Buying the right vanilla ingredient is one decision; determining how much a laboratory result proves about that ingredient is another.
Sources
[1] Authentic Aroma and Compound-Specific Isotope Ratios: full-text XML, DOI 10.3390/molecules30040825