Introduction

The evolution of wine consumption trends is usually addressed independently of wine sensory features, such as packaging, production methods, e-commerce or market development (Bousquet, 2023). In parallel, structural changes in global viticulture worldwide have led to a progressive reduction in grape variety diversity, as vineyards increasingly converge on a limited number of internationally recognised cultivars (Anderson & Pinilla, 2021). The recovery of minority grape varieties therefore represents an opportunity to test whether consumers share the sensory orientations associated with these evolving market dynamics, particularly focusing on intrinsic properties.

Reports on the agronomic and winemaking potential of such varieties have been published over the last two decades (Martínez-Pinilla et al., 2013; Balda et al., 2014). The widespread cultivation of international and market-friendly varieties, such as Cabernet-Sauvignon, Merlot, or Chardonnay, has been primarily driven by their strong name recognition and commercial reliability, rather than by intrinsic oenological quality alone. They have been planted all over the world, mainly with the aim of obtaining commercially successful wines, even in countries with a pre-existing rich diversity of autochthonous grapes (Balda et al., 2014) or whose climatic conditions are not suitable for proper ripening (Jones, 2012). These developments have led to a growing interest in conserving the somewhat neglected ancestral varieties to counteract the homogenisation driven by the world market (García-Muñoz et al., 2011).

An increasing number of reports describing the sensory features of ancestral varieties has been published; these reports originate from traditional wine countries such as Spain (Sánchez-Palomo et al., 2018), Italy (Vercesi et al., 2024), or Greece (Miliordos et al., 2021). Moreover, South American countries are also recovering their original grape varieties (Meneses et al., 2024), while in Australia, using minority varieties is seen as a potential means of coping with climatic change (Mezei et al., 2021); indeed, old grape varieties have been found to adapt well to climate (van Leeuwen & Destrac-Irvine, 2017). In addition, grape varieties with longer vegetative cycles allow more balanced wines to be obtained than those originating from overripe grapes (Augusto et al., 2019; Vercesi et al., 2024), which have been clearly distinguished by untrained consumers from their cooler climate counterparts (Loureiro et al., 2016; Coste et al., 2018; Souza-Coutinho et al., 2020).

Portugal has a high diversity of grape varieties, with 34 main grape varieties (18 white and 16 red) covering about 87 % of the total vineyard area (Brazão et al., 2023). Several reports have addressed the genetic or agronomic characteristics of the less cultivated Portuguese varieties (Alifragkis et al., 2015; Barrias et al., 2023; Sousa et al., 2024), but their sensory profile has been relatively little studied (Piras et al., 2020; Brazão et al., 2023). The study of these varieties therefore offers a valuable framework to explore both sensory diversity and consumer response to non-dominant wine styles, while contributing to the preservation of viticultural biodiversity.

Consumer wine preferences have been widely studied, being dependent on both intrinsic (i.e., sensory features) and extrinsic (e.g., label, mode of production, origin, among others) factors (Francis & Williamson, 2015). Studies have established that consumers are not a homogenous group, their preferences being better understood under the concept of segmentation (Pickering et al., 2014). This concept is based on grouping individuals according to common features that modulate product appreciation. In the field of wine appreciation, consumer segmentation has been achieved through the utilisation of demographic data, including age, gender, and geographical location, as well as lifestyle characteristics, personality traits, cultural background, and/or involvement with wine, among others (Johnson & Bastian, 2015; Pickering & Hayes, 2017; Sena-Esteves et al., 2018; Rodrigues & Parr, 2019; Mora et al., 2019).

Within this framework, consumer involvement has emerged as a key factor influencing wine appreciation. As demonstrated in the existing literature, the experience of wine consumption is influenced by sensory perception, emotional state, and cognitive processes. However, the extent to which these factors contribute to the overall experience is contingent on the individual’s level of involvement with wine. Low-involved consumers tend to prioritise immediate sensory and emotional responses, while wine experts tend towards cognitive evaluation. Highly involved consumers are positioned between these two categories, utilising all three dimensions (Oyinseye et al., 2022). These studies are based on experiments that considered both the extrinsic and intrinsic cues of wine, but less work is focused on understanding the role of the consumer’s involvement in wine acceptance that is based exclusively on the intrinsic, and thus sensory, properties of wine.

Hopfer and Heymann (2014)’s study identified four distinct American consumer segments based on their preference for Cabernet-Sauvignon wines. However, the researchers did not find a common trait within these consumer clusters that could explain the differences in liking based on demographic information (i.e., age, gender, and income), wine consumption, or wine expertise. It is interesting to note that these segments differed significantly from the scores given by the experts, who expressed a preference for fruity attributes (e.g., red fruit, dark fruit) and oak, as well as sweet aromas. Conversely, they penalised marginally detectable vegetal-green, chemical, earthy, sulfur, and animal aromas. These results align with the findings of Sáenz-Navajas et al. (2015), who observed significant discrepancies in quality assessments provided by wine experts and consumers. Both groups exhibited a similar propensity to penalise sensory traits as reported by Hopfer and Heymann (2014). In relation to the attributes that enhance perceived quality, wine experts exhibited a higher valuation of fruity wines, while consumers demonstrated a preference for oaky red wines.

Despite these advances, the empirical evidence for how the level of involvement of non-expert consumers influences intrinsic wine preference is still limited. Therefore, the main purpose of this work was to evaluate the consumer acceptance of wines produced with Portuguese minority varieties and to identify their main sensory drivers of preference. Minority grape varieties are used here in a case study to investigate how consumer involvement modulates hedonic responses to wines differing in sensory style when extrinsic cues have been removed. It was hypothesised that the consumers’ level of wine involvement influences their preference patterns based on intrinsic sensory properties, and that specific consumer segments may show appreciation for sensory profiles associated with minority grape varieties, even if the majority still prefers the dominant commercial styles.

Materials and methods

1. Wine samples

Twenty-four Portuguese wines were selected, 12 of which were white and 12 red. The wine selection was performed by the research team in collaboration with producers, aiming to maximise sensory diversity while ensuring the inclusion of Portuguese minority grape varieties from different regions. The Blend_T_21 and Blend_D_19 samples were made from different grape varieties representing traditional Portuguese wines, while the remaining 22 were single varietals of minor varieties, each representing less than 2 % of the total Portuguese vineyard area (Table S1). Most of the samples were young wines in their first to third year (Table 1), that had been generously provided by the producers. The different wines were produced using simple winemaking methods. According to the winemakers, wines were designed to highlight the unique characteristics of each grape variety in its region.

Given the exploratory nature of this study and the availability of minority-variety wines, samples from different vintages were included. This is acknowledged as a limitation, particularly for the red commercial reference wine (2019 vintage) compared to the minority-variety wines (mainly 2021–2022), and is discussed in the “Discussion” section.

2. Wine sensory description by a panel of wine experts

2.1. Participants

For the sensory characterisation carried out using the PM methodology, 21 expert tasters were selected (4 females and 17 males). The group was composed of wine professionals (oenologists, journalists, wine critics) aged between 25 and 65 (the average age being 45 years old) and with more than five years of professional experience. They had all signed a declaration in accordance with the Declaration of Helsinki, and the faculty Ethic Committee had provided a formal acceptance of the research (report 2/2025). Accordingly, the tasters were informed at the start of the experiment that they would remain anonymous and that the data they provided would only be reported in aggregate. They had to sign a consent form in order to take part. Participants could request more detailed information. They could withdraw from the research if they requested access to their personal data, their rectification and deletion, restriction of their processing and portability, or any other right, and they could withdraw their consent at any time by contacting the experimenters. To take part in the study, they had to be over 18, free from certain health problems, not drink alcohol, not be pregnant and have no involvement with the research. They were not paid for their participation.

2.2. Procedure

The tasting sessions were held on two separate days (3 and 17 June 2024). Each participant attended two sessions on the same day, which were separated by a 30-minute break. The first session comprised the evaluation of white wine samples, and the second the evaluation of red wine samples. The wines were placed on the table in front of each expert in a different, randomised order following a Williams squares design created by RedJade software (version 6.2.3).

Each session consisted of two parts. In the first part, participants were asked to distribute the samples on a white sheet of paper according to their similarities and dissimilarities following the PM strategy (Valentin et al., 2018). All the wines had to be placed on the 60 × 40 cm white sheet in such a way that two wines were placed close together if they were perceived as similar, and further apart if they were perceived as different. In the second part, once the wines had been arranged, they were asked to write a maximum of three descriptors that characterised each sample on the sheet next to each wine (i.e., UFP).

The tasting took place in a laboratory of the Instituto Superior de Agronomia, at 22 °C and, with adequate lighting and ventilation. Thirty mL of wine was served in black Schott Zwiesel Sensus glasses, each marked with a unique three-digit code. The white wines were served at a temperature of 12 °C and the red wines at 18 °C. After serving, plastic Petri dishes were used to cover the glasses. Each professional evaluated each wine and was given water and unsalted crackers to cleanse the palate if necessary.

2.3. Data analysis

For each participant, the position of the samples on the sheet was measured. Therefore, the x and y coordinates (the bottom left part of the sheet representing the origin (0,0)) for each sample and participant were registered. A final wine × assessor matrix was obtained and submitted to Generalised Procrustes Analysis (GPA) (Findlay, 2023) to obtain a two-dimensional consensual map of samples. Further, the consensual coordinates derived from GPA were submitted to Agglomerative Hierarchical Cluster Analysis (AHCA). For AHCA, Ward’s linkage method and Euclidean distances were used. The number of clusters was determined by inspecting the dendrogram and selecting the cut-off at the level showing the largest increase in linkage distance between successive agglomeration steps (i.e., the largest vertical jump in the dendrogram), which indicates a marked decrease in within-cluster similarity when additional merges are forced.

The method described by Ballester et al. (2024) was utilised to process the data derived from UF; therefore a multi-stage approach was adopted. To prevent redundancy, the initial step involved compiling a list of every attribute that the tasters had mentioned and eliminating any duplicate terms. The next step was lemmatisation, which consisted of grouping terms with the same root into a single representative term (for example, “complexity” and “complex” were combined to form “complex”). Then, three researchers separately conducted a triangulation process on the data by grouping terms belonging to the same semantic universe. After comparing the groups formed by the experimenters, a final, consolidated list of terms was determined by consensus.

Finally, a matrix containing the frequency of citations for each attribute was constructed, with the consensual terms in columns and wines in rows. For each attribute, the average of the citation frequency of between the wines belonging to the same cluster derived from the GPA-AHCA was calculated to obtain a raw description of the clusters. To identify any differences between the four clusters in the citation of attributes, chi-square was calculated, and for significant attributes (p < 0.1) the Marascuilo post-hoc test was applied.

The responses were compiled for the 12 white and 12 red wines separately. The final list of descriptors contained 32 terms for white wines and 25 terms for red wines. Attributes cited by at least 10 % of the panellists were considered, an arbitrary threshold used to ensure a minimum level of agreement.

3. Consumer liking and wine characterisation

3.1. Participant selection

Participants were recruited through emailed invitations and social media. Recruitment was limited to participants who lived and had spent most of their lives in Lisbon to ensure consistency in terms of the region of origin effect and familiarity. They completed an online questionnaire on demographic information and their level of involvement (10 items) in wine consumption; inspired by the questionnaire developed by Bruwer and Buller (2013), our questionnaire was adapted to Portuguese consumers, taking into account the results of previous work involving Spanish consumers (Oyinseye et al., 2022) (see Table S2). Participants were asked to rate their agreement with statements using a 5-point Likert scale (1 = strongly disagree; 2 = disagree; 3 = neither agree nor disagree; 4 = agree; 5 = strongly agree). The survey was conducted for a period of two weeks.

A final number of 84 consumers (out of the 95 that performed the demographic and involvement questionnaire) qualified for and participated in the sensory sessions (Table S3). The requirements to participate were: (i) drink wine at least once a month; (ii) reside in the Lisbon metropolitan area for a minimum of 10 years; and (iii) not work in the wine industry.

Regarding the expert panel, all its members had been informed about the objective of the study and had signed a consent form in accordance with the Declaration of Helsinki.

At recruitment, consumers and expert panel participants were informed that they would take part in a sensory study involving wine tasting. At the beginning of the sessions, they were not informed about the nature of the samples (minority vs commercial wines) nor about the specific hypotheses tested, in order to minimise expectation bias.

3.2. Procedure

The tasting sessions took place on different days. Each participant attended two sessions (each lasting a maximum of 30 minutes) on the same day with a 30-minute break between sessions.

The first session was devoted to the evaluation of four white wines and the second to four red wines. Wines were selected based on the results from the PM reported in a previous selection and aiming at having wines with maximal sensory variability. Specifically, one wine located near the centroid of each expert-derived sensory cluster was selected in order to represent the typical sensory profile of each cluster.

In each session, the wines were randomised according to a Williams square design. The tastings took place at the Instituto Superior de Agronomia, in a room at a temperature of 22 °C and with adequate lighting and ventilation. Twenty-five mL of each wine was served in transparent glasses (Schott Zwiesel, model Ivento, Bavaria, Germany) that were coded with three-digit numbers. White wines were poured at 10 °C and red wines at 16 °C. The glasses were covered with plastic Petri dishes. Each taster evaluated each wine individually, and they were provided with water and unsalted crackers to cleanse the palate when necessary. The participants were not informed of the nature of the samples or the purpose of the study.

For each wine, participants were asked to first rate wine liking on a structured 10-point scale (0 = ‘I don’t like it at all’; 5 = ‘I don’t like it or I dislike it’; 10 = ‘I like it very much’) based on visual, olfactory and gustatory appreciation. A 10-point scale was chosen to provide sufficient discrimination for segmentation analyses while maintaining simplicity for consumers. Then, sensory profiling was undertaken following the RATA methodology (Danner et al., 2017; Nguyen et al., 2020).

Participants had to select from a list of 10 attributes referring to flavour (for white wines: citrus/tropical fruit, dried fruit, floral, herbaceous, neutral, oxidation, reduction, yellow/white fruit, woody/spicy; and for reds: animal, defective, dried fruit, fresh fruit, neutral, oxidation, red/black fruit, woody/spicy), taste (for whites: acid), mouthfeel (for reds: astringency), and multidimensional (for reds: complexity) attributes. The assessed sensory attributes had been extracted from the description of the initial 24 wines described in the previous session by the panel of wine experts. The attributes were selected based on their frequency of citation and their relevance to differentiate sensory clusters, while keeping the list short and understandable for consumers. To facilitate the interpretation of potentially ambiguous technical terms, brief associated descriptors were provided in parentheses for the purposes of comprehension and their consistent use, especially by low-involved consumers; for instance, “boiled potato, honey, overripe apple” for oxidation and “rotten eggs, sewer, garbage” for reduction. Each consumer was required to only rate the intensity of attributes they perceived in the wine (i.e., only those that apply) on a 7-point scale, where 1 = ‘extremely low’ and 7 = ‘extremely high’.

3.3. Data analysis

Participant selection. A first Principal Component Analysis (PCA) was carried out on the participants’ answers to the involvement questionnaire to identify the questions with the greatest impact on consumer involvement. This analysis showed that 8 out of 10 questions were most strongly correlated (Table S2 of supplementary data). The standardised Cronbach’s α was calculated. This is a reliability coefficient that ranges from 0 to 1. Higher values of Cronbach’s α indicate higher internal consistency. Cronbach’s α values higher than 0.75 were considered acceptable, otherwise the non-related items were identified and removed and Cronbach’s α recalculated. An average level of involvement score was calculated with the selected eight items (Cronbach’s α > 0.75) by assigning scores of 1–5 to responses ranging from totally disagree: 1, to totally agree: 5.

To define two contrasting involvement groups, participants with scores below the 33rd percentile and above the 67th percentile were classified as low- and high-involved, respectively, while intermediate scores were excluded from the tasting sessions to minimise overlap between involvement levels (Oyinseye et al., 2022).

Consumer liking and wine characterisation. First, the preference scores of all the 8 wines evaluated and the 84 respondents were subjected to a PCA considering consumers as variables to determine the consensus among the responses. Then, an AHCA was run to group the wines according to the similarity of their preference scores.

To identify significant differences between wines in terms of their sensory descriptions, two-way ANOVA with all the scores of RATA attributes were calculated, considering the 84 consumers as random factors and the wine as fixed factors.

Consumer segmentation was carried out by calculating an AHCA on the preference scores. Then, two-way ANOVA was calculated using the preference scores and considering the clusters of consumers derived from the AHCA and the wine, and their interaction as fixed factors. In addition, a one-way ANOVA on the involvement scores and considering the cluster as a fixed factor was calculated to evaluate the effect of the level of involvement on consumer segments. For significant effects, Student–Newman–Keuls post-hoc pairwise comparison (95 %) test was calculated.

Finally, to relate sensory profile and preference scores, a PCA for each wine set was performed, considering the significant RATA attributes as active variables, and the averaged preference score (considering all the consumers together) as well as the average preference scores for each cluster as supplementary variables. These analyses were performed with the XLSTAT software (version 2023.1.1) and the free online statistical software Jamovi (version 2.3.28).

Results

1. Wine sensory description by a panel of wine experts

The results of the AHCA calculated on the data derived from the PM task showed the presence of four distinct clusters for both the white and red wines (Figure 1). The two-dimensional consensual configuration obtained by GPA is presented in Figure S1.

Overall, the white wine clusters significantly differed in their aroma intensity and fruity profile, covering a broad sensory space. The wines in Cluster III showed the most intense and the fruitiest profiles in contrast to the samples in Clusters IV and I (see Table S4 for the average of citation of terms for each cluster and their significance). The studied white wine set thus occupied a distinct sensory space, with profiles ranging from intense and fruity to neutral and subtle.

Cluster I contains the reference sample Blend_T_21, as well as the samples Ramis_PS_22, and Jampal_L_21. Even if there were no significant differences in terms of the attribute citations, most wines in this cluster displayed the highest frequency of citation for the terms “neutral” and “structured”. Cluster II brings together the wines Folga_D_22, Donzel_D_18, Vital_L_22, and Carac_M_23, showing the highest frequency of citation for the terms woody/spicy and yellow/white fruit. Cluster III contains the Sarig_A_22, Callum_BI_22, and FontCal_BI_22 wines. Their highest intensity and fruity character is related to the highest citation for yellow/white and citrus/tropical fruit. This cluster also showed higher citation frequency for acid taste and fine character. Finally, Cluster IV included the RabOve_A_23 and Moscat_D_22 wines, with higher frequency for the terms “neutral” and “reduced profile”, which could explain their significant least intense and fruity character.

For the reds, only the term “animal” showed significant differences between the clusters, which was significantly higher for the wines in Cluster I (Table S4). This group of wines contains the wines Moret_A_22 and Ramis_PS_20, which also showed a higher frequency of citation for neutral and oxidised attributes, indicating a less fruity sensory style that is associated with notes of ageing or evolution. Cluster II consists of the wines TintCao_D_16, Rufet_BI_21, TintFran_D_22, Bastar_D_22, and Maruf_D_21, which showed a higher frequency of citations for the woody/spicy and complex terms. Cluster III contains the wines Cascul_D_19, NegMol_Al_22, and TintCar_A_21 wines, which were defined by their red/black fruit descriptors, with the presence of fresh fruit, albeit neutral, and showing a younger profile. On the other hand, Cluster IV group together the reference sample, Blend_D_19 and TintMiu_A_21, which are characterised by the presence of wood and spices, combined with the descriptors of dried fruit and complexity, suggesting wines with more structure. These groups show a remarkable sensory diversity among the red wines analysed, due to differences in style, ageing, and aromatic profile.

The dendrogram analysis revealed multiple clusters, each representing wines with distinct sensory profiles (Figure 1). Within a cluster, the wines positioned closest to the average centre are those that most effectively capture the representative characteristics of the group. Therefore, selecting one of these central wines ensures the choice of a sample that best embodies the typical profile of its cluster. Accordingly, for the consumer test, the selected white wine samples were Blend_T_21, Donzel_D_18, Callum_BI_22, and RabOve_A_23, and the red wine samples were Blend_D_19, Moret_A_22, TintFran_D_22, and NegMol_Al_22.

2. Overall wine liking

Significant differences in wine liking were observed between the wines. Most wines were scored with values of 4 to 8 (on a 0-10 scale), indicating a neutral to positive liking (Figure S2). Computing all wines, the average scores were higher for white (5.09 ± 2.36) than for red wines (4.41 ± 2.39) (F = 13.779; p < 0.001), driven by the number of 0 scores for red wines (31 responses). These null scores were distributed among all wines, the highest number being given to NegMol_Al_22 by 12 consumers.

Furthermore, in both cases, the reference commercial wine (i.e., blend wine) received significantly higher preference scores compared to the minor varieties (Table 2). However, since one commercial reference wine was included per colour category, these results should be interpreted as comparisons with the selected reference style, rather than as a general statement about commercial wine preference.

3. Liking distribution among consumers

Figure 2a shows the projection of the preference scores of participants on the first two axes of the PCA. The first and second principal components (F1, F2) explains almost 50 % of the total variance of the data. The dissemination of consumers along the quadrants of the plot demonstrates that there was not a homogenous distribution of liking scores, since each vector is a measure of correlation among the scores given to each wine. The projection of the samples is presented in Figure 2b, in which a contrast can be seen between the lower scored red wines in the left-hand quadrants and the higher scores given to the white wines in the right-hand quadrants. The AHCA in Figure 2c shows a clear separation between the appreciation of white and red wines. These results provided statistical justification for the separate analysis of the results from the white and red wines respectively.

The results shown in Figure 2a highlight the absence of homogenous appreciation among the consumers, thus justifying looking for segments with a higher consensus on white and red wines. Therefore, in the next step, a consumer segmentation based on preference scores was applied.

4. Cluster segmentation based on preference scores

4.1. White wines

The consumer preferences for all white wines were subjected to AHCA (Figure S3) showing a separation into three clusters of consumers (demographic characteristics shown in Table S5). Interestingly, a significant effect of the level of involvement of consumers among clusters was observed, with significantly higher scores for Cluster I (average = 3.68) than for both Cluster II (average = 3.20) and Cluster III (average = 3.30) (F = 5.905; p < 0.01), which confirms our initial hypothesis (Table 3). The significant interaction between cluster and sample (F = 17.228; p < 0.0001) shows that the preference for samples depends on the cluster to which each judge belong. The average liking score for each cluster is shown in Table 3, together with a summarised description of the sensory preference styles.

Regarding the consumer description, six significant attributes significantly varied among the white wines; these were citrus/tropical fruit, reduced, dried fruit, acid, oxidation, and neutral (Table 4). The mean liking scores and the sensory characterisation of the white wines were subjected to a PCA (Figure 3); they are listed in Table S6. The ANOVA of RATA scores, with consumers clusters and wine as fixed factors, showed that reduction was the only attribute significantly influenced by the consumer cluster (F = 2.95; p < 0.01). This suggests that the characterisation of wines with this attribute depends on the consumer cluster, with RabOve_A_23 receiving the highest reduction scores from Cluster II, and similar scores being given by Cluster III to both RabOve_A_23 and Callum_BI_22 (Table S6).

Consumers in Cluster I had a higher level of involvement and they preferred all white wines to Donzel_D_18, although the latter’s score was not significantly different from that of Callum_BI_22. This cluster preferred wines dominated by citrus/tropical fruit, such as Blend_T_21 and RabOve_A_23, both of which had higher acidity and lower reduction perception. Donzel_D_18 was slightly penalised due to its lower citrus/tropical fruit perception and higher dried fruit and oxidation notes. Callum_BI_22 was the most neutral wine with lower acidity. Despite this, the consumers in Cluster I showed relatively high appreciation of all the white wines, as the reduction of RabOve_A_23 was low and equal to the other wines in this cluster. As a result, the consumers in this cluster are referred to as intense freshness wine likers who are insensitive to reduction (see Table 3).

Cluster II showed a preference for most wines, except RabOve_A_23. Donzel_D_18, dominated by dried fruits, had lower acidity but was not penalised by oxidation. Callum_BI_22 shared similar characteristics, albeit with lower intensity. Blend_T_21 was also equally liked, likely due to a mild perception of citrus/tropical fruit. In contrast, RabOve_A_23 was the least liked, as its reductive flavours overwhelmed its other sensory attributes, leading to a low neutral score. The consumers in this cluster are referred to as mellowed wine likers and reduction dislikers.

Cluster III was the largest group and its preferences mirrored the overall consumer responses (Figure 3a). Blend_T_21, with its citrus/tropical fruit and acidity, and Donzel_D_18, with its dried fruit and oxidation notes, were the most liked. The difference from Cluster I was related to the lower preference given to RabOve_A_23 and Callum_BI_22, which tended to have higher reductive flavours (Table S6). As Blend_T_21 tended to have higher neutral scores, the consumers in this cluster were referred to as mild fruit likers and reduction dislikers.

4.2. Red wines

The preference scores for red wines were subjected to AHCA, with four different clusters being identified (Figure S5, demographic characteristics shown in Table S5). A relative consensus was observed in the responses of three out of the four clusters, while some of the responses of Cluster II were scattered across the plot (Figure S6). A highly significant interaction between consumer cluster and sample (F = 13.306; p < 0.0001) was observed for red wines, indicating that preference for each sample depended on the cluster to which the judge belonged.

A significant effect of the level of involvement was observed among the clusters (F = 4.397; p < 0.01), with consumers in Cluster IV showing a significantly higher level of involvement compared to the other three clusters (Table 5). This observation is consistent with the involvement effect observed in the white wine set. This difference can be attributed to the higher proportion of high-involved subjects in Cluster IV (Table S5). Unlike white wines, the ANOVA of RATA scores, considering cluster and wine as fixed factors and their interaction, showed no significant effect of cluster on any of the sensory descriptors. The average preference scores for each cluster are shown in Table 5, along with a brief description of the preferred styles. Regarding the description of wines, Table 6 presents the significant attributes that differ among the red wines, including woody/spicy, astringency, neutral, animal, complex, and red/black fruit.

Figure 4 illustrates the PCA with the projection of the significant sensory descriptors, cluster preferences, and the associated wines. Overall, the higher perception of wood/spicy, red/black fruits, astringency, and complexity explains the preference for the commercial blend. By contrast, the animal aroma predominantly characterised Moret_A_22, which was negatively related to preference. TintFran_D_22 was the most neutral wine. NegMol_Al_22 is located in the quadrant opposite the blend and lying between the animal notes of Moret_A_22 and the neutral character of TintFran_D_22. The mean liking scores and the sensory characterisation of red wines are summarised in Table S7.

Cluster I consists of consumers who significantly favoured only the commercial blend wine (bold oaky red wine likers). Consumers in Cluster II gave low scores to all red wines (red wine dislikers). The consumers in Cluster III enjoyed all reds except for the light-coloured NegMol_Al_22, suggesting that colour-linked style perception may have influenced preference. By contrast, the consumers in Cluster IV, composed of a higher proportion (68 %) of high-involved consumers, strongly preferred NegMol_Al_22 and liked Moret_A_22 and TintFran_D_22 less. The high score given to Blend_D_19 indicates that this cluster’s liking was not restricted to a single wine style, contrasting with the preferences of Cluster I (full and light red wine likers).

Discussion

The distribution of the liking scores was qualitatively similar to that in other studies, in which wines were rated by a high number of tasters (Bodington & Malfeito-Ferreira, 2019). The most relevant difference was the assignment of a high number of zeros to red wines, with a concomitant average score lower than that of the whites. Highly reputed wines tend to receive lower ratings when tasted blind (Vecchio et al., 2019; Yang & Lee, 2021; Angelini et al., 2023). In the present study, the overall higher liking scores for white wines compared to red wines may reflect differences in sensory style perception under blind conditions and should not be interpreted as evidence of a broader market trend.

Segmentation according to wine involvement has been described in academic literature using self-reported questionnaires without tasting (Johnson & Bastian, 2015; Pomarici, 2016). In this work, the assignment of wine preferences to the outputs of an involvement questionnaire was followed by wine sample tasting. The findings corroborate our initial hypothesis that wine preferences, based exclusively on intrinsic properties (i.e., sensory profile), are modulated by the level of involvement of the consumer. To our knowledge, these results provide additional empirical evidence for the role of wine involvement in modulating intrinsic preference patterns.

Beyond the specific case study of Portuguese minority grape varieties, the present work also highlights the methodological value of combining rapid expert sensory mapping (PM + UFP) with consumer liking, RATA profiling, and involvement-based segmentation, as a cost-effective strategy to identify sensory drivers of preference and potential niche consumer segments.

The consumers gave consistent wine descriptions, which confirms the fact that consumers can provide wine descriptions comparable to those of trained tasters (Valentin et al., 2012; Varela & Ares, 2012). These responses enabled us to show that most consumers yielded responses consistent with those reported from international wine challenges. The citrus/tropical fruits of white wines are known to elicit the highest appreciation, while for red wines, the red/black fruit, oaky and astringent mouthfeel are typical of the Grand Gold medals (Malfeito-Ferreira et al., 2019).

The lower scores given to the minority wines by the consumers might be a result of their lower familiarity with their sensory features, since preference is affected by degree of familiarity (Yang & Lee, 2020). Nevertheless, the wines from Ramisco, Jampal, and Tinta Miúda might have a commercial advantage, since the descriptions they gave were similar to those of the blends (Long et al., 2023). However, it should be noted that only one commercial reference wine was included per colour category; therefore, conclusions regarding “commercial wines” should be interpreted as comparisons with the selected reference style rather than as a general statement.

In addition, the diversity of vintages included in the initial wine set represents a limitation of the study, particularly regarding the red wines whose commercial reference (2019) differed in age from most of the minority-variety wines (mainly 2021–2022). Wine age and vintage conditions can influence sensory profiles and perceived liking; this should be considered when interpreting the proximity between the samples in the expert-derived clusters, as well as the consumer preference patterns.

Regarding the sensory drivers of preference, in white wines, the contrasting responses were mainly due to the perception of reduction and, to a lower extent, to dried fruits (highly correlated with oxidation). Cluster I, comprising 65 % of high-involved consumers and a higher proportion of older and more frequent consumption individuals (see Table S5), yielded lower reduction scores for RabOve_A_23 than the other clusters. According to Franco-Luesma et al. (2016) and Bekker et al. (2020), the presence of reductive aromas masks fruity and floral aromas, resulting in a more neutral profile. Therefore, the majority of the consumers in Cluster I may have been insensitive to reduction (Lytra et al., 2014). However, reduction can also be perceived under the concept of the so-called mineral character of highly reputed white wines (Rodrigues et al., 2015) explaining the high preference given to RabOve_A_23 (see Table 3). Interestingly, professionals had also recognised a reduction in this wine but with low citation frequency during the PM evaluation. The duality of the concept, where some individuals recognise its high qualitative significance associated with minerality while others associate it with spoilage, may explain the observed conflicting behaviours (Malfeito-Ferreira, 2022). Moreover, reduction may also convey a positive connotation when associated with organic wines (Romano et al., 2020).

Oxidation has been shown to differentiate non-faulty and faulty wines, with few perceptual differences between experts and novices (Franco-Luesma et al., 2019). Accordingly, the white wine that was described using the terms dried fruits and oxidised was from 2018 (Donzel_D_18) and was most appreciated by Cluster II, with an equivalent number of individuals from both involvement levels and a lower frequency of consumption. Indeed, Cluster II only comprised 12 persons, who could be referred to as “oxidation fans” – a term that was applied by Franco-Luesma et al. (2019) to a small group of novices. Nevertheless, wine professionals now adequately recognise the quality of “oxidised” aged white wines, as recently documented by Esteves et al. (2024). For many consumers in the present study, however, the term “oxidation” may have had negative connotations, which could have been minimised if the term “mellowed by age” had been used, as proposed by Marasà et al. (2024) when dealing with proper ageing.

Overall, the sensory discrimination of white wines appears to mainly have been due to dried fruits/oxidation and reduction (see Figure 3) with different quality inferences. In the absence of these attributes, appreciation could not be fully explained by differences in the intensity perception of the consensual flavours of citrus/tropical fruit or acidity. These findings suggest that consumer acceptance of minority-variety white wines may depend not only on conventional fruit-driven freshness cues but also on how consumers interpret sensory notes that can be perceived either positively (e.g., minerality, aged character) or negatively (e.g., reductive/oxidative deviations), depending on involvement and prior experience.

In reds, the main source of conflicting consumer responses was probably wine colour, since the wines were tasted in transparent glasses and no particular attribute could be associated with the different perceptions of each cluster, in contrast to reduction in white wines. The perception of animal aroma might explain the occasional lower preferences, but the behaviour was not so evident as reduction was in the white wines. As a tendency, astringency probably explains the responses of Cluster II (red wine dislikers), in which Blend_D_19 received the highest score for astringency, suggesting that it may have been the penalising factor. The low red colour intensity of the variety Negra Mole elicited higher preference by some individuals, who were clearly outnumbered by those preferring bolder and more aromatically intense wines. Nevertheless, this preference for a light-coloured red, as would be expected for a Burgundian Pinot noir (Valentin et al., 2015), has been mostly associated with high-involved subjects, with a high percentage of women who have high consumption frequency, demonstrating that there are niches that prefer these wines (Yang & Lee, 2021). These results are particularly relevant for minority varieties, as they support the existence of targeted market segments for less conventional red wine styles.

The acceptance or rejection of certain flavours was accompanied by the saliency of their perception, which is an idiosyncratic feature (Barwich, 2017). The fact that liking/disliking could be anticipated, in certain cases, by the magnitude of the sensory response can be linked to olfaction being an emotional sense (Yeshurun & Sobel, 2010). Even before perception and flavour identification, the brain produces an emotional reaction that modulates the flavour perception (Schulze et al., 2017). This was particularly evident with the high intensity grades given to dried fruits and reduction in the most and least liked white wine, respectively, in Cluster II (Donzel_D_18, RabOve_A_23), suggesting that the high variations in preference scores are the result of emotional reactions of affection or distaste. This behaviour was not so clearly observed in red wines for any of the flavours. It may have been possible to associate the animal descriptor with the “horse sweat” off-flavour (Malfeito-Ferreira, 2018), but the red wines were not so markedly affected by this flavour as the white wines by reduction.

The utilisation by the professionals of a sensory attribute with an aesthetic value (complexity) was only valid for reds, being highly correlated with red/black fruit and woody/spicy. The consumer results showed a high positive correlation between complexity and preference, which was expected since neither experts nor non-experts are able to separate enjoyment from aesthetic evaluation (Sackris, 2018).

Conclusion

Preference for the commercial style, which was represented by the control samples, was still dominant for most of the consumers after the responses had been averaged across the full consumer sample. However, given that only one commercial reference wine was included per colour category, these results should be interpreted as a comparison with the selected reference style rather than as a general conclusion regarding commercial wine preference.

Preference patterns were mainly driven by intrinsic sensory cues associated with familiar and widely appreciated profiles, such as fruit-forward character in white wines and red/black fruit combined with woody/spicy notes in red wines.

This behaviour is not surprising, since the pleasantness of the wines dominated by fruit-associated sensory cues – expressed as fresh fruitiness in the white control or red/black fruit and oak in the red control – is often associated with high consumer intrinsic liking.

By contrast, when sensory profiles deviate from these dominant styles – such as wines perceived as being more evolved or lighter in colour – consumers may need to be more familiar with them and have more experience in order to positively interpret them.

The results clearly show that consumers are not a homogenous entity. They demonstrate that the level of involvement of consumers with wine constitutes a determining factor in the modulation of wine preferences based on intrinsic properties. In addition, the preference segmentation revealed the existence of niche consumer groups who show comparable or higher appreciation for specific wines made from minority grape varieties.

These findings highlight the potential of using minority grape varieties to address differentiated market segments and support the usefulness of combining rapid expert sensory mapping with consumer segmentation and involvement stratification to identify sensory drivers of liking. From an applied perspective, targeted communication, and consumer education initiatives may help contextualise and explain less familiar sensory profiles, facilitating a broader acceptance of wines produced from minority grape varieties. Finally, future studies should be carried out to confirm these findings, using a larger consumer sample and a broader set of commercial reference wines, ideally matched by vintage, to better disentangle the effects of wine style, wine age, and seasonal variability on consumer segmentation and sensory drivers of preference.

Acknowledgements

The authors acknowledge all wine companies for kindly donating the wines used in this study.