Mapping the sustainability labels jungle in the food context: an analysis of multi-dimensional multi-level labels Open Access

Food production causes one-quarter of anthropogenic greenhouse gas emissions (Futtrup et al., 2021) and has a huge environmental impact, also because of its use of land and water as well as its effect on biodiversity (Willett et al., 2019). The green transition of the food sector has therefore been high on the political agenda at both national and supra-national levels (e.g. European Commission, 2019). Bringing about the green transition requires behavioral changes from all actors in the food industry, not least consumers (Schulze et al., 2024). The eight billion food consumers, by their choices, can delay or speed up the green transition. Therefore, promoting more sustainable food choices is a major element of the green transition of the food sector.

One attempt to transform consumers' food purchase patterns is to provide and encourage the use of front-of-package labels (see, e.g. Fresacher and Johnson, 2023; Latino et al., 2020; Liu and Wang, 2025). Multi-level front-of-package labels that encourage healthier choices, such as the NutriScore (indicating foods' healthiness based on several dimensions using multiple levels), are well established and researched, and several review papers are available (e.g. Grunert and Wills, 2007; Ikonen et al., 2020; Kelly et al., 2024). In contrast, the sustainability labelling landscape is very fragmented (Futtrup et al., 2021; Torma and Thøgersen, 2021), and currently, no multi-dimensional multi-level sustainability label (MDMLSL) prevails in the food context (Tiboni-Oschilewski et al., 2024). Most existing labels focus on a particular aspect of sustainability and are binary (e.g. with/without genetic engineering; Nes et al., 2024; Weinrich and Spiller, 2016). However, sustainability is a complex and multi-dimensional concept; thus, combining major elements of sustainability in a common label would reduce label clutter and consumer confusion, and encourage label use (Torma and Thøgersen, 2021). Moreover, reducing the multi-facet sustainability into a single attribute risks oversimplification, fails to capture trade-offs between different dimensions (Shaikh et al., 2024; Jürkenbeck et al., 2024) and may lead to so-called waterbed effects, where optimizing products for one dimension increases the negative impacts in another (Torma and Thøgersen, 2024).

Such a multi-dimensional label could increase the comparability of groceries by incorporating not one but several of the most important metrics. However, “despite repeated calls for the development of such a holistic label, to date no label represents sustainability holistically with all of its relevant dimensions” (Futtrup et al., 2021, p. 1412). In addition, while some labels are inherently one-dimensional or binary (e.g. vegetarian), most groceries cannot simply be sorted into “sustainable” and “not sustainable”. Moreover, studies indicate that multi-level labels cause higher consumer satisfaction than binary labels in the food context (Weinrich and Spiller, 2016) and are more effective in guiding toward more sustainable grocery choices (Thøgersen et al., 2024). Similarly, previous research showed a significant interaction effect when combining different dimensions of sustainability labels (Lui and Wang, 2025). One main challenge in implementing a MDMLSL lies in translating complex sustainability information into labels that can easily be understood by consumers (Futtrup et al., 2021).

Some studies have proposed and investigated multi-dimensional sustainability labels (i.e. considering different aspects, such as carbon emissions, water use, etc.), but fall short for two reasons. First, the different dimensions are sometimes outlined as a binary condition (i.e. aspect fulfilled or not). However, merely stating that fair farming conditions exist (or not) does not capture their magnitude or (relative) amount. Second, some multi-dimensional sustainability labels merge various sustainability aspects into one aggregate score. However, this approach can dilute the negative impact of one dimension (e.g. carbon emissions) by having this negative impact compensated by positive scores on other dimensions. In the same vein, recent research attests a lack of overarching guidelines for communicating the impact of various dimensions to consumers in the EU (Zhen et al., 2025). In contrast, a MDMLSL can address these issues by indicating the degree of sustainability both per dimension (instead of binary) and by an aggregated score. To find such a label, two questions need to be addressed:

1. What are the most “relevant” sustainability dimensions for groceries?

2. How should a multi-dimensional multi-level sustainability label be conceptualized/visualized to convey the information in an easily understandable manner?

To address these questions, this article makes the following contributions. First, a literature review is conducted to synthesize previous insights on MDMLSL and on meta-labels, which are often proposed as alternative solutions to MDMLSLs. This literature review is extended by examining those MDMLSL currently being used in selected organic grocery shops. For those labels derived from industry (selected organic shops), we focused on the summary of labels by WBCSD (2021) that are based on life cycle assessments and with a publicly available calculation tool. As a result, it is assessed which sustainability dimensions are currently most often used in the context of sustainable groceries to derive one possible measure of their relevance/importance. Second, an empirical study is conducted to explore different aspects of consumers' preference for MDMLSLs. A Best–Worst Scaling (BWS) design is used to investigate which of the currently existing MDMLSLs and those discussed in the literature is preferred among n = 448 Swedish consumers (representative sample). This addresses the call for research on different design features of sustainability labels, including the trade-off between the amount of information and usability (Taufique et al., 2019). Additionally, the BWS experiment provides a theoretical contribution to Media Richness Theory (Daft and Lengel, 1986) by uncovering the level of media richness (i.e. scope of visualizations displayed in label) and complexity of the communication task (amount of information) preferred most among respondents to enable effective communication for MDMLSLs. Since the labels used in the BWS experiment are modified based on the most frequently employed sustainability dimensions, the aforementioned call for literature for research about a holistic label that shows the degree of sustainability with all of its dimensions will be addressed (Futtrup et al., 2021; Fresacher and Johnson, 2023). Finally, this article responds to the identified lack of empirical investigation on the effectiveness of environmental labels with varying levels of complexity (Hallez et al., 2021). Specifically, by examining multiple MDMLSLs in general, not for a specific food category, the call for exploring consumers’ preference for “a multi-level eco-label on other product categories such as staple food, convenience food, snacks, or beverages” (Kolber and Meixner, 2023, p. 16) is addressed. Consequently, this article helps reduce consumer confusion concerning sustainability labels in the food sector by identifying which MDMLSL holistically considers all important dimensions as suggested by the literature (Stein and Lima, 2022). It helps reconcile the contradicting findings of sustainability labels for groceries (e.g. Vlaeminck et al., 2014; Hallez et al., 2021; Potter et al., 2022) by incorporating all relevant labels proposed both by the literature and those already used in selected organic shops.

For identifying all relevant previous studies that provide insights on food labels with multiple levels, multiple dimensions or the combination of both, a literature research was carried out using a search string that consists of four thematic clusters: (1) synonymously used words for multi-level and/or multi-dimensional, (2) for labels, (3) sustainability aspects and (4) words implying a food context:

We focused on the databases EBSCO, Science Direct, JSTOR, Web of Science, and complemented them by findings from Google Scholar. The literature search took place at the beginning of 2024. This led to a total of 2,964 hits. After screening titles and abstracts and removing duplicates, n = 134 usable studies were identified. However, most of them did not focus on multiple levels or not on multiple dimensions. Therefore, only studies that consider MDMLSL, ML, MD or similar concepts such as meta labels are taken into account (n = 17). These studies are summarized in  Appendix 1.

The literature reveals that MDMLSLs with an aggregated score are being preferred over those without it (Vlaeminck et al., 2014). Labels with traffic-light scheme level visualizations are preferred over specific units of measure (such as exact number of liters for water use) (Vlaeminck et al., 2014; Neumayr and Moosauer, 2021) and result in purchase intentions with lower carbon footprint, less meat and lower water footprint (Hallez et al., 2021). Multiple papers indicate that more information on labels is better (e.g. multi-level design instead of binary labels, as well as displaying multiple sustainability dimensions). Most notably, labels should include the sustainability sub-dimensions (not just one aggregated score) (Vlaeminck et al., 2014). Not only consumers but also food sector stakeholders would welcome a holistic label (Futtrup et al., 2021). Additionally, MLSLs are more effective in reducing greenhouse gas emissions, eutrophication and acidification (Muller et al., 2019), and result in higher consumer satisfaction compared to binary labels (Weinrich and Spiller, 2016). Similarly, MLSLs concerning climate impact are better than no label regardless of their level of additional information (Fresacher and Johnson, 2023). In line with this, MDMLSLs without aggregated scores (showing score for sub-dimensions only), those that only show an aggregated score, and those with a combination of both perform similarly regarding the reduction of greenhouse gas emissions, biodiversity loss, and eutrophication potential, but all of them significantly reduce these compared to when there is no label at all (Potter et al., 2022). Only the water use can be slightly more reduced using an aggregated score label or one with sub-dimensions only, compared with the combined version (Potter et al., 2022).

Based on the literature review, it becomes evident that there are four central limitations that have not yet been addressed. First, some publications discuss different label types without empirically testing them (i.e. conceptual studies, such as Leach et al., 2016; Hélias et al., 2022). Second, some publications report empirical studies but use small samples that do not allow a broader generalization of findings (based on  Appendix 1: 16–84 participants). Third, to obtain a MDMLSL that can be used in several (European) countries, international comparisons should be carried out to increase the findings' generalizability. Only Torma and Thøgersen (2024) recently compared American and German consumers. Fourth, some studies only perform online studies. As a result, a potential self-report bias cannot be excluded (as well as hypothetical bias, since products are not actually purchased). However, given the context of sustainability, a self-report bias is very likely to occur. Our study aims to overcome two of the identified limitations using an empirical study about all relevant MDMLSLs and a large-scale representative sample.

While this article's focus lies on identifying the most important sustainability dimensions (research question 1) and how a MDMLSL should look like, comparing various label formats from academia and industry (research question 2), the literature review also suggests that field experiments (to overcome hypothetical biases) and international studies are research gaps to be addressed in future studies.

Since one-dimensional binary sustainability labels (e.g. vegan or not) fall short of addressing all facets of sustainability for groceries (e.g. greenhouse gas emissions, water use, land use, etc.), the summary of previous work focuses on multi-level and/or multi-dimensional sustainability labels instead (see  Appendix 1). In contrast to multi-dimensional sustainability labels, MDMLSLs do not only consider various aspects, but additionally provide insights on the degree of fulfillment or lack thereof (Torma and Thøgersen, 2021). To clarify how MDMLSLs differ from multi-dimensional only or multi-level only labels, Figure 1 visually categorizes the different types of sustainability labels. For instance, the Fair Trade label takes into account various dimensions (ecological standards; social sustainability by providing fair wages), but it does not allow more granular insights into its level of fulfillment (compared to other multi-level labels, such as the NutriScore). While MDMLSLs' calculation always covers multiple dimensions and multiple levels, they can be illustrated in three different ways. First, they may display an aggregated sustainability score that summarizes the scores for each sub-dimension without explicitly indicating these. Second, they may display the sub-dimensions only without an aggregated sustainability score. Third, they may display both an aggregated score as well as scores for each sub-dimension.

To answer the first research question concerning the most “relevant” sustainability dimension for groceries, the subsequent analysis examines which dimensions are most frequently used by (1) previous MDMLSL literature and (2) in industry (see Table 1). While one might argue that the sustainability dimensions most frequently applied are not necessarily those with the strongest effect regarding the prevention of climate change, it can be assumed that the frequency and relevance may correlate. Moreover, by combining both labels used in industry and those discussed in research, potential biases caused by lobbying diminish. Examining previous literature about which sustainability dimensions are most relevant also responds to the call for analyzing which dimensions should be incorporated into a MDMLSL (Stein and Lima, 2022). Therefore, the subsequent table emerges from analyzing the literature, which examines MDMLSLs, MDSLs, or MLSLs, as well as labels used in selected organic shops (i.e. those MDMLSLs in the WBCSD (2021) summary that are based on life cycle assessments and with a publicly available calculation tool). More specifically, the table outlines which sustainability-related dimension of the related labels has been investigated by which paper.

Based on these findings, the four most widely used dimensions are (1) greenhouse gases, (2) water footprint, (3) land use and (4) – dependent on product type – animal welfare. Additional support for the relevance of these four dimensions is provided by the literature. For instance, the European Commission also found that climate change, water use, and land use are three out of the four most investigated environmental impacts (besides biodiversity) when it comes to sustainability labels for groceries (European Commission, 2024). According to Roa-Goyes and Pickering (2024), greenhouse gases such as CO₂ emissions and methane are the main cause of climate change. Moreover, greenhouse gases are among the influences with the highest environmental impact of groceries (Hallez et al., 2021), which is why they represent the largest share when calculating aggregated sustainability scores for groceries (Vlaeminck et al., 2014). Similarly, the water footprint is among the most important dimensions (Hallez et al., 2021). Other research also considers greenhouse gases and water footprint as crucial determinants for purchasing sustainable groceries (Grebitus et al., 2016). Various studies request incorporating animal welfare (Leach et al., 2016; Stein and Lima, 2022). Accordingly, consumers welcome labels that indicate animal welfare (Weinrich and Spiller, 2016). Furthermore, including animal welfare takes into account a more holistic approach that goes beyond an ecological perspective on sustainability.

All MDMLSLs currently used in industry or discussed in research were implemented in a BWS experiment to reveal the label that solves the trade-off between more information and an easy-to-understand conceptualization in the best possible way. Building on the Media Richness Theory (Daft and Lengel, 1986), this trade-off should balance the media richness (here: extent and type of information displayed) and the complexity of the communication task (here: informing about groceries' sustainability). All relevant MDMLSLs from the previous step were adjusted to comprise all four dimensions. Since some labels reflect multiple dimensions, but are illustrated with varying levels in an aggregated score only (e.g. eco-score by Beelong (WBCSD, 2021), the French eco-score (Marette, 2021)), their visualization will remain with the aggregated score only, while other labels provide graded evaluations for each of the four dimensions. Besides, all labels received an equal score (“B”) to make sure differences in consumers' preferences are caused by the conceptualization, not by the score itself. For MDMLSLs with scores for each sustainability dimension, the scores were selected in a way that their mean score results in a “B” evaluation (e.g., A + B + B + C = B). The labels used from industry were Eco-Score, Eaternity and Planet-Score, since they are the most popular sustainability labels for groceries (BNN, 2022).

Based on the literature review, 20 potential MDMLSLs emerged, including eight from industry. To examine a manageable amount of MDMLSLs for the experiment, three labels from industry were not included (i.e. “M-Check” from Migros, Eco-Score from Beelong, Eco-Score from EIT Food). Instead, we focused on the most well-known MDMLSLs from industry according to a ranking by Verweij-Novikova et al. (2022). In addition to the four most well-known MDMLSLs, the Eaternity label has been evaluated favorably by the literature (Stein and Lima, 2022), resulting in five labels from industry. Additionally, one label resembling the energy efficiency label from the EU was discarded, since several papers found that this type of sustainability label is not preferred compared to other labels (Feucht and Zander, 2018; Neumayr and Moosauer, 2021). The proposed label by Meyerding et al. (2019) and the one by Muller et al. (2019) were discarded, as they visually resemble the one by Vlaeminck et al. (2014), which was included instead. Furthermore, the authors proposed their own MDMLSL suggestion that takes into account various ideas and elements that were found beneficial in previous studies (see Table 2). This selection process resulted in a total of 14 MDMLSLs.

The 14 labels used (Table 2) can be categorized into three groups. In one group, the different dimensions are calculated separately but are displayed by an aggregated score only (i.e. Global label, Eco-Score Industry, Eco-Score Research, Enviroscore, Eco-Impact, Sustain-Rating). The second group of labels illustrates several dimensions with varying levels but without showing an aggregated score (i.e. Eaternity, Table, Guidelines). The third group consists of labels that display both scores for the individual sustainability dimensions but also an aggregated score (i.e. Pie chart, Planet-Score, 10-point Score, Flower-Score, Author's suggestion).

To find the one MDMLSL that consumers are most likely to take into account when shopping for groceries (second research question), a BWS experiment is conducted. This methodological approach proposed by Louviere and Woodworth aims at identifying the best (and worst) option among similar ones (Louviere et al., 2013). The meaning of “best” and “worst” can be adjusted to the specific context (Brand and Kopplin, 2023); in our case, consumers were asked which label is “most likely” and “least likely” to be taken into account when shopping for groceries. The BWS approach allows for a distinct discrimination among similar items, which is especially useful when it comes to comparing slightly different conceptualizations of sustainability labels (de-Magistris et al., 2017). Accordingly, the individual's preference is modeled by the function of the relative frequency with which one option is assessed as superior to its alternatives. Using multiple choice sets with different labels and varying sets, the likelihood of choosing an option can be calculated.

The BWS experiment was part of an online survey. The survey started with a scenario: one enters a supermarket with the intention of buying groceries. Besides focusing on conventional criteria, such as price and taste, one aims to take into account the sustainability of groceries. To do so, one will pay attention to the sustainability label to make an informed decision. To imitate decisions close to reality, the labels were not explained before the choice experiment, since consumers will usually not receive any additional label information before purchasing. The BWS experiment consisted of the 14 sustainability labels (n) derived from industry and literature. For each choice task (exemplary choice task in  Appendix B), four labels were used (m). To counteract potential contextual/order biases, each label was shown three times (k). Based on the formula for calculating the number of choice sets (n/m x k), a total of eleven choice sets were generated (Brand and Kopplin, 2023) using Sawtooth Software. The D-efficiency was 88.67 and the block D-efficiency was 71.92. Subsequent to the BWS, an anchor scaling (direct binary approach) was performed (Brand and Kopplin, 2023) to not only obtain relative preferences (A better than B), but also absolute evaluations (label taken into account in general). Besides the BWS experiment, four control variables environmental concerns (EC) (Kolber and Meixner, 2023), representing an important moderator (Geldres-Weiss et al., 2024), eco-label involvement (EI) (Riskos et al., 2021), green consumption values (GCV) (Brand and Rausch, 2021), Skepticism in sustainability claims (Skept) (Cho and Baskin, 2018) and socio-demographic questions (including income and education, see Liu and Wang, 2025) were asked. The four control variables are based on established constructs from the literature and were measured using a 7-point Likert scale. Before launching the survey, a pre-test with twelve respondents took place, but no adjustments were deemed necessary.

To counteract the limitation of low sample sizes identified ( Appendix 1), a representative sample with regard to age and gender of n = 448 Swedish consumers was obtained using a panel provider (i.e. Kantar). According to a representative study comparing six European countries (Grunert et al., 2014), Swedish consumers use sustainability labels the most when shopping for groceries. Among Swedes, 94% are aware of sustainability labels according to this representative study (Grunert et al., 2014). For translating the survey from English to Swedish, we used the double-backward translation approach. After excluding n = 41 speeders, the final sample consisted of 57% females (due to eliminating speeders and 50% gender-split in age cohorts). The generational cohorts (ranging from 18 to 78 years) match the ones in the Swedish populations with deviations below one percent (all socio-demographic information in  Appendix 3), except for those 78 years or older (−1.2% deviation).

To identify the MDMLSL that is preferred most, we used Hierarchical Bayes estimation (using a total of 30,000 iterations; Sawtooth Software). For the BWS experiment, the utility difference between all possible pairs is calculated and the pair with the maximum difference is selected:

Accordingly, the probability of choosing the pair “i” as the best and “j” as the worst item from a set C of potential item combinations is estimated. The difference for the underlying scale between “i” and “j” is expressed by δ_ij + ε_ij (with additional random error term). The largest difference of all other potential combinations in set C is described by Max(δ_mn + ε_mn). By identifying the best (highest value) and worst (lowest value) item for each set C, the means and variances of these values can be calculated across several choice tasks (Brand and Kopplin, 2023; de-Magistris et al., 2017). The probabilities can then be described by a multinomial logit (MNL) model, when assuming that the random error term follows an independent identical distribution:

The likelihood function can be obtained using the chosen best and worst items throughout all choice tasks of every respondent. Assessing the resulting quality criteria, the root likelihood of 0.605 indicates high internal consistency (compared with the naïve model of 0.25). In addition to the common probability scales, the anchor scaling was assessed following the direct binary approach (see Brand and Kopplin, 2023). As a result, not only relative preferences, but also an absolute evaluation (whether it will be taken into account or not) becomes possible. Additionally, the BW score (de-Magistris et al., 2017; Rausch et al., 2021) was calculated to prevent potentially skewed results from the HB estimation. This score focuses on relative frequencies of choices by subtracting the number of times an item/label was selected as worst from the number of times an item was selected as best and dividing the result by the number of times the item was displayed.

The results (Table 3) indicate that the MDMLSL proposed by Potter et al. (2022) is preferred most (“Flower-Score” with 207.47 absolute and 11.13% relative preference), closely followed by the Planet-Score label (205.05 and 10.61% respectively). Compared with the second most preferred label (“Planet-Score”), which exists in the industry, the Flower-Score does not perform significantly better (z = −0.938, p = 0.348; Mann–Whitney–U test, using SPSS) when focusing on the probability score with anchor, but does so compared to the one without (T(447) = 1.768 with p = 0.039; SPSS). However, the Flower-Score (z = −4.594, p < 0.001) as well as the Planet-Score label (z = −2.157, p = 0.031, using Mann–Whitney–U tests) are significantly more preferred compared with the third best label (“Eco-Score Research”). Generally, labels that combine an aggregated score with sustainability sub-dimensions tend to perform better than those offering only an aggregate score (z = −1.969, p = 0.049) or only sustainability sub-dimensions (z = −2.813, p = 0.005, using Mann–Whitney–U tests). Accordingly, out of the five most favored MDMLSLs, three represent such combined solutions (i.e. Flower-Score, Planet-Score, 10-point Score). Apart from that, almost all labels are taken into account by Swedish consumers when shopping for groceries. Only the “Global label” tested by Potter et al. (2022) would not be considered by consumers.

To allow for better visualization of the main results, Figure 2 summarizes the relative (without anchor) and absolute preference (with anchor) for all MDMLSLs.

When examining the results for different socio-demographic segments, some differences emerge. Accordingly, females preferred the Eaternity label more (mean = 7.88%) than males (mean = 6.60%; z = −2.054; p = 0.040). There were no differences between the generational cohorts and levels of education, which increases the findings' generalizability across age groups. The only other distinction emerged for different income segments. Those with medium to higher income groups (earning 45 to almost 54 thousand SEK) show a lower preference for the Sustain-Rating label with its aggregated score (M = 3.20) compared to those earning 36 to almost 45 thousand SEK (M = 6.38; χ² = 85.924; p = 0.026), to those earning 27 to almost 36 thousand SEK (M = 6.92; χ² = 96.378; p = 0.006), 18 to almost 27 thousand (M = 8.36; χ² = 122.627; p < 0.001), 9 to almost 18 thousand SEK (M = 7.83; χ² = 106.058; p = 0.002) or those with less income than nine thousand SEK (M = 8.13; χ² = 114.688; p = 0.002).

Focusing on the background variables, interesting differences were found concerning consumers' level of EC, their GCV, their Skept and their EI. Accordingly, consumers with high EC are significantly more likely to take any of the proposed MDMLSLs into account (based on median split with EC_high = 170.80; EC_low = 117.22, z = −7.73, p < 0.001). Similar patterns were observed for GCV (GCV_high = 173.09; GCV_low = 115.99, z = −7.36, p < 0.001), Skept (Skept_high = 162.63; Skept_low = 123.66, z = −4.89, p < 0.001) and EI (EI_high = 172.79; EI_low = 126.33, z = −5.72, p < 0.001).

To identify the one MDMLSL that best handles the trade-off between complete complex information on the one hand and being easily understandable on the other hand, this article investigates consumers' preferences for several MDMLSLs for groceries. Since sustainability is not binary and consists of several different dimensions, this study's focus lies on investigating MDMLSLs to provide a more accurate and informative approach to labelling. By incorporating both those sustainability labels for groceries previously examined in research and those that are currently available in some organic stores, this article provides a summary of all recently discussed/employed labels and follows the suggestion by Torma and Thøgersen (2021) to build on labels already existing. Additionally, it follows the call for standardization and harmonization efforts (Geldres-Weiss et al., 2024) by focusing on existing labels.

To answer the first research question, four dimensions are identified as being used most frequently (i.e. greenhouse gases, water footprint, land use, and, depending on product category, animal welfare). Using this as the foundation, the previously detected MDMLSLs are adjusted by incorporating those four dimensions. The results of the related BWS experiment with n = 448 Swedish consumers show that those MDMLSL, which combine displaying both an aggregated score and additional dimensions, tend to be favored (three of these types of MDMLSLs in the top five). Besides this type of label in the top five (i.e. Flower-Score, Planet-Score, 10-point Score), only the Eco-Score research (3rd) and the Guidelines label (4th) are ranked comparably high. This may be due to the visual closeness to the already established NutriScore and the nutritional values boxes. Therefore, these two labels may have benefited from higher levels of familiarity.

The only other labels with aggregated score and a score for each dimension are the label proposed by the authors and the Pie Chart label. One reason why these two were less preferred might be their general design. No other label uses a circle design with separate shares to illustrate the dimensions' sustainability impact. Instead, most labels use boxes and are designed rectangular. Similarly, no other label used a QR code or separate, additional boxes next to the aggregated score box to visualize their impact (as in the authors' suggestion). In contrast to other labels, the use of additional pictograms for these two labels may increase their complexity and cause diffusion. While the Eaternity label is also using small pictograms, this label already exists in selected organic stores and thus, may have benefited from higher levels of familiarity. Thus, the use of such pictograms may not necessarily be beneficial by catching consumers' attention (Leach et al., 2016) but can also cause confusion when it makes the label design too complex.

Additionally, all other MDMLSLs with both an aggregated score and for each dimension display the dimension scores and thus, a more negative evaluation than the average “B”, in a less salient way: The Flower-Score uses petal sizes and color, the Planet-Score white dots on the colored scale, the 10-point Score colored manifestations on the 10-point scale. As a result, the only score directly observable are “B” evaluations for the Flower-Score, the Planet-Score or the 10-point Score (with 7.5 out of 10 score respectively), which represents the second-best score (on the scale from “A” to “E”). Therefore, only the author's suggestion and the Pie Chart label outline more negative scores (e.g., a “C”) directly perceivable. As a result, these relatively negative individual scores might have caused a less favorable evaluation even though the aggregated score is the same.

One label proposed by research (i.e. Flower-Score; Potter et al., 2022) yields the highest preference, closely followed by the Planet-Score label. However, significant differences were found depending on consumers' EC, their EI, Skept and GCV, indicating heterogeneity among consumers. Thus, some consumers may not even consider such labels depending on their EC, EI, Skept and GCV (i.e. probability score below anchor of 100) when shopping for groceries. This corroborates previous research that also found no effect of age on sustainability label preference, but for different levels of environmental concerns (Liu and Wang, 2025).

As a result, this study corroborates the positive effect of the Flower-Score on consumers' preferences (Potter et al., 2022), in line with its positive evaluation by Stein and Lima (2022). Moreover, it confirms that sustainability labels with multiple dimensions and higher information density are being more preferred (Muller et al., 2019). However, this contradicts findings from previous literature claiming that consumers favor simple sustainability labels (Roa-Goyes and Pickering, 2024). This discrepancy may stem from cultural differences (investigating Canadian and Argentinian consumers) compared to European countries (here: Sweden), respondents' age (18–24 years versus 18–87 years) or a combination of both. Besides, there can be differences between which label consumers prefer and which one is actually best for directing choices into the desired direction (see, e.g. Ikonen et al., 2020).

While “sustainability” for groceries means more than just CO₂ emissions/greenhouse gases, many sustainability labels focus on one dimension only. Therefore, the identification of the most important sustainability dimensions for groceries becomes crucial. Among those studies that incorporate multiple dimensions, there seems to be no consensus yet about which dimensions should be taken into account. By systematically analyzing the literature and assessing current label use in industry, this study contributes to the literature by providing a unified understanding of what the most relevant sustainability dimensions for groceries are. Moreover, the dimensions detected were additionally corroborated by literature indicating their importance (see section 2.2). Consequently, this study responds to the call for defining which sustainability dimensions should be incorporated in MDMLSLs for groceries (Stein and Lima, 2022).

In contrast to labels indicating groceries' healthiness (most frequently: NutriScore), no sustainability label prevails in the yet largely fragmented market of various labels. To reveal the one label consumers are most likely to take into account when shopping for groceries, this article contributes by summarizing the labels used in practice and those discussed in literature to disentangle the current “label jungle”. As a result, one specific label stands out as being the most preferred (i.e. “Flower-Score” by Potter et al. (2022)). Additionally, the group of MDMLSLs that combine an aggregated score with scores on sustainability dimensions tends to be consumers' most favorite ones (with three of these labels among the top five). Hence, (Swedish) consumers prefer more information about groceries' sustainability, which may indicate the pursuit of transparency to increase their trust in the corresponding labels. This aligns well with the fact that the group of MDMLDLs with one aggregated score only (lacking scores for sub-dimensions) is less favored and the seemingly least detailed “global” label would not be taken into account at all.

This article also responds to the identified lack of empirical investigations about the effectiveness (by taking the label into account or not using the anchor scaling) of environmental labels with varying levels of complexity (Hallez et al., 2021). Similarly, this article answers the specific calls for exploring “the effects of a multi-level eco-label on other product categories such as staple food, convenience food, snacks, or beverages” (Kolber and Meixner, 2023, p. 16) on consumers' preference. Moreover, the findings revealed in this study answer the call for the development of a holistic sustainability label with all relevant dimensions (Futtrup et al., 2021). While nutrition and health emerge as one of eight topic clusters from recent bibliometric analysis (Latino et al., 2020), whereas the aspect of sustainability does not, this study appears to contribute to a relatively emerging field of research in the context of food label design.

Another contribution is made by providing empirical evidence for the Media Richness Theory (Daft and Lengel, 1986) in the context of sustainability labels for groceries. Accordingly, labels with an aggregated score only inhere a comparably low level of media richness (too simple design for complex information) and thus, fall short of effectively communicating a product's sustainability. Instead, to take into account the multi-facet and complex nature of sustainability, more information and higher media richness (i.e. additionally displaying sub-dimensions and an aggregated score) is needed to increase consumers preferences for sustainability labels.

When investigating consumers' behavior concerning sustainable products, the intention-behavior gap is likely to occur due to social responsibility bias. To counteract such biases, this investigation not only asks for relative preferences regarding the different MDMLSLs, but also includes an anchor scaling that controls for an absolute evaluation of each label (i.e. will it be taken into account or not). While this additional information relies on self-report data, it contributes to the sustainability label literature for groceries nevertheless by separately inquiring consumers' reported actual assessment besides their (relative) preferences from the BWS experiment.

As indicated from the results of the literature review, current research is limited by a lack of international comparisons to increase findings' generalizability and by an overemphasis of online studies. This article's focus is on identifying the most important sustainability dimensions and how an MDMLSL should look like, but future research may focus on field experiments and international studies (see also Geldres-Weiss et al., 2024). While field experiments would allow to overcome a potential attitude-behavior gap (particularly, in the light of sustainability), cross-country studies would increase the findings' generalizability. More precisely, this study uses a representative sample of Swedish consumers; however, the insights gained may be limited to consumers from this particular country, which is why future studies may verify their generalizability across countries or different cultures.

Given that Swedes tend to use sustainability labels for groceries the most (out of six European countries; Grunert et al., 2014) and with Sweden being among the countries with the highest environmental performance (Statista, 2024), they might require more sustainability information than other countries. This could explain why many MDMLSLs with both an aggregated score as well as scores for each dimension are among the most preferred, whereas several labels with just an aggregated score did not perform as good. In contrast to Sweden, countries with lower sustainability awareness might favor simplified MDMLSLs with less information amount instead, which is why future research may replicate this study across countries. We therefore call for international comparisons.

The labels exposed to respondents were not explained beforehand. Hence, some consumers may not have been familiar with the MDMLSLs displayed. However, it appears that this mimics grocery shopping more realistically, since consumers would not receive an explanation about sustainability labels when shopping as well.

Apart from that, this research focuses on identifying the label that provides the best possible trade-off between communicating a sufficient amount of information and yet having an easy-to-understand visualization. Therefore, the label that is most likely to be taken into account when shopping for more sustainable groceries (i.e. Flower-Score) from a consumers' perspective does not necessarily represent the label with the best impact for a more sustainable world. However, even the best product is merely being sold, when it is not being understood by the consumer or using the “voice of the customer” for explaining it. Therefore, future research may contrast which labels have the largest impact on the environment with the label being most preferred by consumers.

Since the label standing out as the most preferred one (i.e. Flower-Score proposed by Potter et al. (2022)) has not been tested in retail stores yet, this label could be examined in practice (and be contrasted with similar labels focusing on the largest impact on the environment). Subsequently, based on the insights gained in this study, the industry may start field experiments by using the MDMLSL proposed by Potter et al. (2022), since it stands out as being the most preferred one (relative probability score: 11.30%). Interestingly, this MDMLSL suggested in research even outperformed sustainability labels currently existing in practice (such as the second-most preferred Planet-Score). However, this may be caused by the sparse distribution of sustainability labels in grocery stores. Currently, only a few hand-picked stores offer groceries with MDMLSLs on top.

In general, those labels existing in industry (i.e. Planet-Score, Eaternity, Eco-Score, Enviroscore, Eco-Impact) performed relatively badly (ranked 2nd, 6th, 7th, 10th, 11th, respectively), given that consumers might have already heard about these labels. As a result, this shows that even so far, completely new/unknown labels can make consumers consider sustainability aspects for purchasing groceries when visualized in a favorable manner and covering all relevant information. Therefore, producers and policymakers do not necessarily need to rely on the few existing labels for illustrating groceries' sustainability holistically, but could consider implementing a new label (such as the Flower-Score) that effectively works and encompasses multiple dimensions, as well as multiple levels (since this appears to be preferred by consumers).

Since several MDMLSLs that combine both an aggregated score as well as a score for each sustainability sub-dimension are among the most preferred (three out of the best five), retailers may start testing the impact of these types of labels (and thus, not only focus on the Flower-Score that performed best).

In order to assess the environmental impact of groceries, consumers prefer labels that contain both an aggregate score and separate scores on the underlying dimensions of greenhouse gas emission, land use, water use and animal welfare. Among all the labels tested, the Flower-Score proposed by Potter et al. (2022) was the most preferred one, closely followed by the Planet-Score. These two labels both fall within the type of MDMLSLs that combine an aggregated score with scores for each sustainability dimension. Preferences vary with EC, EI, Skept and GCV, highlighting consumer heterogeneity. These findings provide practical guidance for further testing in real-world settings and for designing clear and informative sustainability labels that effectively support the green transition to a more sustainable food system.