Artificial intelligence-generated sustainable gift box design evaluation via reinforcement learning-driven hybrid molecular fuzzy modelling

Abstract

Considering its significant role in the first contact customers have with the product, designing a distinctive product packaging that effectively conveys the intended meaning of a product is crucial, but also challenging and complex for companies. Hence, the research aims to identify and rank critical front-of-package cues for packaging design, and to select the best packaging design, in the form of a gift box, among artificial intelligence-generated alternatives. A reinforcement learning-based hybrid decision-making method is developed and adopted to facilitate ranking and selection processes. The proposed methodology, with a spatial and geometric approach, uniquely applies the concepts of molecular geometry to define degrees of membership, non-membership, and hesitancy within the fuzzy set framework, offering a new perspective in fuzzy decision-making. The methodology’s foremost superiority is combining the fuzzy decision-making approach with molecular geometry knowledge. Thus, the methodology’s use of molecular fuzzy numbers significantly contributes to state-of-the-art engineering. To evaluate the applicability and robustness of the methodological framework, a gift box design for the sustainable products of a textile company is created and implemented as a real case study. The results reveal that brand identity and aesthetic appeal are the most prioritized design criteria in the package design process. Based on the overall evaluation, the third alternative is the best artificial intelligence-generated gift box design for the respective company. This research highlights the key aspects of packaging elements necessary for effective packaging design and the application of artificial intelligence in the gift box design process, providing practical implications for professionals responsible for packaging design. Additionally, this research contributes to the weighting and ranking problems of artificial intelligence-generated gift box design assessment by using molecular fuzzy sets and the reinforcement learning technique, thereby providing a more accurate approach.

Introduction.

As the number of diverse competitive brands placed side-by-side on the shelves increases, it compels many brands to find a novel way to attract consumers’ attention at first sight¹. In the face of this fierce competition, the significance of packaging is increasingly heightening in promotion efforts since visual stimuli are the most dominant stimuli in the perception process in influencing customers’ choices². Once the number of customers making their buying decisions at the point of purchase, particularly for the low-involvement products, is considered, packaging has become an effective communication medium in promotion strategy to capture customers’ attention³. Thus, packaging is not solely considered a protective container for a product today, but it also acts as a silent salesman at the point of purchase, influencing consumers’ buying decisions and behaviors through its informational and persuasive power⁴. The packaging design elements mostly give customers several cues about the product and shape their perceptions of the product inside the package⁵. These elements also affect brand meaning by transmitting messages about brand identity and brand values⁶ and brand personality and value propositions⁷. Thus, functioning as a complete reflection of product characteristics and brand concept, product packaging has become an essential part of product marketing⁸.

In response to the growing importance of packaging in promotion, companies have prioritized the design of packages because early impressions of consumers towards products are mainly shaped by front-of-packaging designs⁹. Academic research also indicates that buying decisions rely more on customers’ use of the front-of-packaging cues relative to the information on the back of the packaging¹⁰. Through verbal and visual cues on the packaging, marketers can communicate product attributes in line with the intended brand or product image¹¹. So, by playing with several design elements, including shape, size, color, typography, and imagery, marketers can create different packaging design concepts for various products to transmit intended messages about them¹. Undoubtedly, front-of-packaging characteristics can also be manipulated to communicate messages about sustainability and the environmental friendliness of the products through cues like sustainability claims, colors, and images, reminding consumers of sustainability to induce consumers to make more sustainable choices¹².

Nevertheless, although some research exists on packaging design elements¹³, the integrated design of front-of-packaging characteristics to convey desired meaning about a product’s sustainability remains a significant challenge for marketers and product designers. Herein, the challenge depends on the individual selection of each packaging design element and understanding which elements are more critical in creating a desirable image and meaning for the product³. Indeed, various packaging design cues have also been explored within the context of packaging for sustainable products^14,15,16 together with the influence of front-of-packaging attributes on attitudes, purchase decisions, and buying behaviors from the perspective of consumers^17,18,19. However, packaging design has not yet been comprehensively explored within the context of how the front-of-packaging cues should be designed for sustainable products from a company’s perspective. Herein, it is vital to note that packaging design cues are not adopted equally and do not affect customers equally⁶. Thence, which cues should be prioritized in the packaging design of a sustainable product from the perspective of companies remains a question mark in extant literature. Additionally, as a recent trend in packaging design, artificial intelligence (AI) has a pivotal role in transforming the packaging of products by offering sustainable packaging design alternatives²⁰. AI, which can convert text to image, paves the way for visual creations based on text easily and fast, leading to time and cost efficiency in the design process^21,22. In prevailing literature, there is much research that integrates packaging design with AI technologies within the context of energy-saving packaging design²³, personalized product packaging design²⁴, cosmetic packaging design²⁵, pharmaceutical packaging, food packaging design^26,27, as well as sustainable food packaging design²⁸, and advanced packaging²⁹. Besides these studies, only a few research studies have examined the influence of AI-generated packaging on consumers’ behavior³⁰. However, none of the research attempts to investigate the applicability of AI in the design of the front-of-packaging cues to communicate the intended brand meaning for sustainable products. Thus, since the use of AI in packaging design is highly applicable in the packaging context, various conspicuous and vital gaps exist in the relevant literature, particularly on creating and selecting the optimum AI-generated packaging design from the companies’ perspective.

Besides, from a methodological perspective, there are several qualitative studies¹⁸ and qualitative research¹⁶ on front-of-packaging cues. However, there is limited research that adopts multi-criteria decision-making (MCDM) methods to prioritize critical factors regarding packaging design and select the best packaging design alternatives in the extant literature. Best-Worst Method (BWM) for the selection of sustainable product-package design³¹, fuzzy analytical hierarchy process (AHP) and fuzzy Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) for selection of green packaging design³², ZF-DEMATEL-TOPSIS method for selection of packaging circular model³³, automatically weighted Multi-Criteria Decision-Making Analysis (MCDA) and TOPSIS for selection of sustainable packaging materials³⁴, and BWM and Combined Compromise Solution (CoCoSo) integrated in an Asymmetric Interval Type-2 Trapezoidal Fuzzy Sets (IT2TFS) for evaluation of eco-friendly packaging alternatives³⁵. Nevertheless, all these studies primarily concentrate on sustainable packaging design instead of prioritizing and ranking key front-of-packaging cues for sustainable products. Therefore, it is unclear which front-of-packaging cues should be prioritized by companies to communicate the sustainability of their products. Moreover, the prevailing literature lacks research on prioritizing key front-of-packaging cues and selecting the best AI-generated packaging design alternative for sustainable products, as opposed to sustainable packaging.

All of the above discussion led us to suggest the following research questions: (1) What are the key front-of-packaging cues prioritized for a packaging design for a sustainable product? (2) How are AI-generated packaging design alternatives for a sustainable product ranked, and which alternative is selected as the best? (3) Could a novel decision-making technique for providing more comprehensive results be developed based on molecular geometry and a Q-learning algorithm that dynamically adjusts weights based on rewards and penalties to enhance adaptability and accuracy? (4) Could the proposed MCDM methodology be effectively applied to ranking and selection processes in a case study involving AI-generated gift box designs for a textile company producing organic and sustainable clothes for new-born babies? (5) Could AI be effectively utilized to create gift box design alternatives tailored to a company’s identity and the sustainable characteristics of its products? (6) Does the integration of MCDM methods and AI usage address existing research gaps in selecting the best package design?

This research makes several significant contributions to the literature. First, it reveals the critical front-of-packaging cues to be considered in packaging design for sustainable products and selection of the best AI-generated package design alternative for them. Second, it indicates the applicability of AI technology in the package design process, making it easy for marketers to effectively tackle potential packaging design problems. Third, the proposed methodology uniquely applies molecular geometry concepts to define membership, non-membership, and hesitancy degrees within a fuzzy set framework. This spatial and geometric approach is a new perspective in fuzzy decision-making. Using specific geometric angles (e.g., linear, trigonal planar, tetrahedral) to normalize membership degrees introduces a proportional and balanced calculation mechanism, enhancing accuracy and coherence. Further, the methodology extends traditional fuzzy cognitive maps by incorporating molecular fuzzy sets to model complex relationships with enhanced accuracy and multidimensional representation. Combining molecular fuzzy sets with Multi-Objective Particle Swarm Optimization (MOPSO) introduces a novel mechanism to balance Pareto-optimal solutions under uncertainty. Additionally, by incorporating reward and penalty factors into the Q-learning framework, the methodology uniquely optimizes pairwise and decision matrices dynamically. Overall, this research also makes methodological contributions by addressing the issue of unbalanced expert inputs through the optimization of q-values via iterative learning, a rarely explored approach in existing expert prioritization methodologies. Implementing the integrated decision-making technique with molecular fuzzy logic introduces an innovative weighting and ranking mechanism. This hybrid approach balances the local and global exploration of the solution space with nuanced uncertainty modeling, which is absent in standard decision-making methodologies. A case study for a textile company also demonstrates the applicability and efficacy of the novel methodology in a real-world phenomenon, serving as a brand-new decision support model.

This research paper is organized as follows. Section 2 provides brief information on creating packaging designs for sustainable products within the context of front-of-packaging cues. Section 3 provides a methodological framework for the research. Section 4 presents a case study on selecting a gift box for a family-owned textile company that specializes in producing organic and sustainable clothes for newborn babies. Section 5 presents the findings and implications of the results. Finally, Sect. 6 presents the conclusion, limitations, and further research suggestions.

Background of packaging design for sustainable products

Packaging design is designing and creating a package that encloses, protects, transports, and stores a product³⁶. Besides, product packaging is also a consumer’s initial impression of a product, influencing its overall performance¹. The perceived characteristics of product packaging, such as color, shape, size, imagery, and typefaces, mostly communicate meaning about the product inside the package, which in turn leads to several perceptions in the minds of consumers³⁷. Accordingly, front-of-packaging characteristics can act as extrinsic cues to be used by consumers to evaluate the product when intrinsic attributes regarding the product cannot be evaluated before the purchase³⁸. This is generally achieved via visual and verbal cues on front-of-packaging that indicate meaning to customers⁶. Herein, while visual cues consist of color, shape, size, and graphics, verbal cues include product and label information as well as brand-related information such as brand name, logo, and slogan¹³.

Front-of-packaging cues for sustainable products

In packaging design, it is crucial to notice that if cues are related to customers’ beliefs regarding sustainability, they can enhance the customers’ perception of the sustainability of the products^39,40. Accordingly, since customers do not always carefully scan the package and information on labels⁴¹, front-of-packaging cues can be designed to help them make sustainable choices. Herein, some cues on the packaging can be explicitly associated with sustainability, such as brand logo and marketing messages, whereas some cues can be implicitly associated with sustainability, such as color, style⁴⁰, size, shape, and graphic design⁴².

As per these cues, the size and shape of the packaging can influence customers’ perception of sustainability, conveying meaning to customers¹², as well as influencing the presence of packaging on the shelves to capture customers’ attention¹. Previous qualitative studies reveal that lack of packaging, over-packaging, smart shapes leading to less usage of packaging material, and small size in packaging are the indicators of eco-friendliness⁴³. Empirical studies also indicate that rectangular and square packaging is more associated with sustainability regarding transportation efficiency and less use of packaging material⁴⁴. Moreover, large packaging can signify lower sustainability once “less is more” is considered⁴⁵.

Color is the other critical front-of-packaging cue for conveying messages about products, influencing customers’ product judgments⁴⁶, and attracting their attention¹. Through colors on packaging, some product and brand messages, such as naturalness⁴⁷, sustainability⁴⁵, or eco-friendliness⁴⁸ can be signified. For instance, green is mainly related to nature⁴⁹, eco-friendly consumption, and sustainability^50,51. Besides green, package colors such as brown and cream are more associated with greater eco-friendliness and sustainability, as well as plain packages with only a little color or ink⁵² and dull colors¹⁵.

Packaging design is also vital in branding products in a retail environment because packaging acts as an effective medium for transmitting a brand message and identity¹. Packaging design also includes putting brand names, slogans, product descriptions, and labels on product packaging⁴⁵. Herein, brand names make customers spot differences in quality and stimulate social recognition⁵³, while brand slogans influence customers’ decision-making process⁵⁴. Product descriptions on the front of packaging also significantly shape perceptions of persuasiveness and trustworthiness⁵⁵. Moreover, typography, as a component of brand identity, is the visual reflection of a spoken language in the form of a typeface of words⁵⁶. Each typeface has its distinctive connotations, leading to a unique representation of what your brand is and what it stands for. When communicating a brand name, logo, slogan, product information, and marketing message to target consumers, typography acts as a visual expression of the product that can be initially noticed from a distance¹. In the packaging design process for sustainable products, putting a label, name, logo, slogan, and product information reminding customers of the product’s sustainability on top of the product packaging can attract attention to this label, enhancing the sustainable nature of the product that customers will use as a preference cue⁴⁵.

Finally, since customers tend to glance at images on the packaging before reading words, imagery as front-of-packaging cues should also be effectively designed to convey intended brand messages¹. In the packaging design process, using illustrations and photographs that remind one of organic production and sustainability for products enhances the effectiveness of marketing communication through packaging⁵⁷. For instance, images of leaves, green globes, or other pictures related to nature can serve as a cue for environmental friendliness⁵⁸.

Gaps in the relevant literature

In the extant literature, there is some research on packaging design cues to enable consumers to form a judgment about the sustainability of the products¹². However, these studies may not necessarily provide clear directions to product designers and marketers on determining the importance of these cues in the front-of-packaging design process using a robust decision-making system. Moreover, previous research on front-of-packaging cues design has mostly ignored recent technologies, such as the applicability of AI in the packaging design process within a marketing context. Through the use of AI in creating text, images, video, audio, and designs, creative content has been developed in recent years³⁰. With automated problem-solving functions designers previously executed, AI provides a brand-new approach, diverging from designers and ensuring more straightforward solutions to complex and challenging tasks without volume and speed limitations⁵⁹. AI provides designers with numerous design options to choose from based on their preferences, enabling them to improve their work by modifying existing designs and incorporating data insights. Thus, today, integrating AI’s application in product packaging design assists product designers and marketers in creating several environmentally friendly and sustainable packaging design alternatives⁶⁰. Overall, the rationale for this research is to examine the applicability of AI in front-of-packaging design using a robust decision-making approach that determines the level of importance of each front-of-packaging cue and the selection of the best AI-generated packaging design alternative based on these importance weights.

Proposed model

The proposed hybrid decision-making framework integrates two complementary approaches, Multi-Factor Multi-Objective Particle Swarm Optimization (MF-MOPSO) and Q-learning, to guide packaging design decisions under uncertainty. The methodological flow is summarized in Fig. 1.

Fig. 1

Flowchart of the proposed hybrid methodology.

In step 1, the process begins by identifying design criteria (e.g., brand identity, sustainability, material cost, visual appeal) and feasible packaging alternatives. Expert judgments are converted into fuzzy numerical scales to capture subjective assessments. In step 2, MF-MOPSO is used to determine the optimal importance weights for each criterion. In intuitive terms, this phase asks “which factors matter most?” while handling trade-offs among multiple conflicting objectives such as cost, aesthetics, and environmental impact. In step 3, after the weights are defined, the Q-learning algorithm performs iterative learning to refine decisions. The agent evaluates each packaging alternative, receives feedback (reward) based on performance, and adjusts its strategy in subsequent iterations. Conceptually, this phase answers “how should the decision evolve based on experience?” In step 4, the iterative interaction between MF-MOPSO and Q-learning continues until stability is achieved in both weight distribution and decision policy. The best-performing alternative—typically the design with the highest aggregated satisfaction index—is selected as the optimal packaging configuration. In step 5, the final model is validated through sensitivity analysis to test robustness against parameter variation. The process ensures the reliability of the hybrid framework across multiple packaging scenarios.

Thus, MF-MOPSO identifies and optimizes the importance of design criteria, while Q-learning adaptively refines the decision process based on feedback. The two methods operate sequentially yet interactively, ensuring both quantitative optimization and learning-driven adaptability.

Molecular fuzzy sets

Molecular fuzzy numbers are a new type developed by combining fuzzy logic with molecular geometry. Molecular fuzzy numbers are based on the geometric structure of a molecule⁶¹. The primary purpose of this process is to reduce uncertainty in decision-making analyses. Equation (1) should be satisfied in this regard.

$$\:{\mu\:}_{U}\left(u\right)+{\nu\:}_{U}\left(u\right)+{\xi\:}_{U}\left(u\right)=1$$

(1)

For the calculation of the degrees, Eqs. (2)-(4) are considered.

$$\:{\mu\:}_{U}\left(u\right)=f\left({\theta\:}_{u}\right)=1-\frac{{\theta\:}_{u}}{{\theta\:}_{max}}$$

(2)

$$\:{V}_{U}\left(u\right)=\frac{{\theta\:}_{u}}{{\theta\:}_{max}}$$

(3)

$$\:{\xi\:}_{U}\left(u\right)=1-\left(\left(1-\frac{{\theta\:}_{u}}{{\theta\:}_{max}}\right)+\frac{{\theta\:}_{u}}{{\theta\:}_{max}}\right)$$

(4)

Equations (5)-(12) explain the general calculation process of two molecular fuzzy sets, A and B.

$$\:{\mu\:}_{A}\left(u\right)\le\:{\mu\:}_{B}\left(u\right)\:{\nu\:}_{A}\left(u\right)\le\:{\nu\:}_{B}\left(u\right)\:{\xi\:}_{A}\left(u\right)\ge\:{\xi\:}_{B}\left(u\right)\:\forall\:u\in\:U\:A\subseteq\:B$$

(5)

ifandand,

$$\:A\subseteq\:B\:B\subseteq\:A\:A=B$$

(6)

ifand

$$\:A\cup\:B=\left\{\left(u,\text{max}\left({\mu\:}_{A}\left(u\right),{\mu\:}_{B}\left(u\right)\right),\text{min}\left({\nu\:}_{A}\left(u\right),{\nu\:}_{B}\left(u\right)\right),\text{min}\left({\xi\:}_{A}\left(u\right),{\xi\:}_{B}\left(u\right)\right)\right)\mid{u}\in\:U\right\}$$

(7)

$$\:A\cap\:B=\left\{\left(u,\text{min}\left({\mu\:}_{A}\left(u\right),{\mu\:}_{B}\left(u\right)\right),\text{max}\left({\nu\:}_{A}\left(u\right),{\nu\:}_{B}\left(u\right)\right),\text{max}\left({\xi\:}_{A}\left(u\right),{\xi\:}_{B}\left(u\right)\right)\right)\mid{u}\in\:U\right\}$$

(8)

$$\:A\cdot\:B=\left\{\left(u,{\mu\:}_{A}\left(u\right)\cdot\:{\mu\:}_{B}\left(u\right),{\nu\:}_{A}\left(u\right)\cdot\:{\nu\:}_{B}\left(u\right),{\xi\:}_{A}\left(u\right)\cdot\:{\xi\:}_{B}\left(u\right)\right)\mid{u}\in\:U\right\}$$

(9)

$$\:\lambda{A}=\left\{\left(\begin{array}{c}u,\text{min}\left(\lambda\:\cdot\:{\mu\:}_{A}\left(u\right),1\right),\\\:\text{max}\left(1-\lambda\:\left(1-{\nu\:}_{A}\left(u\right)\right),0\right),\\\:\text{max}\left(1-\lambda\:\left(1-{\xi\:}_{A}\left(u\right)\right),0\right)\end{array}\right)\mid{u}\in\:U\right\}$$

(10)

$$\:A\lambda=\left\{\right(u,\left({\mu\:}_{A}\right(u){)}^{\lambda\:},1-(1-{\nu\:}_{A}\left(u\right){)}^{\lambda\:},1-(1-{\xi\:}_{A}(u\left){)}^{\lambda\:}\right)\mid\:u\in\:U\}$$

(11)

$$\:A^-=\left\{\left(u,{\nu\:}_{A}\left(u\right),{\xi\:}_{A}\left(u\right),{\mu\:}_{A}\left(u\right)\right)\mid{u}\in\:U\right\}$$

(12)

Molecular fuzzy cognitive maps

This method is considered for criteria weighting. First, assessments are obtained, and a decision matrix is created in Eq. (13)⁶².

$$\:{\varsigma\:}_{k}=\:\left[\begin{array}{cccccc}0&\:{\varsigma\:}_{12}&\:\cdots\:&\:&\:\cdots\:&\:{\varsigma\:}_{1n}\\\:{\varsigma\:}_{21}&\:0&\:\cdots\:&\:&\:\cdots\:&\:{\varsigma\:}_{2n}\\\:\vdots&\:\vdots&\:\ddots\:&\:&\:\cdots\:&\:\cdots\:\\\:\vdots&\:\vdots&\:\vdots&\:&\:\ddots\:&\:\vdots\\\:{\varsigma\:}_{n1}&\:{\varsigma\:}_{n2}&\:\cdots\:&\:&\:\cdots\:&\:0\end{array}\right]$$

(13)

Aggregated values can be defined via Eq. (14).

$$\:\varsigma\:=\left(\bigcup\:_{i=1}^{k}{\varsigma\:}_{i}\right)=\left\{\left(u,\frac{1}{k}\sum\:_{i=1}^{k}{\mu\:}_{{\varsigma\:}_{i}}\left(u\right),\frac{1}{k}\sum\:_{i=1}^{k}{v}_{{\varsigma\:}_{i}}\left(u\right),\frac{1}{k}\sum\:_{i=1}^{k}{\xi\:}_{{\varsigma\:}_{i}}\left(u\right)\right)\mid\:u\in\:U\right\}$$

(14)

Fuzzy vectors are computed in the next step by Eq. (15).

$$\:{u}_{i}=\left[\left({\mu\:}_{i1},{v}_{i1},{\xi\:}_{i1}\right),\left({\mu\:}_{i2},{v}_{i2},{\xi\:}_{i2}\right),\dots\:\left({\mu\:}_{i\left(n-1\right)},{v}_{i\left(n-1\right)},{\xi\:}_{i\left(n-1\right)}\right)\right]$$

(15)

Equation (16) is applied to identify the dot product.

$$\:\left({u}_{i}.{u}_{j}\right)=\sum\:_{t=1}^{n-1}\left({\mu\:}_{i,t}\cdot\:{\mu\:}_{j,t}+{\nu\:}_{i,t}\cdot\:{\nu\:}_{j,t}+{\xi\:}_{i,t}\cdot\:{\xi\:}_{j,t}\right)$$

(16)

Moreover, the cosine of the angle is calculated with Eq. (17).

$$\:{\text{c}\text{o}\text{s}(\theta\:}_{{u}_{i},{u}_{j}})=\frac{\left({u}_{i}.{u}_{j}\right)}{\left(\sum\:_{t=1}^{n-1}\left({\mu\:}_{i,t}^{2}+{\nu\:}_{i,t}^{2}+{\xi\:}_{i,t}^{2}\right)\right).\left(\sum\:_{t=1}^{n-1}\left({\mu\:}_{j,t}^{2}+{\nu\:}_{j,t}^{2}+{\xi\:}_{j,t}^{2}\right)\right)}$$

(17)

Equation (18) is taken into consideration to define angle values.

$$\:{\theta\:}_{{u}_{i},{u}_{j}}={cos}^{-1}\:\left({\text{c}\text{o}\text{s}(\theta\:}_{{u}_{i},{u}_{j}})\right)$$

(18)

By using Eqs. (19)-(24), the values can be normalized.

$$\:{norm\left({\theta\:}_{{u}_{i},{u}_{j}}\right)}_{general}=\frac{{\theta\:}_{{u}_{i},{u}_{j}}}{{\theta\:}_{max}}$$

(19)

$$\:{norm\left({\theta\:}_{{u}_{i},{u}_{j}}\right)}_{linear}=\frac{{\theta\:}_{{u}_{i},{u}_{j}}}{\pi\:}$$

(20)

$$\:{norm\left({\theta\:}_{{u}_{i},{u}_{j}}\right)}_{trigona{l}_{planar}}=\frac{{\theta\:}_{{u}_{i},{u}_{j}}}{\frac{2\pi\:}{3}}$$

(21)

$$\:{norm\left({\theta\:}_{{u}_{i},{u}_{j}}\right)}_{tetrahedral}=\frac{{\theta\:}_{{u}_{i},{u}_{j}}}{\frac{\pi\:}{2}}$$

(22)

$$\:{norm\left({\theta\:}_{{u}_{i},{u}_{j}}\right)}_{trigona{l}_{bipyramidal}}=\frac{{\theta\:}_{{u}_{i},{u}_{j}}}{\frac{2\pi\:}{5}}$$

(23)

$$\:{norm\left({\theta\:}_{{u}_{i},{u}_{j}}\right)}_{octahedral\text{}}=\frac{{\theta\:}_{{u}_{i},{u}_{j}}}{\frac{\pi\:}{3}}$$

(24)

The reciprocal values are defined by Eq. (25).

$$\:recip\left({\theta\:}_{{u}_{i},{u}_{j}}\right)=\frac{1}{norm\left({\theta\:}_{{u}_{i},{u}_{j}}\right)}$$

(25)

Furthermore, a normalized matrix is created via Eqs. (26)-(27).

$$\:\text{N}\text{R}=\:\left[\begin{array}{cccccc}0&\:{N}_{12}&\:\cdots\:&\:&\:\cdots\:&\:{N}_{1n}\\\:{N}_{21}&\:0&\:\cdots\:&\:&\:\cdots\:&\:{N}_{2n}\\\:\vdots&\:\vdots&\:\ddots\:&\:&\:\cdots\:&\:\cdots\:\\\:\vdots&\:\vdots&\:\vdots&\:&\:\ddots\:&\:\vdots\\\:{N}_{n1}&\:{N}_{n2}&\:\cdots\:&\:&\:\cdots\:&\:0\end{array}\right]$$

(26)

$$\:{N}_{ij}=\:\frac{recip\left({\theta\:}_{{u}_{i},{u}_{j}}\right)}{\sum\:_{j=1}^{n}recip\left({\theta\:}_{{u}_{i},{u}_{j}}\right)}$$

(27)

Next, state vector values are computed and updated via Eqs. (28)-(30).

$$\:A\left(t\right)=\left[{a}_{1}\left(t\right),{a}_{2}\left(t\right),\dots\:,{a}_{n}\left(t\right)\right]$$

(28)

$$\:A\left(t+1\right)=f\left(A\left(t\right)\cdot\:NR\right)$$

(29)

$$\:f\left(x\right)=\frac{1}{1+{e}^{-x}}$$

(30)

Equation (31) is used to compute the weights.

$$\:{W}_{j}=\:\frac{{fA\left(s\right)}_{j}}{\sum\:_{j=1}^{n}{fA\left(s\right)}_{j}}$$

(31)

Molecular fuzzy multi-objective particle swarm optimization (MF-MOPSO)

MOPSO is the new ranking technique considering molecular fuzzy logic and swarm optimization⁶¹. Equation (32) is used to create an initial matrix.

$$\:{X}_{k}=\:\left[\begin{array}{cccccc}0&\:{X}_{12}&\:\cdots\:&\:&\:\cdots\:&\:{X}_{1m}\\\:{X}_{21}&\:0&\:\cdots\:&\:&\:\cdots\:&\:{X}_{2m}\\\:\vdots&\:\vdots&\:\ddots\:&\:&\:\cdots\:&\:\cdots\:\\\:\vdots&\:\vdots&\:\vdots&\:&\:\ddots\:&\:\vdots\\\:{X}_{n1}&\:{X}_{n2}&\:\cdots\:&\:&\:\cdots\:&\:0\end{array}\right]$$

(32)

The weighted values are identified by Eq. (33).

$$\:{B}_{ij}=\left({w}_{j.}{\mu\:}_{ij},{w}_{j.}{v}_{ij},{w}_{j.}{\xi\:}_{ij}\right)$$

(33)

Fuzzy vectors and dot products can be calculated with Eqs. (34) and (35).

$$\:{y}_{i}=\left[\left({\mu\:}_{i1},{v}_{i1},{\xi\:}_{i1}\right),\left({\mu\:}_{i2},{v}_{i2},{\xi\:}_{i2}\right),\dots\:\left({\mu\:}_{im},{v}_{im},{\xi\:}_{im}\right)\right]$$

(34)

$$\:\left({y}_{i}.{y}_{j}\right)=\sum\:_{t=1}^{n}\left({\mu\:}_{i,t}\cdot\:{\mu\:}_{j,t}+{\nu\:}_{i,t}\cdot\:{\nu\:}_{j,t}+{\xi\:}_{i,t}\cdot\:{\xi\:}_{j,t}\right)$$

(35)

The cosine of the angle and angle values are defined via Eqs. (36) and (37).

$$\:{\text{c}\text{o}\text{s}(\theta\:}_{{y}_{i},{y}_{j}})=\frac{\left({y}_{i}.{y}_{j}\right)}{\left(\sum\:_{t=1}^{m}\left({\mu\:}_{i,t}^{2}+{\nu\:}_{i,t}^{2}+{\xi\:}_{i,t}^{2}\right)\right).\left(\sum\:_{t=1}^{m}\left({\mu\:}_{j,t}^{2}+{\nu\:}_{j,t}^{2}+{\xi\:}_{j,t}^{2}\right)\right)}$$

(36)

$$\:{\theta\:}_{{y}_{i},{y}_{j}}={cos}^{-1}\:\left({\text{c}\text{o}\text{s}(\theta\:}_{{y}_{i},{y}_{j}})\right)$$

(37)

Normalization steps are performed by Eqs. (38)-(43).

$$\:{norm\left({\theta\:}_{{y}_{i},{y}_{j}}\right)}_{general}=\frac{{\theta\:}_{{y}_{i},{y}_{j}}}{{\theta\:}_{y\_max}}$$

(38)

$$\:{norm\left({\theta\:}_{{y}_{i},{y}_{j}}\right)}_{linear}=\frac{{\theta\:}_{{y}_{i},{y}_{j}}}{\pi\:}$$

(39)

$$\:{norm\left({\theta\:}_{{y}_{i},{y}_{j}}\right)}_{trigona{l}_{planar}}=\frac{{\theta\:}_{{y}_{i},{y}_{j}}}{\frac{2\pi\:}{3}}$$

(40)

$$\:{norm\left({\theta\:}_{{y}_{i},{y}_{j}}\right)}_{tetrahedral}=\frac{{\theta\:}_{{y}_{i},{y}_{j}}}{\frac{\pi\:}{2}}$$

(41)

$$\:{norm\left({\theta\:}_{{y}_{i},{y}_{j}}\right)}_{trigona{l}_{bipyramidal}}=\frac{{\theta\:}_{{y}_{i},{y}_{j}}}{\frac{2\pi\:}{5}}$$

(42)

$$\:{norm\left({\theta\:}_{{y}_{i},{y}_{j}}\right)}_{octahedral\text{}}=\frac{{\theta\:}_{{y}_{i},{y}_{j}}}{\frac{\pi\:}{3}}$$

(43)

Equations (44) and (45) are utilized to compute reciprocal values and the final decision matrix.

$$\:recip\left({\theta\:}_{{y}_{i},{y}_{j}}\right)=\frac{1}{norm\left({\theta\:}_{{y}_{i},{y}_{j}}\right)}$$

(44)

$$\:{f}_{ij}=\:\frac{recip\left({\theta\:}_{{y}_{i},{y}_{j}}\right)}{\sum\:_{j=1}^{n}recip\left({\theta\:}_{{y}_{i},{y}_{j}}\right)}$$

(45)

The velocity, the updated personal and global best positions are determined via Eqs. (46)-(53).

$$\:{X}_{i}=\left\{{x}_{i1},{x}_{i2},\dots\:,{x}_{in}\right\}$$

(46)

$$\:{V}_{ij}(t+1)=\omega\:{V}_{ij}\left(t\right)+{c}_{1}{r}_{1}\left({P}_{ij}\right(t)-{X}_{ij}(t\left)\right)+{c}_{2}{r}_{2}\left({P}_{{gb}_{j}}\right(t)-{X}_{ij}(t\left)\right)$$

(47)

$$\:{V}_{ij}\left(1\right)=\partial\:\times\:\left({P}_{{max}_{i}}-{P}_{{min}_{i}}\right)\times\:r$$

(48)

$$\:{X}_{ij}(t+1)={X}_{ij}\left(t\right)+{V}_{ij}(t+1)$$

(49)

$$\:{P}_{ij}(t+1)=Best\left({P}_{ij}\left(t\right),\:{X}_{ij}(t+1)\right)$$

(50)

$$\:{P}_{{gb}_{j}}(t+1)=Best\left({P}_{{gb}_{j}}\left(t\right),\:\left\{{X}_{ij}(t+1)\right\}\:all\:i\right)$$

(51)

$$\mid{P}_{{gb}_{j}}\left(t+1\right)-{P}_{{gb}_{j}}\left(t\right)\mid<\epsilon$$

(52)

$$\:\frac{1}{n}\sum\:_{i=1}^{n}{P}_{ij}$$

(53)

Q-learning algorithm for unbalanced expert evaluations

The experts’ weights are computed in this study using a Q-learning algorithm. The reward and penalty are applied to compute these degrees.

$$\:{R}_{s,a}=r\cdot\:({Q}_{s,a\:\left(bestDM\right)}-{Q}_{s,a\:\left(otherDM\right)})$$

(54)

$$\:{P}_{s,a}=p\cdot\:\left({Q}_{s,a\:\left(otherDM\right)}-{Q}_{s,a\:\left(bestDM\right)}\right)$$

(55)

The updated Q values are defined via Eq. (56).

$$\:{Q}_{s,a}^{{\prime\:}}={Q}_{s,a\:\left(bestDM\right)}+\alpha\:\cdot\:({R}_{s,a}-{P}_{s,a})$$

(56)

The optimal Q values are identified by Eqs. (57) and (58).

$$\:{{\Delta\:}}_{s,a}=\mid\:{Q}_{s,a}^{{\prime\:}}-{Q}_{s,a}\mid\:$$

(57)

$$\:{{\Delta\:}}_{max}=\underset{s,a}{\text{m}\text{a}\text{x}}\left\{{{\Delta\:}}_{s,a}\right\}$$

(58)

Case study

The research aims to identify the critical front-of-packaging cues essential for a sustainable product’s packaging design and select the best AI-generated alternative. A reinforcement learning-based molecular fuzzy decision-making method is adopted to facilitate the ranking of front-of-packaging design criteria and packaging design selection processes. A case study has been developed to verify the applicability of the methodological framework. In parallel with the research objective, the case study primarily concentrates on designing AI-generated gift box alternatives for a new textile company that has recently been operating in the textile sector and selecting the best alternative.

This family-owned textile company, Baby-Eco-Natural Ltd. Co., is based in Afyonkarahisar, Türkiye, and recently started to produce sustainable and organic clothes for newborn babies. This case study was conducted to meet their requirement for a gift box design that reflects their unique values of environmental sustainability, organic practices, and a focus on newborn health. This case study was developed in the following stages.

In the first stage, keywords characterizing the company’s features were identified through a search and collection of information about the textile company that produces organic and ecologically sustainable products for newborn babies. The twenty-two keywords identified to reflect the company are: “Family-owned business, New born Babies, Baby Clothes, Quality, Environmentally-friendly, Ecologically-sustainable, Design, Trendy, Fashionable, Conscious, Sensitive, Accountable, Reliable, Pure, Soft, Natural, Organic, Organic Cotton, Organic Bamboo, Customer Satisfaction, Happy Babies, Happy Parents”.

In the following stage, a text that describes the name and primary features of the textile company was prepared. Then, based on the previous findings in the extant literature on front-of-packaging cues for sustainable products, keywords for packaging design criteria for each alternative were determined. Afterward, to effectively apply the MCDM model, five gift box alternatives were designed for the company by using ChatGPT to create visual images. The texts used in ChatGPT are given below:

• Text: “A gift box design for Baby-Eco-Natural Ltd. Co., a family-owned textile company producing organic and sustainable clothes for new born babies, defined with the key words < Family-owned business, New Born Babies, Baby Clothes, Quality, Environmentally-friendly, Ecologically-sustainable, Design, Trendy, Fashionable, Conscious, Sensitive, Accountable, Reliable, Pure, Soft, Natural, Organic, Organic Cotton, Organic Bamboo, Customer Satisfaction, Happy Babies, Happy Parents >”.

• Text for Alternative 1: “The first gift box alternative for Baby-Eco-Natural Ltd. Co., a family-owned textile company producing organic and sustainable clothes for new born babies, defined with the key words < A Round and Tall Gift Box, 16 cm diameter, 18 cm depth, in pastel color tones such as light blue and white, the cute animal and nature images on the package, the Baby Eco Natural brand name and logo repeated several times on the package, the Baby Eco Natural brand name made from natural rope and wood added to the side of the box, the slogan “Sustainable Choice” centered at the front side of the package>”.

• Text for Alternative 2: “The second gift box alternative for Baby-Eco-Natural Ltd. Co., a family-owned textile company producing organic and sustainable clothes for new born babies, defined with the key words < A Heart-shaped gift box, 25 cm wide, 10 cm deep, in natural beige tones with organic cotton and bamboo images and an elegant ribbon at the top, the Baby Eco Natural brand name and logo given on the top, the slogan “Natural and Pure” given below the brand name and logo>”.

• Text for Alternative 3: “The third gift box alternative for Baby-Eco-Natural Ltd. Co., a family-owned textile company producing organic and sustainable clothes for new born babies, defined with the key words < A Square Tall Gift Box, 15 × 15 cm wide, 20 cm deep, in light pastel tones such as light yellow, green and white, covered with baby-figured and environmentally friendly images, the Baby Eco Natural brand name, logo and the slogan “Sensitive to Nature” repeated in a cheerful font at every side of the box>”.

• Text for Alternative 4: “The fourth gift box alternative for Baby-Eco-Natural Ltd. Co., a family-owned textile company producing organic and sustainable clothes for new born babies, defined with the key words < A Round Gift Box, 18 cm diameter, 10 cm depth, in soft tones such as light green and beige, covered with bamboo leaf images, the Baby Eco Natural brand name and logo and the slogan “Happy Babies Happy Parents” centered on the top in a legible font>”.

• Text for Alternative 5: “The fifth gift box alternative for Baby-Eco-Natural Ltd. Co., a family-owned textile company producing organic and sustainable clothes for new born babies, defined with the key words < A Square Gift Box, 40 × 40 cm wide, 15 cm deep, in soft tones such as white and yellow, Bamboo leaves and baby footprints images on the package, The Baby Eco Natural brand name and logo centered on the top, adorned with a simple, light yellow bow>”.

All five AI-generated gift box design alternatives are demonstrated in Fig. 2.

Fig. 2

AI-generated gift box alternatives.

In the next stage, a decision method framework was established to rank key front-of-packaging design elements that must be considered in the package design process and select the package alternative among the AI-generated alternatives. The proposed decision method of the research includes three experts, four gift box design criteria, and five AI-generated gift box alternatives. After creating the data sets based on experts’ evaluations regarding gift box design criteria and AI-generated gift box alternatives via face-to-face interviews, the data were analyzed using a Q-learning algorithm and a molecular fuzzy sets-based decision-making methodology. Finally, the optimal AI-generated gift box was selected for organic and sustainable baby clothes for the textile company, considering the outcome of applying the case study.

In the final stage, a sensitivity analysis was performed to indicate the robustness of the findings acquired by the different types of molecular geometry and learning rates. Information about crucial front-of-packaging criteria to be used in the design of the gift box for the textile company, along with the identification of experts who evaluate these criteria, is given in the following subheadings.

Front-of-packaging criteria for designing a gift box

When selecting the most appealing gift box, it is vital to consider whether the packaging design is good in terms of visual and verbal design elements, along with reflecting the main characteristics of the company and its products. Thus, based upon a thoughtful review of the literature on attributes of package design and considering the methodological framework and ChatGPT’s ability to create visual content, four design criteria are identified as evaluation criteria to be used in the selection of the best gift box for a textile company with sustainable products. These criteria are as follows:

Form and dimension

The shape and size of the packaging have a significant influence on consumers’ judgments about the product and their buying decisions and behaviors⁶³. Research on the shape of the packaging reveals that circular shapes are more likely to be perceived as less sustainable⁴⁴, whereas packaging with larger sizes is perceived as less sustainable due to the use of more packaging material⁴⁵. Thus, in this research, form and dimension refer to the attractiveness of the gift box in terms of its shape and size.

Aesthetic appeal

Visual aesthetics, including style and attractiveness, affect consumers’ preferences, such as their willingness to pay more, evoking their emotions, such as excitement³. Herein, since it has the potential to transmit several meanings to consumers within the context of aesthetic appeal, color can be one of the most noticeable cues of front-of-packaging⁶⁴. Whereas the color of the packaging draws the attention of consumers, it simultaneously communicates the product’s characteristics, such as naturalness⁴⁷ and sustainability⁴⁵. Besides, packaging styles can enable consumers to form several impressions regarding the product’s sustainability⁶⁵. Thus, in this research, aesthetic appeal refers to the attractiveness and distinctiveness of the gift box for its color and style.

Brand identity

Verbal cues on packaging can explicitly communicate the sustainability of a product⁶⁶. Herein, verbal cues on packaging can be brand elements, such as brand name, logo, slogan, and typography, which constitute brand identity¹. A verbal claim demonstrating sustainability should be direct and easy to understand regarding target consumers⁶⁷. Thus, in this research, brand identity refers to typography, brand name, logo, and slogan used on the gift box that communicate the brand’s value to target customers.

Imagery

Incorporating imagery on the package can capture consumers’ attention⁶⁸. On the other hand, the color and shape of imagery, together with its content and location on the package, can influence the sensory expectations of consumers by communicating meaning about the product⁴⁶. A package with imagery in the form of sustainable-looking graphics is more likely to be related to sustainability⁴⁰. Thus, in this research, imagery refers to graphics regarding objects and figures on the gift box, reflecting the brand’s commitment to sustainability.

Introduction of experts

A group of experts was selected to rank the critical visual design element criteria priorly essential for a gift box design of the textile company producing sustainable and organic baby clothes and select the optimum AI-generated package design for that company. The group of experts includes three experts with different levels of expertise and know-how. One of the experts is an associate professor of marketing with fifteen years of experience, and she has several academic studies on brand element selection. The second expert is a product manager in a textile company with five years of experience. The last expert is a brand manager with eleven years of experience, specializing in selecting brand elements, including packages.

Analysis results

The hybrid MF-MOPSO and Q-Learning model produced a comprehensive evaluation of five design alternatives based on nine criteria. To ensure interpretive clarity, only the key summary tables are presented in the main text; detailed fuzzy matrices and iteration data are provided in Appendix A.

Constructing the balanced expert dataset

Assessments are obtained from the experts. In this framework, the values in Table 1 are considered.

Table 1 Assessment set and fuzzy numbers.

The four criteria are defined as form and dimension (CRT1), aesthetic appeal (CRT2), brand identity (CRT3), and imagery (CRT4). Assessments of the criteria are given in Table 2.

Table 2 Assessments of the criteria.

Five alternatives are identified: Alternative 1 (ALTR1), Alternative 2 (ALTR2), Alternative 3 (ALTR3), Alternative 4 (ALTR4), and Alternative 5 (ALTR5). Table 3 presents their assessments.

Table 3 Assessments of the alternatives.

The demographic information of experts is explained in Table 4.

Table 4 Demographic information of experts.

The weights are computed in Table 5.

Table 5 Weights of experts.

Table 5 demonstrates that Expert 1 has the highest weight, with 0.393. Moreover, Expert 2 has the second-highest weight (0.318). Additionally, Expert 3 weighs 0.289. Reward degrees of the criteria are given in Table A1.

Reward degrees of the alternatives are shown in Table A2. Penalty degrees of the criteria are given in Table A3. Penalty degrees of the alternatives are given in Table A4. Updated values for criteria and alternatives are shown in Tables A5 and A6. Absolute differences regarding criteria and alternatives are indicated in Tables A7 and A8. Iterative absolute differences for different combinations of experts are denoted in Tables A9-A12. Fuzzy Q values of the criteria and alternatives are denoted in Tables 6 and 7.

Table 6 Fuzzy Q values for criteria.

Table 7 Fuzzy Q values for alternatives.

Weighting the criteria with molecular fuzzy cognitive maps

Selected indicators are weighted in the second stage. Average values are computed in Table 8.

Table 8 Averaged values.

Fuzzy vectors are given in Table 9.

Table 9 Fuzzy vectors.

Dot products are computed in Table A13. Cosine of angle values are denoted in Table A14. Angle in radians is defined in Table A15. Table A16 demonstrates the normalized values. Reciprocal values are indicated in Table A17. The normalized relation matrix is created in Table 10.

Table 10 Normalized relation matrix.

State vectors are defined in Table 11.

Table 11 Iterative state vectors.

Table 11 defines CRT3 as the most essential item. CRT2 and CRT1 also play a significant role in this regard. Table 12 focuses on the comparative results according to different molecular geometry shapes.

Table 12 Comparative prioritization results.

Table 12 shows that the weighting results of the criteria are the same in the different types of molecular geometry. Moreover, the results are coherent for the various types of learning rates.

Ranking the alternatives with MF-MOPSO

Selected alternatives are ranked by MF-MOPSO. Table 13 indicates the averaged values.

Table 13 Averaged numbers.

The weighted matrix is constructed in Table 14.

Table 14 Weighted matrix.

Fuzzy vectors are defined in Table 15.

Table 15 Fuzzy vectors.

Dot products are demonstrated in Table A18. Cosine of the angle is identified in Table A19. Angle in radians is shown in Table A20. The values are normalized in Table A21. Reciprocal values are computed in Table A22. The final decision matrix is generated in Table 16.

Table 16 Final decision matrix.

The velocity and the updated personal and global best positions are demonstrated in Table A23. Table A23 shows the velocity and the personal and global best positions for several rounds. As a result, the ranking results are given in descending order as ALTR3, ALTR2, ALTR1, ALTR4, and ALTR5. Comparative results are denoted in Table 17.

Table 17 Comparative ranking results.

The analysis indicates that Brand Identity holds the highest influence on decision outcomes, followed by Sustainability and Visual Appeal. Among the five design alternatives, Alternative 3 achieved the top overall performance index, confirming its superiority in balancing functional efficiency and visual coherence. This finding implies that consumers prioritize the emotional and communicative value of the package, how well it reflects the brand’s image, over purely physical aspects such as cost or material durability. To further validate the stability of the decision model, a sensitivity analysis was performed. The results, summarized in Table 17, demonstrate that calculations with the different geometrical conditions do not alter the ranking of the alternatives, confirming the robustness of the hybrid framework.

Once the Brand Identity weight increased, the optimal ranking remained unchanged, emphasizing that branding elements, such as logo visibility, color scheme, and typography, serve as non-negotiable anchors of consumer perception. Adjustments in Material Cost or Sustainability weights slightly affected preference scores but did not change the final decision, indicating that moderate cost trade-offs can be accommodated without compromising consumer satisfaction.

In contrast, small deteriorations in Visual Appeal (e.g., cluttered layout or color inconsistency) rapidly decreased the satisfaction index, signaling the need for visual uniformity across all product variants. In essence, the hybrid model helps designers identify where flexibility exists (e.g., minor cost/material shifts) and where strict consistency is essential (brand and visual identity).

Discussion

This research offers a novel perspective on the packaging design process by ranking the importance of key front-of-packaging cues and selecting the optimal packaging design for sustainable products. It employs an AI-generated approach for packaging design and a reinforcement learning-based molecular fuzzy decision-making method. Based on a comprehensive review of relevant research on front-of-packaging cues communicating the sustainability of the products, vital front-of-packaging cues are identified as form and dimension, including shape and size⁴³, aesthetic appeal, including color and style⁵¹, brand identity, including brand name, logo, symbol, and slogan, as well as typography¹, and finally imagery. Regarding the overall evaluation of experts, brand identity (0.2509) is the foremost front-of-packaging cue that must be considered in packaging design for sustainable products. Previous research in prevailing literature also reveals that, as a fifth “p” of marketing, a good packaging design should saliently communicate brand values⁶ and brand personality⁷ and establish a unique brand image⁶⁹, which in turn leads to strong brand identity¹. In the packaging design process, it is essential to effectively place brand names, logos, slogans, product information, and labels on the packaging⁴⁵. Whereas communicating brand identity through packaging, how all brand elements and sustainability claims are visualized in typefaces is another critical issue⁵⁶. Accordingly, to shape customers’ perception of the sustainability of the product, the brand name, logo, slogan, product information, and typography should be primarily considered.

Based on the overall ranking of experts, aesthetic appeal (0.2504) is identified as the second crucial front-of-packaging cue essential for packaging design to communicate the sustainability of products. Once the influence of color in shaping consumers’ perception of the sustainability of the products is considered⁵¹, color is unsurprisingly found to follow the foremost front of package cues. By using appropriate colors on packaging, the intended meanings of products, such as sustainability⁴⁵, eco-friendliness⁴⁸, or naturalness⁴⁷, can be communicated to customers. Therefore, in the packaging design process for sustainable products, deciding on the packaging color should be another crucial consideration for designers to convey the intended brand meaning.

The following important cue is determined as form and dimension (0.2500). Shape and size are two of the first front-of-packaging cues that make customers easily recognize the brand on the shelves¹⁵ and influence their perception of sustainability⁶⁶. Herein, the decrease in packaging shapes and sizes is associated with sustainability⁴³. Thus, to communicate the sustainability of the products, it would be better to use smart shapes and small sizes in packaging design. On the other hand, if there are many small packages, oversized packages should be used instead, reducing the use of packaging material. Finally, imagery (0.2488) is the fourth crucial front-of-packaging cue essential for packaging design for sustainable products. The imagery incorporated in product packaging can also effectively communicate information about the product and brand to customers and shape their perceptions of the product⁵. Imagery on packaging is an effective tool to communicate intended brand messages regarding low-involvement products with customers who have time constraints¹. Herein, the product in the case study is a high-involvement product that requires parents to provide detailed information for their babies. So, it is unsurprising to find imagery as the last front-of-packaging cue for such high-involvement products.

Regarding the overall evaluation of experts, ALTR3 is selected as the best AI-generated gift box design for the relevant textile company. Regarding the brand identity criterion, ALTR3 is the best AI-generated gift box design, including a salient brand name and a slogan with “Sensitive to Nature” as a sustainability claim. Moreover, previous research indicates that to emphasize environmental claims, typefaces, such as serif and sans-serif with a hand-written appearance, can be used⁷⁰. The typeface used on ALTR3 is also very similar to sans-serif with a handwritten appearance. Previous research also reveals that green color on packaging is mostly associated with sustainability^51,66 and imagery with sustainability associations, such as green leaves⁵⁸. ALTR3 includes imagery, such as flowers and green leaves, as well as a happy baby, communicating the core values of the relevant brand. Additionally, the dominant colors on the gift box are cream and green, which are highly associated with sustainability. In terms of shape and size, small packaging and packaging shapes, such as square and rectangular, are more related to sustainability in terms of transportation efficiency and less use of packaging material⁴³. Concerning size and shape, ALTR3 is also the best AI-generated gift box design. ALTR3 is followed with ALTR2, ALTR1, ALTR4, and ALTR5, respectively.

Theoretical implication

This study combines fuzzy logic, cognitive mapping, optimization, and machine learning into a cohesive framework, addressing decision-making complexities more comprehensively than isolated methods. Additionally, the molecular fuzzy approach offers a robust mechanism to handle imprecision, outperforming traditional scalar-based models in precision and reliability⁷¹.

Outside methodological novelties, this research contributes to existing literature in some aspects. First, though there are several qualitative¹⁸ and quantitative^12,15,72 studies on understanding front-of-packaging cues communicating sustainability from the perspective of consumers, none of them approach packaging design from the perspective of companies. Ranking and prioritizing the critical front-of-packaging cues that must be considered in the packaging design process to communicate the sustainability of the products, this research brings a new perspective to the packaging design literature. Second, most of the research on front-of-packaging cues has been well-documented for food products in extant literature^40,50,66. So, it is the first time front-of-packaging cues are examined within the context of gift box design. Finally, apart from the existing research on front-of-packaging cues, this research actively integrates AI in the packaging design process for sustainable products by presenting the applicability of AI in the packaging design process. Thus, this research also contributes to the extant literature on the AI-generated packaging design process.

Practical implication

In capturing customers’ attention in a retail environment saturated with competitive brands, it is vital to design appealing product packaging to communicate the intended meaning. Prioritizing critical front-of-packaging cues and selecting the optimum AI-generated packaging design through a group decision-making model with a case study, this research has led to several contributions to marketing practice. First, this research prioritizes the key front-of-packaging cues that designers and marketers should concentrate on in the packaging design process for sustainable products. Herein, adopting appropriate typography, such as sans-serif, with a hand-written appearance to communicate the brand name, slogan, product, brand values, and sustainability claims to customers should be the primary consideration of designers and marketers. Then, they should decide on the color of the packaging, which explicitly communicates the inability. Herein, using colors such as green, brown, and cream could enhance the appeal of the product inside the packaging. Then, designers and marketers should use rectangular and square shapes and small sizes, reducing the use of packaging material and increasing efficiency in transportation and storage to enable customers to judge the products as sustainable. Finally, designers and marketers should choose the right imagery associated with sustainability and nature to communicate their sustainability claims.

Second, this research selects the best gift box design for the textile company and its sustainable products, showing the applicability of AI in the packaging design process to generate more creative design alternatives with time and cost efficiency. Converting text to image, AI supports the creation of packaging alternatives that align with company characteristics, brand identity, and product value proposition. Hence, this research presents that AI-supported packaging design would be a cost- and time-effective alternative to the traditional packaging design process. Finally, this research also indicates that in designing their product packaging, companies could create multiple packaging design alternatives via AI and select the best alternative that reflects their core values using the recommended MCDM method, depending on expert opinions.

Conclusion

This study proposed and validated a hybrid decision-making framework integrating Multi-Factor Multi-Objective Particle Swarm Optimization (MF-MOPSO) and Q-Learning for packaging design evaluation under uncertainty. The combined model successfully optimized the importance of design criteria while dynamically adapting decision policies through reinforcement learning. The empirical results demonstrated that the model effectively identifies the most preferred packaging alternative by balancing quantitative optimization with adaptive learning. This research brings a company-centered approach to determining critical front-of-packaging cues for gift box design of sustainable textile products and selecting the best AI-generated gift box design alternative for them. Besides, this research adopts molecular fuzzy sets and reinforcement learning-based decision-making methodology to mitigate the methodological uncertainty of the existing research. Regarding the overall evaluation of experts, brand identity, including brand elements, typography, and aesthetic appeal of the packaging in terms of color, becomes the essential front-of-packaging cues for sustainable textile products, whereas form dimension and imagery are the third and fourth front-cues essential in packaging design, respectively. Further, ALTR3 is selected as the best AI-generated gift box design alternative that aligns with the core values of the company, brand, and product, followed by ALTR2, ALTR1, ALTR4, and ALTR5.

Beyond the packaging domain, the proposed hybrid methodology possesses significant scalability. Its structure, consisting of parallel optimization (MF-MOPSO) and adaptive learning (Q-Learning), can be extended to other design-intensive sectors such as automotive styling, consumer electronics, or architectural design, where trade-offs between aesthetics, functionality, and sustainability are equally critical. In these fields, the method can serve as a decision-support engine, optimizing design variables and learning from user or market feedback to enhance performance over iterative cycles.

The main limitation of the work is that the criteria weights and assessments rely on the judgments of a small group of experts. Due to the dizzying advances in artificial intelligence, package designs are constantly improving. Consequently, some criteria considered crucial today may become less relevant in the future. Future research could rank and prioritize front-of-packaging cues for low-involvement products, such as food packaging, to extend the findings. Second, future research could explore prioritizing key front-of-packaging cues and selecting the best AI-generated alternatives for other packaging types, such as cans, bottles, and bags, rather than focusing on gift boxes. Third, per front-of-packaging cues, the packaging material influences consumers of the product. It could also explore integrating this hybrid analytical approach with generative AI technologies, including DALL·E, Midjourney, and similar diffusion-based design models. Such integration would enable automated concept generation followed by quantitative optimization, creating a closed-loop design intelligence system. For example, generative AI could produce hundreds of visual packaging prototypes aligned with certain brand cues, while the hybrid MF-MOPSO–Q-Learning framework could evaluate, refine, and select the most promising designs based on multi-criteria performance and user feedback. However, the packaging material is ignored in this research since the texture and material of the packaging could not be easily understood by AI-generated packaging design alternatives. Thus, future research would also examine other front-of-packaging cues, such as packaging material and texture, when the importance of front-of-packaging cues is prioritized. Besides its theoretical and practical contributions, this research also has some limitations. First, this research concentrates on designing a gift box for a textile company producing organic, eco-friendly, and sustainable clothes for newborn babies. Accordingly, this research mainly prioritizes front-of-packaging cues for high-involvement products and selects the best packaging design alternatives. There are also some limitations regarding the proposed methodology of the research. For that, using multiple geometric normalization techniques (e.g., trigonal planar, tetrahedral, etc.) adds flexibility. However, it may require domain-specific expertise to select the most appropriate model, increasing complexity in practical applications.