Technological evolution has changed life for all people, achieving greater efficiency, products and services that were not previously available to users, thereby improving conditions and the quality of life in general (Funk et al., 2018; Chen et al., 2021). People use electronic tools on a daily basis, which in turn provide different Applications (hereafter Apps) that make it possible to access various services and products simply by registering (Sturm et al., 2018; Tao and Edmunds, 2018). For example, using Apps to move around the city (Uber, Cabify, car sharing Apps) records information about where a person moves, shopping Apps (supermarkets, clothes shops, make-up shops, etc.) store information about a person’s tastes and preferences, or simply searching for a term in a search engine stores the searched information. On the basis of these searches and purchases a database is created for a specific person and all adds or offers then take into account the previously collected information (Goldenberg et al., 2012; Fainmesser et al., 2019). Because of this, companies have the possibility to design specific offers for their customers (Gelashvili et al., 2021), but the big question is whether this affects or not user’s privacy and security on digital platforms (Saura et al., 2021a,b) since, actually more and more personal information is being processed, shared or disseminated by companies (Ribeiro-Navarrete et al., 2021).

Therefore, one of the drawbacks of this technological advance is a lack of privacy, in particular digital privacy (Munuera, 2017). Privacy is understood as a matter of controlling one’s own data, including information disclosure (Sarikakis and Winter, 2017). Study elaborated by Sarathy and Robertson (2003) indicates that individualized information based on users’ activities on online platforms allows companies to offer personalized marketing, but this may cause an alarm about the private data of individuals.

The problem of data privacy is particularly visible in social networks (Kayes and Iamnitchi, 2017). Increasingly, users face a lack of control over their information on social networks as they cannot control what companies or other users do with the information they share online. This is why the control of individual users’ information has become more difficult (Trepte, 2020). As stated by Beigi and Liu (2018), the daily use of social networks by users has generated an enormous amount of user-generated data that is used by researchers and service providers being an opportunity for them, but that in turn creates a risk of exposing individuals’ privacy. According to Fainmesser et al. (2019) in recent years, there has been growing concern about the privacy and security of users’ data, for this reason governments have designed and implemented new laws and regulations that do not provide 100% protection but are partially effective at best.

Nowadays, the accessibility of online platforms has enabled users to change their opinions about products and services. On the basis of these opinions, many customers decide whether or not to consume a service and purchase a product (Jiménez-Castillo and Sánchez-Fernández, 2019; Nuseir, 2019). This method of exchange or expression of consumer-generated opinions on the Internet is called E-WOM. Usually these opinions are addressed to other users. One of the first definitions of E-WOM (Hennig-Thurau et al., 2004) states that E-WOM is any positive or negative statement made by potential, actual or former customers about a product or company, which is made available to a multitude of people and institutions via the Internet. E-WOM has a major impact on both businesses and their consumers (López and Sicilia, 2014). Through social networks and electronic channels consumers can express their opinion which is visible to the rest of the users, if these comments are negative other consumers will avoid buying or acquiring those products and services which will generate a decrease in sales for the companies (Sa’ait et al., 2016; Li et al., 2019). In addition to this, if companies will not take into account the importance of E-WOM for businesses, there will be a high probability with customer satisfaction, loyalty and trust (Rialti et al., 2017; Duarte et al., 2018; Mishra and Singh, 2019; Tran and Strutton, 2020).

According to Reyes-Menendez et al. (2019) there are different social network sites or platforms like Facebook, Instagram, Twitter, Amazon, Expedia or TripAdvisor that can be useful to inform about the features or benefits of products and services. With a special focus on TripAdvisor, it can be considered as one of the most important E-WOM platforms in the tourism sector. The comments and opinions generated by users generate trust in other users, for this reason it is necessary to manage correctly the content on platforms since there may be some false opinions that can affect the reputation of companies.

But E-WOM is not only in the tourism sector that has an importance. Study elaborated in the banking sector have shown that E-WOM is positively related to loyalty, trust and purchase intention (Guping et al., 2021; Zhang et al., 2021). Another study elaborated by Tjhin and Aini (2019), has found that E-WOM has a significant effect on purchase decision and consumer trust in clothing industry. This means that in almost all industries E-WOM has an important significance. For this companies need to be aware of the huge impact E-WOM has on consumers and to avoid problems they need to pay special attention to customer satisfaction and complaints (López and Sicilia, 2014).

Nowadays user satisfaction plays an important role for the performance and growth of the companies (Eklof et al., 2020; Otto et al., 2020). User satisfaction is defined as the users’ opinion of the service or products received compared to the service or products expected (Jani and Han, 2011). This means that if the customer has expectations of receiving a product or service of a certain quality and when they receive these products and services in line with expectations, satisfaction is generated. Therefore, through customer satisfaction companies can know the tastes and preferences of their customers (Alsini, 2017).

Through satisfaction it is possible to study the trust and loyalty generated by users toward the company or products. According to Gelashvili et al. (2021) when using mobile Apps for restaurant reservation has a positive and direct impact on users’ trust in restaurants. Likewise, researches carried out on tourism companies have shown that when the user of these companies is satisfied with the services, trust is generated (Lai, 2014; Martínez-Navalón et al., 2019). In addition to trust, another variable that has been studied along with satisfaction by several studies is loyalty (El-Adly, 2019; Thamrin et al., 2020). These studies have shown that user satisfaction is positively related to user loyalty toward the brand. In other words, if satisfaction is generated toward a product or service, this positively affects the loyalty of customers, who will continue to buy the same products and services. Therefore, we can say that these two variables are important for companies to be taken into account. Apart from these two variables, the direct and positive relationship of customer satisfaction with the variables like continuance intention or service quality among others has been studied (Chumpitaz and Paparoidamis, 2007; Liang et al., 2018; Hossain et al., 2019).

In short, user satisfaction is a key factor for companies, because a satisfied customer is a customer who intends to repurchase, trusts the brand and is loyal to the company’s products and services. This implies a long term relationship between the customer and the company.

The variable trust is one of the most important variables in analyzing customer behavior and has been the subject of research over the decades (Swan et al., 1985; Komiak and Benbasat, 2004; Leninkumar, 2017; Al-Khalaf and Choe, 2020). According to McKnight and Chervany (2001) it is not easy to define the term trust, as most definitions are confusing and for this reason some researchers have chosen not to go further in defining the concept. This may be because each discipline defines and analyses trust from its own perspective. In social sciences customer trust is referred to as the willingness of the consumer to rely on a service brand because he or she has confidence in the reliability and integrity of that brand (Zhang and Bloemer, 2008). In other words, trust could be defined as the users’ confidence or faith in the brand’s products and services. According to Martínez-Navalón et al. (2019) a customer who has trust in the brand’s products and services could help the company improve its profitability and achieve other business goals. Another advantage for companies is that the trust variable creates a stable and collaborative user-company relationship (Bilgihan and Bujisic, 2015). As we can see, it is essential for companies to create the trust of their users toward their products and services, which will help to improve their economic and financial situation as well as competitiveness in the market.

Study carried out by Aurifeille et al. (2009), on the measurement of trust indicates that it is not easy to measure thisvariable since there are several studies that analyze customer trust based on various dimensions, among them it can be seen the following dimensions: reliability, credibility, honesty, benevolence, expectation or confidence among others. Lassala et al. (2010) to measure customer trust, uses the following three dimensions: competence, benevolence and honesty, where honesty is defined as the belief that the other party will fulfill the promises and obligations undertaken. Competence refers to the professionalization of companies in terms of promised quality, know-how and ability to do their job well. And finally, the dimension of benevolence refers to the belief that the other party is interested in achieving joint benefits and will make decisions that will be beneficial to both parties. Several recent studies (Martínez-Navalón et al., 2019; Gelashvili et al., 2021) have supported the use of these three dimensions to measure customer trust toward a company’s products or services.

The objective of this study was to study users who use the TripAdvisor website with the aim of finding managerial implications for the company’s future strategy approach. Therefore, it analyses how privacy and E-WOM affect the satisfaction and trust of users of this review website in tourism and hospitality.

In order to carry out this study, an online questionnaire data source was used. This questionnaire has had a self-administered format that has collected the opinion of users of this Apps living in central Spain. Specifically in the regions of Castilla-La Mancha and Madrid. Focusing on localities such as Albacete, Chinchilla de Montearagón, Guadalajara, Cuenca, Madrid, Chinchón, Arroyomolinos, Móstoles, Fuenlabrada, etc. This sample is motivated by the intention of finding out the opinion of the population in the interior of Spain. Data collection took place during the months of May and June 2021. In order to have a greater veracity of the data, the IP of the users was checked, avoiding that the same user fills in the form twice.

The data collection process was divided into two stages. In the first stage, a sample of 50 questionnaires was collected in order to carry out a pre-test. Once it was verified that the measurement scale was well done, the second part was developed, which was the collection of the majority of data, where 400 questionnaires were obtained. Of these sample, 390 were valid.

The questionnaire used in this study was divided into two parts. In the first part, questions were asked to classify the users. For this classification, questions were asked about gender, employment status, age and education. For the second part of the questionnaire, a measurement scale was used, based on the bibliographical analysis carried out in this study. It analyses the variables privacy, E-WOM, satisfaction and trust. It should be noted that the trust variable is a multidimensional variable, so an analysis by dimensions (Benevolence, Honesty and Competence) had to be carried out. A total of 19 questions were asked to measure these variables.

The type of questions used were Likert scale questions from 0 to 6 (1 = “strongly disagree”; 6 = “strongly agree”). The choice of these types of questions is motivated by the fact that they are able to analyze the degree of sensitivity of each user and are the most commonly used type of question in social science studies (Fernández-Fernández et al., 2021).

In order to carry out the validation of the measurement scale and the validation of the hypotheses put forward in the model, a structural equation analysis based on variances has been carried out (Liengaard et al., 2021). The PLS-SEM method through the SmartPLS software has been used to process the data and test the hypotheses (Del-Castillo-Feito et al., 2020).

This is a multivariate analysis method whose main purpose is the prediction of dependent variables by estimating robust models. This program allows to analyze and determine the estimation of the measurement model and structural model taking into account their dependent variables. It also allows for the calculation and quantification of the size of both indirect and direct effects that some of the variables in this model have on others (Cachón-Rodríguez et al., 2021). The method has the advantage of deciding whether or not to impose the direction of the hypotheses and is considered to be the most reliable and advisable according to Hair et al. (2019).

The composition of the proposed model, which includes reflective and formative variables, makes this technique and software optimal for the analysis proposed in the study (Martínez-Navalón et al., 2020).

Before starting with the analysis of the measurement scale and the analysis of the model, the classificatory data analysis of the sample obtained from the questionnaires is carried out. This analysis shows the data collected in the first part of the questionnaire and shows the characteristics of the sample.

Gender, employment status and age are studied in this analysis. This helps to have knowledge of the sample obtained. This makes it possible to specify to which levels of the population or styles of the population the results obtained in this work can be applied. The sample consists of 390 questionnaires (400 collected and 390 valid) (Table 1).

The descriptive analysis visible in Table 1 shows that the composition of the sample is mostly female. The female gender occupies 56.92% of the individuals surveyed, compared to 43.07% of the male gender. With regard to the age composition of the sample, it can be seen that the majority of the sample is made up of individuals aged between 21 and 35, followed by individuals aged between 36 and 55. Marginally the third group of individuals surveyed are in the age range 56–65 years. Finally, with regard to the employment status of the individuals, it can be seen that the majority of the sample is made up of employees (47.43%), followed by 34.41% of students.

To carry out the analysis of the proposed model, a validation analysis of the measurement scales must be carried out, as specified above, in order to subsequently carry out the analysis of the proposed relationships (Nitzl et al., 2016). Firstly, and given that this is an analysis with multidimensional variables, we begin by validating the measurement scale of the first-order model and, once the dimensions of the trust variable have been validated, we proceed to the validation of the complete measurement scale, known as the second-order scale (Hair et al., 2018). In the first-order model we find that all items have a reflective character, while in the second-order model we observe that the variables privacy, E-WOM and satisfaction have a reflective character and that the variable trust has a formative character (Martínez-Navalón et al., 2020).

When carrying out the validation of the measurement scale, different criteria are applied for the reflective and formative items. In the case of the reflective items, an analysis of individual reliability, composite reliability, convergent validity and discriminant validity is carried out. These analyses can be found in Tables 2, 3 of the paper. In the case of the formative variables, the collinearity of the indicators is assessed through the analysis of the variance inflation factor (VIF) and their weights (Rodríguez et al., 2019).

First, we proceed to the individual reliability analysis of the study. In this analysis, the item loadings (λ) are analyzed. These charges according to the criterion of Carmines and Zeller (1979) should be above 0.707. In this study, all items of the proposed first-order model exceed this cut-off threshold. Secondly, the validation analysis of the composite reliability scale is carried out, where the Cronbach’s alpha analysis is taken into account (Martínez-Navalón et al., 2019). The criterion used is that of Nunnally and Bernstein (1994) which sets an item validation cut-off at 0.7. Here too, all of the items raised pass this cut-off, as they all exceed 0.7. To analyze composite reliability, the Dijkstra-Henseler analysis (rho_A) is used, a more up-to-date criterion that is more robust in its analysis (Dijkstra and Henseler, 2015). Becoming the only real measure of reliability (Hair et al., 2019). The cut-off score is 0.7. In both composite reliability criteria all the items tested pass the analysis.

The convergent validation analysis uses the average variance extracted (AVE) in which we study how much of the variance of a variable is generated by its indicators (Martínez-Navalón et al., 2020). It is analyzed to ensure that the proposed indicators account for at least 50% of the variance of the underlying variable. The cut-off rate is 0.5 (Rodríguez et al., 2019). In this analysis, all of the items in this study exceed the cut-off threshold of 0.5 (Table 2).

Finally, for the validation of the measurement scale of the first-order variables, discriminant validity analysis is carried out. This study is carried out using Heterotrait-Monotrait analysis (HTMT). This criterion is the most current and allows the study to be more robust (Dijkstra and Henseler, 2015). It takes into account that the amount of variance of a variable captured from the indicators must be greater than the variance it shares with other variables (Ghasemy et al., 2021). Values in the study must be less than 0.90 (Henseler et al., 2015). Having carried out the analysis in the proposed model, the items SAT-3, CONH-1, CONB-1, CONC-2 are eliminated as they do not comply with the proposed criteria (Table 3).

Once the validation of the measurement scale of the first-order model has been completed, the validated items of the multidimensional variable (trust) are grouped together. The validation analysis of the measurement scale is carried out again, but in this case of the second-order model. In this second analysis, the reflective variables (privacy, E-WOM and satisfaction) meet all the criteria previously analyzed, and the scale for measuring these variables is considered to be validated. Subsequently, the validation of the formative variable (trust) is carried out, as the measurement scale of the formative variables is validated with different criteria. First, the weights are analyzed to see if they are significant (Gelashvili et al., 2021). All three items of the trust variable in the second order are significant. Once the weighting criterion has been overcome, the collinearity study of the items is carried out. This analysis is performed using the VIF (Hair et al., 2018). The chosen cut-off threshold is the one advocated by Hair et al. (2019) where it is advised that VIFs should be close to value 3, preferably with a lower value. Applying this criterion, the three items of the trust variable are retained (Table 4).

The measurement analysis is carried out, where it is analyzed whether the hypotheses are fulfilled or rejected.

In order to carry out the structural analysis of the model previously proposed, it is necessary to analyze whether there is multicollinearity between the antecedent variables and the endogenous variables. For this purpose, structural variance inflation is analyzed using the (VIF) criterion, where the structural model must have values less than 3 according to Hair et al. (2019). In this case, the structural model meets this criterion as there is no structural multicollinearity since all values are below 2. The goodness-of-fit index of the model must also be measured beforehand. For a model to be considered a good fit, the SRMR index must be less than 0.08 (Hu and Bentler, 1998). The fit index of this model is 0.06 and therefore meets this criterion.

Once the previous analyses have been carried out, a bootstrapping analysis of 50,000 samples is applied, which allows us to see the algebraic sign, the magnitude and the significance of the hypotheses put forward. In the analysis of the algebraic sign, it can be seen that all the hypotheses have the same value as the hypothesis that has been put forward. The Student’s t-test and significance analysis shows that hypotheses 1, 3, 4, and 5 meet the criteria for both analyses, while hypothesis 2 does not meet the criteria. Therefore, the hypothesis H.2 (Privacy → Trust) is rejected and H.1 (Privacy → Satisfaction), H.3 (E-WOM → Satisfaction), H.4 (E-WOM → Trust), H.5(Satisfaction → Trust) are accepted (Table 5).

Once the hypotheses have been accepted and rejected, the analysis of explained variance (R²) and effect size (F²) is carried out to measure the predictive relevance of the model (Aldás-Manzano, 2014; Hair et al., 2019). The R² for trust and the R² for satisfaction show positive values. In the case of trust it has high values according to Chin (1998) and in the case of satisfaction it has average values. This indicates that the predictor variables have a high predictive power. F² shows how an exogenous variable contributes to explain another endogenous variable. The E-WOM variable exerts a large explanatory effect on the satisfaction and trust variables. In the case of the privacy variable, it has a low explanatory effect on satisfaction. Finally, we measure the predictive relevance of the model by (Q2), obtaining a high predictive relevance (Geisser, 1975; Martínez-Navalón et al., 2019).

Table 5 and Figure 2 show the results obtained in the study.